Evaluating uncertainties in an integrated approach for chemical risk ...

Evaluating uncertainties in an integrated approach for chemical risk assessment under REACH: more certain decisions?

De evaluatie van onzekerheden in een geïntegreerde benadering voor de risicobeoordeling van chemicaliën onder REACH : betere beslissingen? (met een samenvatting in het Nederlands)

Proefschrift

ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van de rector magnificus, prof.dr. J.C. Stoof, ingevolge het besluit van het college voor promoties in het openbaar te verdedigen op donderdag, 11 juni 2009 des middags te 12.45 uur door Theodorus Gabriël Vermeire geboren op 3 juli 1953 te Bleiswijk

Promotoren: Prof.dr. W. Seinen Prof.dr. W. Slob

Dit proefschrift werd mede mogelijk gemaakt met financiële steun van het Rijksinstituut voor Volksgezondheid en Milieu (RIVM) Layout and Printing: Gildeprint Drukkerijen - Enschede. The Netherlands ISBN: 978-90-39350-58-4

Voor Anita

Table of Contents

Chapter 1 Introduction

7

Chapter 2 Integrated human and ecological risk assessment. A case study of organophosphorous pesticides in the environment

25

Chapter 3 European Union Systen for the Evaluation of Substances (EUSES): the second version

43

Chapter 4 Opportunities for a probabilistic risk assessment of chemicals in the European Union

71

Chapter 5 A probabilistic human health risk assessment for environmental exposure to dibutylphthalate

85

Chapter 6 Assessment factors for human health risk assessment: a discussion paper

103

Chapter 7 Evaluating uncertainties in an integrated approach for chemical risk assessment under REACH: more certain decisions?

163

Abbreviations

197

References

203

Summary

229

Samenvatting

237

Curriculum Vitae

245

List of publications

249

Acknowledgements

259

1 Introduction

Chapter 1

Regel 1 Regel 2 Regel 3 Regel 4 Regel 5 Regel 6 Regel 7 Regel 8 Regel 9 Regel 10 Regel 11 Regel 12 Regel 13 Regel 14 Regel 15 Regel 16 Regel 17 Regel 18 Regel 19 Regel 20 Regel 21 Regel 22 Regel 23 Regel 24 Regel 25 Regel 26 Regel 27 Regel 28 Regel 29 Regel 30 Regel 31 Regel 32 Regel 33 Regel 34 Regel 35 Regel 36 Regel 37 Regel 38 Regel 39 8

Introduction

1. Introduction

1

1.1 Chemicals control Chemicals are important in our everyday life. The chemical industry is the third largest industrial sector in the world. Chemicals add value to our lives, but also costs due to adverse effects to health and the environment from exposure at production and use and at the waste stage. The annual production of industrial chemicals classified as toxic and harmful in the EU-25 in 2005, was 212 million tonnes and is still growing (EEA, 2007). Notorious examples such as the Minimata disease due to methyl mercury (1956), the adverse effects of widescale use of pesticides like DDT on the environment highlighted by Rachel Carson (1962), the release of methylisocyanate in Bhopal (1984), the Sandoz fire with release of chemicals into the river Rhine (1986) and the adverse effects of organotin antifoulants on sea life (known from the 1970s) and many other examples all show the need for chemical control (Nichols and Crawford, 1983; Lönngren, 1992; EEA, 2001). Chemicals control covers all regulatory and voluntary measures taken to minimize the risks of chemicals to human health and the environment (Lönngren, 1992). A more precise definition was given to the analogous term ‘risk management’: risk management is a decision-making process that entails weighing political, social, economic, and engineering information against risk-related information to develop, analyse and compare regulatory options and select the appropriate regulatory response to a potential health or environmental hazard (Van Leeuwen, 2007). In this process, risk managers, both in the public sector and in industry, need to be advised by experts of various disciplines and a major one is risk assessment. Both risk managers and risk assessors experts nowadays recognise that uncertainties are part and parcel of risk assessment advice. This thesis is about chemical control and the challenges posed to both decision-makers and risk assessors in framing the questions and formulating the answers in the light of the predominant uncertainties. Therefore, important developments in both chemical control and risk assessment are introduced with a focus on industrial chemicals, before the aim of this thesis is explained. The 1972 United Nations Conference on the Human Environment was the UN’s first major conference on international environmental issues. Chemical pollution was recognized as one of the side effects of economic development and this followed the increasing demand for more chemical control from environmental movements (Lönngren, 1992). Subsequently, the Brundtland Commission, formally the World Commission on Environment and Development (UN-WCED, 1987), was convened by the United Nations in 1983 to address growing concern ‘about the accelerating deterioration of the human

9

Regel 1 Regel 2 Regel 3 Regel 4 Regel 5 Regel 6 Regel 7 Regel 8 Regel 9 Regel 10 Regel 11 Regel 12 Regel 13 Regel 14 Regel 15 Regel 16 Regel 17 Regel 18 Regel 19 Regel 20 Regel 21 Regel 22 Regel 23 Regel 24 Regel 25 Regel 26 Regel 27 Regel 28 Regel 29 Regel 30 Regel 31 Regel 32 Regel 33 Regel 34 Regel 35 Regel 36 Regel 37 Regel 38 Regel 39

Chapter 1


environment and natural resources and the consequences of that deterioration for economic and social development’. The global nature of environmental problems was coined in the term ‘sustainable development’. The Commission defined sustainable development as development that ‘meets the needs of the present without compromising the ability of future generations to meet their own needs’. In 1992, sustainable development with regard to chemicals and ‘sound management of chemicals’ were specifically addressed in Chapter 19 of Agenda 21 of the UN Conference on Environment and Development (UN, 1992). The first recommendation was to expand and accelerate the international assessment of chemical risks, which requires mutual acceptance of data and of hazard and risk assessment methodologies. In 2006, the Strategic Approach to International Chemical Management (SAICM), adopted by the UN International Conference on Chemicals Management and the Global Ministerial Environmental Forum renewed the target of global implementation of the sound management of chemicals, ‘so that, by 2020, chemicals are used and produced in ways that lead to the minimisation of significant adverse effects on human health and the environment’ (IOMC, 2006). Particularly influential in the development of instruments for chemicals control at the international level was, and still is, the Chemicals Group of the Organisation for Economic Co-operation and Development (OECD), established in 1971. A basic element of the OECD chemicals programme, and a prerequisite for harmonisation of hazard and risk assessment, is the mutual acceptance of data (MAD). MAD is built on the OECD test guidelines, the OECD Principles of Good Laboratory Practice and the Minimum Pre-marketing set of Data. Further to this work on MAD, the OECD established a hazard assessment programme for existing high production volume chemicals in 1987 and a new chemicals programme in 2006. The OECD risk assessment programme focuses on release estimation and exposure assessment and reporting. In the area of risk management, the focus is on guidance development, e.g. on socio-economic analysis, risk communication, prioritisation and the development of environmentally benign chemicals (Nichols and Crawford, 1983; Diderich, 2007). These global developments had their parallels at regional and national scale. Legally binding directives for the control of chemicals started in the 1960’s. In the European Union (EU), environmental policy started in 1973 with the adoption of the first five-year European Community Environmental Action (EC, 1973). New chemical control legislation, adopting the concept of pre-manufacturing or pre-marketing assessment of the effects of a new chemical, taking exposure into account as well as provisions for extended control measures, was in place in the European Union (EU) from 1979 (‘the 6th Amendment’), in the United States (USA) from 1976 (Toxic Substances Control Act, TSCA), in Canada from 1988

10

Introduction

(CEPA-1988), and in Japan from 1973 (Chemical Substances Control Act) ( Lönngren, 1992; Van Leeuwen et al., 2007). Other important regulatory instruments in the EU were Directive 67/548/EEC on the classification, packaging and labelling of dangerous substances, and Regulation (EEC) 793/93 on the evaluation and control of existing substances. All these regulatory initiatives were accompanied by the development of risk assessment guidance and tools such as the EU Technical Guidance Documents (TGD; EC, 2003a) and the European Union System for the Evaluation of Substances (EUSES, Chapter 3). In the EU, after many years of debate and negotiation, new legislation on the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) was adopted and entered into force on June 1, 2007. REACH is the European contribution to the sustainable development goals of SAICM and will be further explained in Box 1.1 (EC, 2007). Another feature of this modern chemical legislation is a shift from a hazard based approach, i.e. an approach based on the inherent capacity of a chemical to cause adverse effects, to a more risk based approach, i.e. an approach based on both hazard and exposure. What the impact of this is will be explained in the next section. Box 1.1 REACH (EC, 2007) REACH is the new European Regulation for Registration, Evaluation, Authorisation and restriction of CHemicals. It entered into force on June 1, 2007 to streamline and improve the former legislative framework on new and existing chemical substances. It replaced approximately forty Community Directives and Regulations by one single regulation. In principle REACH covers all substances, but some classes of substances are exempted (e.g. radioactive substances). REACH applies to chemicals as such, as components in preparations and as used in articles. The aim of REACH is ‘to ensure a high level of protection of human health and the environment, including the promotion of alternative methods for assessment of hazards of substances, as well as the free circulation of substances on the internal (EU) market while enhancing competitiveness and innovation’. REACH places greater responsibility on industry to manage the risks that chemicals may pose to health and the environment and to provide safety information that will be passed down the supply chain. All substances must be classified and labelled. The extent of the information obligations depends upon the quantity of the substances manufactured or imported. Companies which manufacture or import more than 1 tonne per year of a substance will be required to register the substance at the new European Chemicals Agency (ECHA) located in Helsinki. Since one of the goals of REACH is to limit vertebrate testing and reduce costs, sharing of data derived

11

1


Chapter 1


from in vivo testing is mandatory. Key information in registration is the risk assessment for all identified uses, the Chemical Safety Assessment, CSA. A CSA is required for production volumes above 10 tonnes per year and should include an exposure assessment and risk characterisation for all substances classified as dangerous or assessed to be PBT or vPvB1. An essential starting point for the exposure assessment is the exposure scenario, describing the operational conditions and recommended risk management measures for each identified use. Following registration, REACH will allow the further evaluation of substances where there are grounds for concern or on the basis of spot checking. Evaluation will be performed on registration dossiers, to check the testing proposals and the compliance with the requirements of registration. In addition, substances which are suspected of being a threat to human health or the environment can be evaluated by a Member State. REACH also foresees an authorisation system for the use of substances of very high concern (CMR, PBT, vPvB2), and a system of restrictions, where applicable, for substances of concern. The authorisation system will require companies to switch progressively to safer alternative substances or technologies where a suitable alternative exists. Restrictions may apply to all substances, regardless of tonnage level. Existing use restrictions from earlier legislation will remain under REACH. A detailed package of guidance documents has been prepared for authorities, industry and other stakeholders and is available from ECHA (www.echa.europa.int). REACH is based on the Precautionary Principle which may be used in environmental policies to legitimate decisions and actions in situations characterised by uncertainty (see Section 1.4). Under REACH, industry has to prove that chemicals are safe for man and the environment and need to document this in a registration dossier and communicate this up and down the supply chain. The conclusions on safe use need to be based on relevant information and a risk assessment, the latter which may include -additional- risk reduction or risk management measures. The safety assessment is based on the evidence that gives rise to highest concern. In the process of drafting the registration dossier, uncertainties with regard to exposure and the cause-effect relation should be resolved by either more data or by -more- risk management measures. In this precautionary approach, full scientific proof is not needed for a decision. Under REACH, the authorities may submit substances of concern to a risk management regime, such as the authorisation or restriction regimes, based on scientific evidence on the hazardous properties 1 Classified as dangerous in accordance with Directive 67/548/EEC on classification and labelling; PBT = persistent, bioaccumulative, and toxic according to REACH criteria. vPvB = very persistent, very bioaccumulative according to REACH criteria. 2 CMR = classification of a substance as carcinogenic, mutagenic, reprotoxic category 1 and 2 according to Directive 67/548/EEC.

12

Introduction

and complemented with information on use and exposure. The authorisation procedure is meant

1

for substances of very high concern, i.e. substances with CMR, PBT or vPvB properties. Industry is allowed to market those chemicals only when risks are limited or when a socio-economic analysis and an analysis of suitable alternative substances or technologies show that there is no alternative yet. In case of the restriction regime, the authorities have to prove that a substance poses a risk to man or the environment and also has to show that no other option remains than a marketing and use restriction. The authorities may use a socio-economic analysis to strengthen their case for a proposal to restrict the chemical. The Precautionary Principle thus seems more intertwined in the registration and authorisation regimes under REACH than in the restriction regime, since in the former two regimes remaining uncertainties due to lack of knowledge will not prevent a decision to act or not to act (risk reduction measures, authorisation or no authorisation).

1.2 Risk assessment Before the emergence of chemical risk assessment, many ancient civilizations contributed to the assessment of hazards of chemicals by investigating poisons, antidotes, and effects of medicinal substances. In maturing to a scientific discipline, toxicology passed from an observational science (lists of poisons and antidotes, management of poisoning) to an experimental and analytical science. The concept of exposure as an important element in toxicology was introduced by Paracelsus (1493-1541) and was applied to industrial health problems and to forensic issues by Ramazzini (1633-1714) and Plenck (1739-1807), respectively. Toxicology was definitely established as an experimental science by Orfila (1787-1853) (Borzelleca, 2001). Ecotoxicology as a separate discipline emerged much later in the 1960s in the wake of Rachel Carson’s book ‘Silent Spring’ and was defined by Truhaut (1977). Risk assessment began with the insurance of merchant ships in England and The Netherlands in the 17th century. From there it spread to the assessment of risks to the safety of people and property (Bernstein, 1996; Suter, 2007). Chemical risk assessment for humans really started to be applied in the 1950s with the introduction of the concept of the Acceptable Daily Intake (ADI) in the standard setting and risk assessment of food additives and pesticides (Rodricks and Taylor, 1990; Borzelleca, 2001). Risk assessment in the environmental context originated in the 1970s as a response to the enactment of a series of environmental laws (Suter, 2007). In contrast to risk assessment for decision support in areas like insurance, engineering and external safety, the quantitative estimation of risks as a product of probability and outcome or impact (e.g. costs, effects) is rare for chemicals.

13


Chapter 1


There is a multitude of definitions of risk. With regard to chemicals, risk can be defined as the probability of an adverse effect in an organism, system or (sub)population caused under specified circumstances by exposure to an agent (IPCS, 2008a). In risk assessment of chemicals the following basic questions, related to this definition, are relevant (Renn, 2008): 1. What are the undesirable outcomes and who determines what undesirable means? 2. How can we specify, qualify or quantify the likelihood of undesirable outcomes? What is the uncertainty in these specifications, qualifications and quantifications? 3. How do we aggregate outcomes, likelihood, and other risk-related factors into a common concept or framework that allows risk comparisons, the setting of priorities, the inclusion of social or cultural context, and effective risk communication? It is already indicated above that, except for non-threshold chemicals, ‘probability’ and ‘likelihood’ of adverse effects are often only addressed in a qualitative way, e.g. using exposure versus no-effect level ratios. These are, of course, no true risk levels, since both probabilities and adverse effects are not properly defined. This will be further explained in Chapter 4. Risk assessment can be defined as a process which entails the following elements: hazard identification, dose-response assessment, exposure assessment and risk characterisation. Hazard is the inherent capacity of a chemical or a mixture to cause adverse effects in humans or the environment under the conditions of exposure. Risk characterisation is an estimate of the incidence and severity of the adverse effects likely to occur in a human population or environmental compartment due to actual or predicted exposure to a chemical, and may include ‘risk estimation’, i.e. the quantification of that likelihood (EC, 2003a; Van Leeuwen, 2007). These definitions are technical (but see above note of caution on risk levels for chemicals) and do not include the interaction between risk assessors, risk managers and other stakeholders3 needed to answer the second part of above question 1 and question 3. As pointed out by many authors (a.o. US NRC, 1996; Health Council, 1996a; Van Asselt, 2000; Slovic, 2000; Klinke and Renn, 2002; Renn, 2008;) other dimensions of risk are relevant here such as risk comparison, costs and benefits, human values and risk perception, acceptability of risk, and risk communication or deliberation.

3 Throughout this thesis, risk assessors are experts providing decision-support through performing risk assessments. They can be affiliated to the public sector, industry or other stakeholders. Risk managers are considered to be regulatory decision- and policymakers on chemicals. If ‘risk manager’ is used more broadly, also including decision-makers in industry, this will be made clear. Stakeholders are members of the society outside the government such as representatives of industry, downstream users, non-governmental organisations, public interest groups, and private citizens.

14

Introduction

The close interaction between risk assessment and risk management (Section 1.1), or technical analysis and ‘deliberation’ (US NRC, 1996), is the subject of much debate and many risk assessment and management frameworks exist for both human health and environmental risks or a combination thereof (for overviews, see: Power and McCarty, 1998, and Jardine et al., 2003). Most, however, are based on the report ‘Risk assessment in the Federal Government: Managing the process’, also called the ‘Red Book’ by the US National Research Council (US NRC, 1983). This framework was originally designed for human health risk assessment only, but was later developed into a separate ‘ecological risk assessment framework’ (US EPA, 1992). All frameworks define the roles of science and political and societal values as essential parts of the process. They differ in the degree of separation of science from policy, the different roles of stakeholders at each stage of the process, the relative emphasis on risk management and risk assessment and on the use of quantitative methods of risk characterisation and uncertainty analysis (Power and McCarty, 1998; Suter et al., 2003). Three leading frameworks, the redefined US NRC framework (US NRC, 1996), the EU risk assessment framework for industrial chemicals (EC, 2003a) and the framework developed by the Food and Agricultural Organisation (FAO) and the World Health Organisation (WHO) for risk management and risk assessment of food additives (FAO/WHO, 1997) demand a major role for risk managers and stakeholders in the risk assessment process. A further step was taken in trying to develop a common framework for both human health and the environment. This will be discussed in the next section.

1.3 Framework for Integrated Risk Assessment (IRA) The WHO International Programme on Chemical Safety (WHO/IPCS), in collaboration with the United States Environmental Protection Agency (US EPA) and the Organisation of Economic Co-operation and Development (OECD), has developed a framework for integrated human and ecological risk assessment (IPCS, 2001; Suter et al., 2003 and 2005; Vermeire et al., 2007). In this framework, integrated risk assessment (IRA) is defined as: a science-based approach that combines the processes of risk estimation for humans, biota, and natural resources in one assessment (Figure 1.1). Risk management measures to be taken should be based on an integrated assessment of the risks, e.g. abatement techniques recommended for the protection of workers should not lead to unacceptable risks for the environment and vice versa. The underlying thought is that decisions concerning the management of chemicals and materials in the environment will be inadequately informed until assessments are sufficiently integrated to address effects on the environment as well as on humans. History learns that these effects have usually been studied independently. For example: the first reports on adverse effects of organochlorine compounds like DDT and PCBs on 15

1


Chapter 1

environmental species were already published in the early 1960s, whereas concerns about the possible adverse effects on human health only appeared in 1974. Imposex in marine organisms has been related to exposure to organotins since the early 1980s, but the first risk evaluation of human intake of organotins was only published in 1999. Sulphur dioxide emissions became first of all regulated because of human health concerns, but regulatory action towards ecological concerns was much slower. Likewise, regulatory actions against the use of hormonal growth promoters in livestock in the 1970s and 1980s were based on human health effects, but the discussion on the environmental impact of the use of these compounds was only gradually attracting attention (EEA, 2001). It can be argued in all these cases that both the scientific discussion and the regulatory responses could have benefited from a more integrated, interdisciplinary approach leading to sharing of information, decreased uncertainties and fully informed decisions. IRA acknowledges the interdependence of risks to humans and to non-human species that result from communalities in the sources and routes of exposure, the conservative nature across species of many toxic mechanisms and the link between the quality of human life and the environment and vice versa. (Vermeire et al., 2007). IRA closely matches developments in the EU, where some regulatory frameworks, notably those on industrial chemicals and biocides, already require a partly integrated risk assessment (EC 2003a).

INTEGRATED RISK ASSESSMENT RISK MANAGEMENT

STAKEHOLDER PARTICIPATION


Problem Formulation With Hazard Identification (Figure 1.2)

Analysis Exposure Assessment

Characterization Characterization Doseof of Response Exposure Effects Assessment

Risk Characterization

Figure 1.1: IPCS framework for integrated risk assessment (IRA)4. 4 In the IRA framework, hazard identification is part of problem formulation in which data are being evaluated and stressors of concern selected. In the EU, hazard identification is part of the analysis stage, in which adverse effects are identified.

16

Introduction

This requirement also extends to the new EU chemicals policy dealing with the registration, evaluation and authorisation of chemicals (REACH; EC, 2007). Political pressure has further led to a European Environment and Health Strategy (EC 2003b) proposing an integrated approach involving closer cooperation between health and environment research. This Strategy aims at the development of an EU system integrating information on the state of the environment, the ecosystem and human health, taking into account mixture effects, combined exposure, and cumulative effects. The Strategy is connected to the European Environment and Health Action Plan 2004-2010 (EC 2004a) and builds upon the aims of the Commission’s Sixth Environment Action Programme, a specific target of which is that levels of pollution in Europe should not give rise to deleterious effects on human or environmental health. IRA is claimed to facilitate achieving the goals of greater efficiency and quality and better informed decisions (Munns et al., 2003). The major components of IRA are shown in Figure 1.1 and show the following features: 1. The activities of risk assessors, risk managers and stakeholders are shown as parallel activities which are concerned with the same issues but may address them differently. Interactions are encouraged and can take place in various ways, and at various points in the process, depending on the legal and social context and the assessment problem. For example, while risk assessors focus on risks to humans and the environment, risk managers must also evaluate the legal, economic and technological constraints on management actions. These, as well as stakeholders’ concerns, may all contribute to the risk assessment. 2. The problem formulation component (Figure 1.2) is the process by which the assessment is defined and the plan for analysing and characterising the risk is developed. The first step is a planning dialog that clarifies the management goals, the purpose and scope of the assessment, and the resources available. On the basis of all information available, assessment endpoints, i.e. the specific attributes and entities to be protected, are selected as well as an agreed conceptual model, which shows the relations between sources, stressors, and assessment endpoints (e.g. Figure 2.1). 3. A common analysis plan may allow for efficiency in data generation and modelling. 4. In the analysis, the Weight-of-Evidence (WoE) process in combining multiple lines of evidence, for instance in establishing exposure characteristics, mode-of-action, causeeffect and dose-effect relationships and extrapolations can potentially be based on the same criteria and using all information. The use of alternative methods in stead of vertebrate tests potentially can benefit a lot from such a holistic approach since their acceptance strongly depends on a successful integration of in vivo, in vitro, in silico and exposure data.

17

1


Chapter 1

5. At the risk characterisation stage, IRA may provide the opportunity to present the results in an integrated fashion, such as using the same spatial (e.g. local or regional level) and temporal (e.g. long-term versus short-term) scales, that makes the consequences of alternative actions clear to all parties. It can also provide clearer and more useful results by expressing variability and uncertainty due to lack of knowledge in the same terms for all endpoints. Integration is not always possible or desirable. No doubt, there are traditional disciplinary barriers between the ecological and human health research areas. There are also institutional and regulatory boundaries preventing interaction. Certain problems might be associated narrowly with either humans or environmental species. Costs and complexity may also be considered too high in relation to expected benefits. Therefore, further demonstration of these benefits is needed (see Chapter 7). Since the assessment of uncertainty is key to risk assessment and to decision-support as discussed in this thesis, this subject will be introduced in the next section.

Integration of Available Information

RISK MANAGEMENT

STAKEHOLDER PARTICIPATION


Source and Stressor Characteristics

Systems Potentially at Risk

Assessment Endpoints (Hazard ID)

Human Health/ Ecological Effects

Conceptual Model

Analysis Plan

Problem formulation

Analysis

Figure 1.2: The problem formulation stage of IRA.

18

Introduction

1.4 Uncertainty and risk The general concept of the infallibility of science of the Enlightenment in the 17th and 18th century gradually has been replaced by the notion that scientific knowledge has limitations and is always subject to some degree of uncertainty. Therefore, the concept of risk is intertwined with the concept of uncertainty (Funtowicz and Ravetz, 1990; EEA, 2001; Van Asselt, 2000; Verdonck et al., 2007). As will be pointed out later, there are different types of uncertainty, some quantifiable and others not, some reducible and others not. Ignoring this may lead to incomplete risk assessments, poor decision-making and risk communication. Risk assessors and risk managers have to take uncertainty into account and they should realise that ‘high quality (in policy-related science) does not require the elimination of uncertainty, but rather its effective management’ (Funtowicz and Ravetz, 1990). An uncertainty management framework is needed parallel to the IRA framework described above. Another reason why the concept of uncertainty should be addressed in risk assessment and management is the possible application of the Precautionary Principle (PP). In issues of health or environmental policy characterised by high uncertainty, the PP is frequently discussed, whereas at the same time common understanding of its principles is lacking (Sandin, 1999; Rogers, 2003). An evaluation of historical examples of decision-making under uncertainty for chemicals such as asbestos, tributyltin, benzene and PCBs (EEA, 2001) shows the complex interplay between science, regulatory appraisal and policy making and to what extent the PP could have prevented harm before conclusive scientific evidence became available. REACH is claimed to be based on the PP (see Box 1.1). The PP is defined by the European Commission as follows (EC, 2000): ‘the PP applies where scientific evidence is insufficient, inconclusive or uncertain and preliminary scientific evaluation indicates that there are reasonable grounds for concern that the potentially dangerous effects on the environment, human, animal or plant health may be inconsistent with the high level of protection chosen by the EU’. The PP is considered to be a risk management instrument and should not be confused with applying prudential approaches (e.g. assessment factors) in risk assessment (EC, 2000). Common elements of PP definitions are that the PP applies when there exist considerable scientific uncertainties about causality, magnitude, and nature of harm and that some form of scientific analysis, including a thorough uncertainty analysis, is mandatory. In such situations, characterised by lack of scientific knowledge, the possibility of certain outcomes is sufficient to legitimate actions. So plausibility reasoning should be an important element of scientific evaluations and should be expressed in a transparent way. Interventions are required before possible harm occurs or before uncertainty can be reduced. Application of the PP is limited to those hazards that are unacceptable according to defined criteria, for instance the ‘level of 19

1


Chapter 1


protection in the EU’ in the definition given. Interventions should also be proportional to the chosen level of protection, the magnitude of possible harm and the costs (in relation to the benefits) (COMEST, 2005). The PP is sometimes accused of being unscientific. However, the scientific evaluation should be based on reliable scientific data and lead to a conclusion on the plausibility, the likelihood and the severity of a hazard’s impact. Essential is an evaluation of the reliability of the assessment as well as the remaining uncertainties and a correct typology of the uncertainties. (EC, 2000; EEA, 2001; Van Asselt and Vos, 2005; COMEST, 2005). As will be shown in this thesis, uncertainty in the EU risk assessment for industrial chemicals is, so far, largely taken into account by applying ‘realistic worst case’ exposure situations with average or typical values for the model parameters, conservative assumptions and assessment factors leading to deterministic risk estimates. With some exceptions, only known and quantifiable uncertainties are considered and others hidden in assumptions or ignored. The degree of conservatism in these estimates is unknown. (see Chapter 3; Bodar et al., 2002; Verdonck et al., 2007).The Technical Guidance Documents (TGD) under REACH (ECHA, 2008a) recognize the importance of uncertainty analysis and advise a tiered approach for the analysis of quantifiable uncertainty in accordance with WHO/IPCS principles (IPCS, 2008b; see Chapter 7). How can the results of the uncertainty analysis in risk assessment be incorporated in decision-making? An ‘uncertainty management framework’ for model-based decision support was conceptually described by Walker et al. (2003) and subsequently developed further into a guidance document for application in integrated environmental assessments (Van der Sluijs, 2003). The uncertainty management framework closely parallels the stages in the IRA scheme en is very useful for highlighting different types of uncertainty in the risk assessment of chemicals . The stages are (not necessarily in this sequence): 1. Problem framing & context analysis: this requires identification of the problem, the context and the history in a dialogue between decision-makers, stakeholders and scientists. Possible solutions can be discussed and a first analysis of uncertainties made. 2. Process assessment: given the problem as decribed at stage 1, the risk assessor should know what the implications are for the process of the risk assessment. He should identify the stakeholders and risk managers and map their views, values and interests. Their interests in the assessment should be defined as well as when they should be involved. 3. Communication: The information flow between all groups concerned is mapped and communication pathways identified. 4. Environmental assessment methods: Identification of methods and tools used for the assessment and characterisation of the uncertainties associated with these methods. 20

Introduction

5. Uncertainty identification and prioritisation: For the steps above key uncertainties should be identified as well as an analysis of the best method to approach each uncertainty. Any gaps between the uncertainty methods required and those used should be highlighted. The uncertainties should be prioritised. 6. Uncertainty analysis: The uncertainty analysis is carried out. 7. Review and evaluation: An overview of the analysis made and an evaluation of the robustness of the results. 8. Reporting: The results and implications should be communicated to the decision-makers and stakeholders identified. It is noted that the first five of these stages closely match the problem formulation issues of the IRA framework. Stage 6 belongs to the analysis phase. Stages 7 and 8 are included in risk characterisation. This framework was first of all developed for and applied to integrated environmental assessments. It is believed to contribute significantly to the area of risk assessment of chemicals since it highlights the importance of classification, assessment and management of uncertainties. The uncertainty framework (Walker et al., 2003; Van der Sluijs et al., 2003) classifies types of uncertainty according to the dimensions ‘location’, ‘nature’ and ‘level’. Other typologies exist but are comparable, be it less detailed and focussed on quantifiable uncertainties (EFSA, 2006; IPCS, 2008b; ECHA, 2008b).  ‘Location’ refers to the sources of uncertainty. Uncertainties may originate from: 1. The context: context deals with choices of the boundaries of the system to be modelled and the framing of the issues and formulation of the problems to be addressed. It includes uncertainty in the model boundaries and scenarios and in the description of the chemical agent. 2. In expert judgement: uncertainty in interpretations and weight-of-evidence analysis (for example in hazard assessment). 3. The model and model parameters: this refers to the conceptual model as applied to the model scenario and includes uncertainty with regard to model structure (e.g., pathways, formulations of exposure or dose-response model, model assumptions), the technical model implementation (the software and hardware), model input (e.g. assessment factors, chemical-specific parameters) and internal model parameters. 4. The data: measurements, monitoring and survey data, including data used for calibration of the models. 5. The output: the accumulated uncertainty or prediction error in the final outcome of the assessment. If the true values would be known, a validation exercise could establish the prediction error.

21

1


Chapter 1


 ‘Nature’ refers to whether uncertainty in each location is primarily a consequence of the incompleteness and lack of scientific knowledge or due to variability. Only the first type, also called true or epistemic uncertainty, may be reduced by more research. Often, uncertainty is a mix of both types. Uncertainty in our knowledge about actual variability is reducible.  ‘Level’, expresses how the uncertainty in each location can be classified on a scale from statistical uncertainty or inexactness, which can be quantified adequately in statistical terms, to scenario uncertainty, which has unknown probabilities and can only be depicted as a range of possible outcomes (e.g. worst case, best case), to recognized ignorance (fundamental uncertainty, known unknowns). Van der Sluijs (2003) also include other manifestations of uncertainty: qualification of the knowledge base, actually characterising the reliability of the information, and valueladenness of choices, or degree of subjectiveness. The latter uncertainty type is related to differences in ideological views and in views on how the problem should be framed and solved.

1.5 This thesis This thesis concentrates on uncertainty and variability in the risk assessment methodology for industrial chemicals as applied within the current regulatory framework in Europe, REACH. In spite of this restriction to industrial chemicals, it is believed that this study also provides significant results for other regulatory frameworks. The methodological approaches discussed address the risk assessment for both humans and the environment. The role of probabilistic uncertainty analysis in risk assessment is highlighted as well as its specific contribution to decision-making. The aim of this thesis is to investigate in what way the scientific process of risk assessment can improve decision-making in the light of uncertainty. The following research questions will be addressed: 1. What is the past and present approach towards uncertainty analysis in the methodology for risk assessment of industrial chemicals? What risk assessment methodology is in place and how can uncertainty be approached both qualitatively and quantitatively? 2. What does this mean for past and current scientific advice from risk assessors to risk managers in the public sector and to stakeholders? 3. What are the future perspectives for incorporation of uncertainty in decision making on chemicals?

22

Introduction

For answering these questions it is considered necessary to build on the two frameworks explained above: the framework for Integrated Risk Assessment (IRA) and the framework for uncertainty management. This thesis will discuss the most relevant components of these frameworks and investigate what contribution can be expected from their implementation in risk assessment and risk management. Table 1.1 shows how this work is structured. The IRA-framework was described in Section 1.3 and first of all will be further explained and exemplified in Chapter 2. This chapter will show how the IRA framework could be applied to a human and ecological risk assessment of organophosphorous pesticides (Vermeire et al., 2003). The methodology used in Chapter 2 is based on risk assessment methodology developed in Europe, described in the EU TGD (EC, 2003a) and implemented in the European Union System for the Evaluation of Substances (EUSES; EC, 2004b). This EUSES methodology will be described in Chapter 3 (Vermeire et al., 2004 and Vermeire et al., 1997). This is necessary to be able to understand the approaches to uncertainty analysis for exposure assessment, effects assessment and risk characterisation described in chapter 4 (Jager et al., 2001a) and applied in chapter 5 (Vermeire et al., 2001a). One important aspect of uncertainty analysis is the widely used application of assessment factors to derive human no-effect levels and ecotoxicological no-effect levels. This subject will be reviewed extensively in Chapter 6 with regard to human assessment factors. In addition, this chapter will discuss more extensively the probabilistic approach to establish a human health limit value based on Benchmark modelling as applied in Chapter 5 (Vermeire et al., 1999a). Chapter 7 finally discusses the relevant methodological aspects of risk assessment and uncertainty analysis. It will show how the IRA-framework can be improved, taking into account the current understanding of the role of uncertainty analysis in risk assessment. This improved framework will be put in the context of REACH. The thesis concludes with a discussion of the implications for decision-making.

23

1


Chapter 1


Table 1.1: Structure of this thesis IRA-element Integrated Risk Assessment framework Uncertainty Analysis framework, uncertainty classification Problem formulation Analysis o Characterisation of exposure o Characterisation of effects o Quantitative uncertainty analysis  General approaches  Application of probabilistic risk assessment  Uncertainty in assessment factors for human effects assessment  Probabilistic human dose-response assessment  Risk characterization o Review and evaluation of results o IRA and uncertainty management o Implications for decision making    

24

Thesis-chapter Chapters 1, 2 Chapter 1 Chapter 2 Chapter 3 Chapter 3 Chapters 4 - 6 Chapters 4 and 6 Chapters 5 Chapter 6 Chapter 6 Chapter 7

2 Integrated human and ecological risk assessment A case study of organophosphorous pesticides in the environment

Theo Vermeire1, Robert McPhail2, Mike Waters3 RIVM, The Netherlands US EPA - National Health and Environmental Effects Research Laboratory, USA 3 US EPA – National Institute of Environmental Health, USA

1

2

Published in: Journal of Human and Ecological. Risk Assessment 9: 343-357 (2003)

Chapter 2


2. Integrated human and ecological risk assessment: A case study of organophosphorous pesticides in the environment This study was chosen as an example of integrated risk assessment because organophosphorous esters (OPs) share exposure characteristics for different species, including human beings and because a common mechanism of action can be identified. The “Framework for the integration of health and ecological risk assessment” is being tested against a deterministic integrated environmental health risk assessment for OPs used in a typical farming community. It is argued that the integrated approach helps both the risk manager and the risk assessor in formulating a more holistic approach towards the risk of the use of OP-esters. It avoids conclusions based on incomplete assessments or on separate assessments. The database available can be expanded and results can be expressed in a more coherent manner. In the integrated exposure assessment of OPs, the risk assessments for human beings and the environment share many communalities with regards to sources and emissions, distribution routes and exposure scenarios. The site of action of OPs, acetyl cholinesterase, has been established in a vast array of species, including humans. It follows that in the integrated approach the effects assessment for various species will show communalities in reported effects and standard setting approaches. In the risk characterisation, a common set of evidence, common criteria, and common interpretations of those criteria are used to determine the cause of human and ecological effects that co-occur or are apparently associated with exposure to OPs. Results of health and ecological risk assessments are presented in a common format that facilitates comparison of results. It avoids acceptable risk conclusions with regard to the environment, which are unacceptable with regard to human risk and vice versa. Risk managers will be prompted to a more balanced judgement and understanding and acceptance of risk reduction measures will be facilitated.

26

Integrated human and ecological risk assessment

2.1. Background This study was chosen as an example of integrated risk assessment for two reasons: first because organophosphorous esters (OPs) share exposure characteristics for different species, including human beings, such as sources, emissions, distribution and pathways of exposure, and secondly because a common mechanism of action can be identified. Similarities in exposure and effect provide a means of evaluation and comparison across species. Additionally, other types of integration such as cumulative and aggregated exposure can be considered for OPs. Aggregate and cumulative exposure to OPs is relevant in integrated approaches since both humans and environmental species can be exposed through various, not seldom common pathways of exposure to OPs, which are believed to share a common mechanism of action. The US-National Research Council (US NRC, 1993) performed a risk assessment for children exposed through residues in food to five commonly used organophosphates (OPs: acephate, chlorpyrifos, dimethoate, disulfoton, ethion) and used actual data on their presence on eight foods and three juices to explore the development of methods for assessing exposure to multiple chemicals. The US Environmental Protection Agency (US EPA, 2001) released a preliminary OP cumulative risk assessment for 24 OPs incorporating exposure of humans via food, drinking water and residential/non-occupational pathways. In this case study information package the possibility is explored to include other routes of human exposure (direct exposure, exposure via the environment) and exposure of aquatic, terrestrial and wildlife species. The analysis includes the use of both monitoring and modelling results. The aim therefore is to explore: 1. The integration of risk assessments for man and the environment (integrated and aggregated exposure) 2. The integration of risk assessments of several OPs together (cumulative exposure).

2.2 Problem formulation 2.2.1 Impetus for the assessment Organophosphorous pesticides (OPs) are widely used pesticides and are believed to act through a common mechanism of action. There are ample reasons for integrating research and risk assessments for the OPs. OP exposure pathways overlap for many wildlife species and humans. For example, the spraying of crops with OPs can cause pesticide drift to nearby communities. Similarly, pesticide run-off into water bodies can cause harmful

27

2


Chapter 2


effects on aquatic species, terrestrial species that forage around water bodies, and humans that reside or recreate in the vicinity. OP contamination of well water can harm humans, long after the adverse impact of spraying on wildlife has occurred. In many instances it may be possible to use wildlife species as sentinels of the imminent or impending risks of OPs to human health. In addition, the site of action of OPs (acetyl cholinesterase) has been well established in a vast array of species, including humans. Moreover, cholinergic receptors, which are stimulated indirectly by cholinesterase inhibition, are found throughout most taxonomic groups of animals. The risk manager may have been made aware of the fact that, although exposure to a single compound may not exceed the level considered being without acceptable risk for humans or environmental species, concurrent exposures to numerous OP-compounds could exceed a safe level because of increased cholinesterase (ChE) inhibition. Moreover, absence of a probable risk for one species, e.g. man, does not automatically imply absence of a probable risk for other species. These considerations will prompt the risk manager to formulate a question to risk assessors advocating an integrated approach and, if needed, a risk reduction strategy benefiting all organisms. 2.2.2 Assessment questions The risk manager will work out the basic questions to be addressed together with the risk assessor. In this particular case the question may look like this: Given the considerations above (see under ‘Impetus for the assessment’), present a deterministic, integrated environmental health risk assessment for a group of commonly used OPs in a typical farming community. This local scale assessment should consider the risk for humans, wildlife and other environmental species resulting from both direct and indirect exposure at and following application of OPs. The assessment should include both short-term and long-term risks. Poisoning is not considered in this case information package. 2.2.3 Assessment endpoints Coherence in endpoints used to assess health and ecological risk can be specified for this case to pertain to acetyl cholinesterase inhibition and differences in susceptibility in man, wildlife, aquatic species and terrestrial species as a result of exposure via dietary and nondietary sources. Cholinesterases as the site of action of OPs have been identified in a vast array of species, including humans, but not in plants and micro-organisms. Cholinergic receptors, which are stimulated indirectly by cholinesterase inhibition, are found throughout

28


most taxonomic groups of animals. Though acetyl cholinesterase inhibition in itself is not an adverse effect, studies have been performed in different species to investigate the relation between toxicity and cholinesterase inhibition. 2.2.4 Conceptual models The conceptual model involves sources and pathways of exposure of: 1. Environmental organisms and applicators directly exposed after spraying/granulate treatment; this route includes exposure of water- and sediment organisms via spray drift. 2. Environmental/domesticated organisms and humans indirectly exposed via: • residues in food derived from crops on which the pesticide is applied directly; • crops, ambient air, and soil in non-target areas; • indoor air/surfaces during and following use of OPs in sprays/foggers/etc. and perhaps via medication and personal care products (head lice treatment, via contaminated lanolin). This route obviously only is applicable to man and domesticated animals. The conceptual model for the local problem may be put in a wider context of a regional or watershed approach. 2.2.5 Analysis plan An integrated approach offers substantial opportunities for more efficient data gathering activities. This case offers the following opportunities: 1. Sharing of data on emissions 2. Sharing of distribution, fate and exposure models and the parameter values and distributions needed for these models 3. Sharing of data on concentrations in environmental media and food 4. Sharing of analytical activities to obtain the data mentioned above 5. Sharing of toxicokinetic and physiologically-based pharmacokinetic models 6. Sharing of dose-response models for ChE-inhibition 7. Sharing of analytical activities to obtain information on dose-response and variability across and within species. 2.2.6 Summary An integrated approach already offers advantages in the problem formulation stage: 1. It helps both the risk manager and the risk assessor in formulating a more holistic approach towards the risk of the use of OP-esters. It avoids conclusions based on incomplete assessments or on separate assessments using unnecessarily different assumptions, parameter values, distribution and fate models and exposure scenarios.

29

2


Chapter 2


2. It helps in identifying opportunities in increasing the database available for both the human and the environmental risk assessment risk (increased efficiency/resourceeffective) 3. It helps to identify a coherent expression of the results across species in terms of exposure (common pathways), adverse effects (in relation to ChE-inhibition), dose response and eventually the risks.

2.3. Characterisation of exposure 2.3.1 Sources and emissions In an integrated approach it is useful to consider the whole life cycle of OPs and all possible sources. This inventory will allow a well-founded selection of sources and relevant life cycle stages with potentially significant emissions. The expertise required here is the same for both the human and the environmental risk assessment. Potentially, emissions of OPs can occur at production, formulation, use and disposal. For convenience, we will leave out production, formulation and disposal in our example, assuming these processes occur outside the geographic area of interest. Specifically, the exposure assessment concentrates on sources and emissions following use of OPs as pesticides and biocides. Sources with possible emissions to the environment or direct exposure can be identified at spraying or application of granulates in agriculture, use in dips in animal husbandry, use as a biocide and use in medications (US NRC, 1993). Biocidal uses include use in indoor sprays and foggers and in flee control products for pets. OPs are used medicinally in head lice treatment products. Residues of OPs may be present in lanolin originating for sheep treated in dips. 2.3.2 Distribution pathways Figure 2.1 shows distribution towards the environmental media ambient air, soil, surface water, groundwater, sediment, and soil. For communalities: see Section 3.4.

30


Humans, wildlife, plants

Terrestrial organisms

0 Direct exposure

0

Applicators, Consumers, Wildlife

Micro-organisms

2

Aquatic organisms

Terrestrial organisms

Bees

0 = direct exposure 1 = emission 2 = drift 3 = deposition 4 = sludge application 5 = leaching 6 = drainaige 7 = volatilisation 8 = dilution 9 = sedimentation

Aquatic organisms

Sediment organisms

Figure 2.1: Distribution routes for agricultural pesticides. Grey boxes contain the receptor organisms; dotted lines and boxes are used if the only release is via the Sewage Treatment Plant (STP) 2.3.3 Transport and fate models Data needed are measured concentrations in the environmental compartments or input for distribution models to estimate environmental concentrations. The latter requires physico-chemical properties, partition coefficients, degradation rates, deposition rates, and environmental characteristics (Van Leeuwen and Hermens, 1995). General local distribution screening model specific for agricultural and non-agricultural pesticides are available (e.g., RIVM et al., 1998).

31


Chapter 2


2.3.4 External and internal exposure models granules, seeds insects

spray air crops Humans dairy products

Birds, mammals

soil cattle

groundwater

meat

drinking water earthworm

surface water fish

Fig. 2.2: Schematic representation of possible routes of environmental exposure of humans and wildlife Examples of common distribution and exposure routes, sharing common transport models, fate models and monitoring data: • Direct exposure of applicators, by-standers and wildlife and invertebrates to OPs in spray drift; • The route from OPs spray application towards surface water leading to exposure of aquatic organisms and, through drinking water and food, of human beings, birds and mammals; • The route from OPs application towards non-target soil leading to exposure of terrestrial organisms and, through crops and drinking water prepared from groundwater, of human beings, birds and mammals. Figure 2.2 shows possible routes of environmental exposure of humans and wildlife. Further communalities can be found in the estimation of aggregate exposure of human beings, wildlife and birds through food (oral), drinking water (oral) and air (inhalation, dermal). Cumulative exposure to several OPs together is an issue relevant to both human beings and wildlife. 32


Direct exposure estimation need quantification of the potential exposure in the form of concentrations in the exposure media that may contact the human body. Quantification may follow from exposure data. Otherwise, simple models may be applied to obtain a reasonable worst case prediction of the exposure. This modelling involves quantification of the contact with the exposure media containing the substance by defining exposure routes and exposure patterns, including contact durations, contact frequencies and site of contact. Both a chronic exposure measure and an acute exposure measure may be needed, depending on the expected effects and exposure patterns. Consumer exposure models are available and may be modified to be applicable to wildlife as well (Van Veen, 1996; Van Veen 1997; Pandian et al., 1997; EC, 1997). Local screening models for the estimation of environmental exposure of man and other organisms to pesticides are also available (e.g., RIVM et al., 1998). This estimation involves bioaccumulation and biotransformation models (Van Leeuwen and Hermens, 1995). Aggregate exposure assessment methodology is specifically addressed by ILSI (1998). There is little experience with cumulative exposure methodology. With respect to OPs, a cumulative risk assessment has been published by US NRC (1993). Most risk assessments dealing with OPs rely upon the administered dose or the external dose because of insufficient understanding of how to estimate the internal dose and the relation between the internal dose and effects. In some cases, however, PBPK models may be available and may be used for a risk assessment based on the internal dose for human beings or wildlife (Paustenbach, 2000). 2.3.5 Measures of exposure related parameters Examples of not-receptor-specific parameters with common values and units are: • Emission rates: e.g., the fraction of the amount of OPs applied in kg per ha, emitted to air and water at spraying; • Concentrations in environmental media: e.g., Predicted Environmental Concentrations (PECs) in air, soil and surface water; • Biotic and abiotic degradation/disappearance rates in environmental media: e.g., the rate of biodegradation in soil and surface water, the rate of hydrolysis and photodegradation, the rate of volatilisation, resuspension, and sedimentation; • Characteristics of environmental compartments: e.g., the fraction organic carbon in soil and sediment, temperature, rainfall, dilution rates, water flows etc.; • Partition coefficients: e.g., the octanol-water partition coefficient and the air-water partition coefficient.

33

2


Chapter 2


2.3.6 Analytical tools Integrated exposure assessment requires the application of the same quantitative methods such as methods for sensitivity and uncertainty analysis. Monitoring of environmental media, drinking water and food (target and non-target crops) is relevant both for the human and the environmental risk assessment of OPs. Monitoring strategies for OPs should, however, consider the spatial and time scales relevant for common exposure routes, e.g., consider the need for local, short-term concentrations in surface water as medium of exposure for fish, mammals and birds just after spraying and the need for long-term averages at a larger spatial scale to estimate the exposure of birds, mammals and human beings (following treatment) away from the application area. 2.3.7 Summary In an integrated exposure assessment of OPs, the risk assessments for human beings and the environment share many communalities with regards to sources and emissions, distribution routes and exposure scenarios. Taking these into account will be an efficient way to deal with the risks to all organisms, including humans. Monitoring efforts can be more cost-effective.

2.4. Characterisation of effects 2.4.1 Reported effects and modes of action Organophosphates (OPs) are widely used pesticides that are applied primarily to crops for the control of agricultural pests, as well as in and around residences and offices for the control of urban pests. OPs are also one of the two major classes of cholinesterase-inhibiting pesticide. The main mode of action of the OPs is inhibition of acetyl cholinesterase, the enzyme that terminates the action of acetylcholine neurotransmitter, which is released by nerve stimulation, on postsynaptic cholinergic receptors in the nervous system. OPs produce an irreversible inhibition of acetyl cholinesterase, in contrast to the carbamates (the second major class) that produce a reversible inhibition. Since the principle site of action of OPs is the nervous system, it is not surprising that OPs have produced a variety of toxic effects. These effects have been documented in humans, laboratory animals and several wildlife species (aquatic and terrestrial), due to either accidental or intentional exposures (Ecobichon and Joy, 1994; Mineau, 1991). Acute exposures produce a well-known syndrome of autonomic distress including salivation, lacrimation, urination and defecation (the SLUD syndrome). In addition, acute exposure compromises neuromuscular function, decreases motor activity and body temperature, and alters cardiovascular function. Extremely high doses produce convulsions and death, due to interference with brain-stem structures involved in respiration. 34


Acute exposure to some OPs also produces a delayed neuromuscular effect, seen mainly in the extremities, which is irreversible and can lead to paralysis (Johnson, 1975). Organophosphate-induced delayed neuropathy (OPIDN) has been shown in several susceptible species, including birds and humans, although the basis for species differences in sensitivity is unknown at this time. Repeated exposure to OPs can produce tolerance to the acute effects due mainly to a downregulation of muscarinic cholinergic receptors in the central nervous system. While the observation of tolerance may indicate a reduced risk of exposure, there may be residual risk factors that are operative. For example, organisms made tolerant to an OP are often more sensitive to the effects of muscarinic blocking agents (e.g., belladonna alkaloids). Furthermore, available evidence indicates many OPs are not interchangeable; tolerance to one OP may either confer tolerance to another, have no effect on the actions of another, or may increase susceptibility to still other OPs (Costa and Murphy, 1983). Currently, there is much concern over age-related susceptibility to the OPs (US NRC, 1993). Evidence in support of this concern comes mainly from studies on laboratory rodents, although there is considerable evidence of developmental toxicity in avian reproduction studies (Mineau et al., 1994). 2.4.2 Biomarkers and indicators Since the main mode of action of OPs is inhibition of acetyl cholinesterase, enzyme inhibition has been widely used as biomarker of exposure in both human-health and ecotoxicological research. Whether enzyme inhibition can be used as a biomarker of effect, on the other hand, is debatable. One complication for arriving at a consensus is the observation of a threshold for inhibition above which toxic effects are produced. For example, it is widely assumed that toxic effects ensue only when the inhibition exceeds 20% in brain. Reviews of the literature, however, provide little empirical support for such a sweeping generalisation. As a consequence, little concerted effort has so far been made at exposure-response modelling. Despite the advantages of integrated risk assessments for OP pesticides, some caveats are in order. For example, there are numerous methods used to determine OP-induced cholinesterase inhibition. There are, however, no clear indications at this time on the comparability of the inhibition as determined by the different analytical methodologies. In addition, there are a number of other esterases that OPs inhibit to varying degrees. Some of these esterases are considered sinks for OPs that diminish the inhibition of acetyl cholinesterase. Understanding the relative abundance and activity of esterases in general will be necessary in order to make firm predictions of risk in receptor species. 35

2


Chapter 2


2.4.3 Exposure-response modelling Since OPs produce systemic toxicity, involving primarily the nervous system, the Reference Dose (RfD) approach is widely used for OP standard setting (see MacPhail and Glowa, 1999 for details). An RfD is calculated by dividing a no-adverse-observed-effect level (NOAEL) by a series of uncertainty factors (UFs). Similar approaches are being applied to determine no-effect levels for environmental organisms using NOAELs, NOECs (NoObserved-Effect Concentrations), LC50s or EC50s. These approaches generally ignore the shape of the dose-response curve. New approaches make use of species sensitivity distributions in ecotoxicological risk assessments (OECD, 1992). In human risk assessment, Benchmark dose modelling (Slob and Pieters, 1998: see Section 6.4.3) and categorical regression (Teuschler et al., 1999) have been proposed. Teuschler et al. (1999) applied categorical regression to OPs. 2.4.4 Extrapolations Standard-setting for OPs based on adverse effects on human health involve a number of extrapolations. This is because the data used for standard setting ordinarily come from experiments on laboratory organisms (most often rodents) exposed to relatively high doses and for relatively brief durations UFs are ordinarily a factor of 10 and are included to compensate for limits in our understanding of how toxic substances works, which severely compromises our ability to make accurate predictions of risk. One UF is included when the human health standard is set using laboratory animal data. In other words, as a conservative assumption, humans are assumed to be maximally 10 times more sensitive to OPs than are laboratory rodents. Another UF of 10 is included to compensate for individual differences in susceptibility to OPs, implying no more than one to two orders of magnitude difference in the range of sensitivity. Recently, an additional UF has been recommended for inclusion in standard setting in order to protect children from the risk of OPs and other pesticides. It is important to note, however, that despite the widespread use of UFs in regulatory decisionmaking there is little evidence available for the biological plausibility of UFs (see Chapter 6). The situation could be remedied by empirical selection of UFs and their magnitude(s). In this regard, an empirical approach has recently been described to selecting an UF for interspecies variation in sensitivity that is based on analysis of avian toxicity data used for pesticide regulation purposes (Mineau et al., 1996). Contrary to standard assumptions Mineau et al. found that smaller species were more sensitive. Extrapolation methods for environmental organisms are also available: see for example EC (1996) and Crommentuijn et al. (2000). The evolutionary conservation of enzyme and receptor make OPs an ideal candidate for comparative studies on physiology, biochemistry, metabolism and susceptibility to OPs. Such studies may make it possible to establish species-specific toxic equivalency factors (TEFs) for the OPs (US NRC, 1993). 36


2.4.5 Direct and indirect effects OPs are indirectly acting agents in that inhibition of acetyl cholinesterase causes an accumulation of acetylcholine with subsequent overstimulation of cholinergic receptors. Recent evidence suggests, however, some OPs have direct stimulatory effects on cholinergic receptors at extremely low concentrations (Huff and Abou-Donia, 1995; Ward and Mundy, 1996). Recent evidence also suggests that acetyl cholinesterase may serve as a tropic factor that guides development of the nervous system in several species (Bigbee et al., 1999). The disruption of brain development by OPs may explain their developmental toxicity in avian species. Behavioural decrements may result in increased susceptibility to predation, reduced provisioning of the young, and reduced feeding. There may be other indirect effects produced by OPs. For example, some OPs have repellent actions that underlie the avoidance of OP-contaminated food displayed in avian species Bennett (1989). The reduction of forage species can affect growth of other species or, vice versa, the reduction of certain species can lead to excessive growth of forage species (algal bloom following reduction of crustacea). Finally, toxic effects in wildlife may alter community composition and food-web dynamics leading to further indirect effects of a magnitude and impact on the environment and human beings that is presently unknown. Clearly, only an integrated approach may reveal such interactions. 2.4.6 Summary The site of action of OPs, acetyl cholinesterase, has been established in a vast array of species, including humans. This may lead to a better understanding of effects of different levels of acetyl cholinesterase inhibition in different species. It follows that in an integrated approach the effects assessment for various species will show communalities in reported effects and standard setting approaches on the basis of no (observed) adverse effect levels. Species-specific TEFs have been proposed for OPs. However, caution must be exercised since acetyl cholinesterase inhibition may differ for different OPs and different species.

2.5. Risk characterisation 2.5.1 Combining exposure and effects The results of the characterisations of exposures to OP compounds and associated effects are combined to estimate the risks to each endpoint. The uncertainties associated with the risks are determined, and summarised for presentation to the risk manager and stakeholders. In this relatively simple case, an exposure estimate is used to estimate the likelihood of adverse effects by comparing the exposure value to a limit value. A common set of evidence, common criteria, and common interpretations of those criteria

37

2


Chapter 2


are used to determine the cause of human and ecological effects that co-occur or are apparently associated with exposure to OPs. A best estimate of risk is derived from results of toxicity tests of different species, results of single chemical and mixtures toxicity tests, and exposure estimates derived from different fate models and from environmental measurements. These lines of evidence are quantitatively weighted and combined evidence from ecological and human health risks is integrated. 2.5.2 Uncertainty Through uncertainty analysis, the risks of various stressors are expressed in a common form (e.g., the probability of occurrence of cholinesterase inhibition in humans and in the ecological setting). The integrated assessment starts with a common concept and terminology of uncertainty (e.g., distinguish variance from true uncertainty), and as far as appropriate uses common analytical methods. In a deterministic assessment upper-bound estimates are based on conservative estimates of exposure and risk. In our example, worst case values will, for instance, be used for anatomical and dietary properties of humans and cattle, partition coefficients, bioconcentration factors and biotransfer factors. In uncertainty analysis, not only this upper-bound estimate will be estimated, but the full distribution of intakes of the affected population. This allows the risk manager to choose an appropriate level of uncertainty (e.g., the 50th or 99th percentile of the intake distribution), to separate individual variability (e.g. in human body weights or food intake factors) from true scientific uncertainty (e.g., in estimates of partition coefficients) and to consider benefits, costs and comparable risks. In our example the uncertainty analysis would require the following: • Definition of statistical distributions of key input parameters such as: • variability in application rate, human body weights and food and drinking water intake factors, inhalation rates, fractions of food home-grown, and fat contents; • variability and uncertainty in ingestion of grass, soil and air by cattle; • uncertainty in percentage of drift, leaching/deposition/degradation/dilution rates, the ratio of plant dry mass to fresh mass, partition coefficients, bioconcentration and biotransfer factors. • Generation of a distribution of exposure through simulation. • Comparison of this exposure distribution with a fixed value of the Acceptable Daily Intake and determination of the probability that this ADI is exceeded. Note that the assessment may be further developed by taking into account the variability and uncertainty in the humans effects assessment. • A similar approach can be undertaken for environmental species.

38


2.5.5 Presentation of results Results of health and ecological risk assessments are presented in a common format that facilitates comparison of results, i.e., a common presentation of results with an explanation of differences in the magnitude of effects. Similarly, the uncertainties are presented in a common form (e.g., cumulative frequency). This integrated risk characterisation facilitates the task of communicating risks to risk managers and the public. An example of a common risk measure that could be used in this example is the risk characterisation ratio of the exposure estimate (PEC) and the no-effect level for acetyl cholinesterase inhibition (PNEC, ADI, RfD). Alternatively, an uncertainty analysis may result in the common risk measure being the probability that the exposure estimate (PEC) exceeds the effects estimate (PNEC, ADI, RfD). If dose-response relations are known, a probability distribution of effects can be obtained and decisions can be taken on the basis of an acceptable level of effects (e.g., Klepper et al., 1998, for ecosystems and Slob and Pieters, 1998, for humans). Integration of the results could include an integration of effects over different OPs and over the geographic area of interest, e.g., the farming community in this case. 2.5.1 Summary An exposure estimate is used to estimate the likelihood of adverse effects by comparing the exposure value to a limit value. In an integrated approach, a common set of evidence, common criteria, and common interpretations of those criteria are used to determine the cause of human and ecological effects that co-occur or are apparently associated with exposure to OPs. Results of health and ecological risk assessments are presented in a common format that facilitates comparison of results. It avoids acceptable risk conclusions with regard to the environment, which are unacceptable with regard to human risk and vice versa.

2.6. Risk management and stakeholder participation Risk management is the process of deciding what actions should be taken to mitigate or reduce the risk. It involves making decisions concerning actions in response to estimated risks to humans or ecological systems. Risk managers may be concerned on the potential health effects of occupational and nonoccupational exposure to OPs among farmer families in an area known for intensive use of these substances (e.g. Azaroff, 1999; Simcox et al., 1995). Other risk managers may be interested in effects of OPs on wildlife in such an area (e.g. Custer and Mitchell, 1987). In an integrated approach these studies would be combined addressing both issues and making effective use of the available exposure and effects assessment expertise, exposure models,

39

2


Chapter 2


monitoring data and monitoring strategies. The risk managers may be able then to make a balanced judgement on the risks for not only the farmer families, but also their environment on the basis of the estimated risks and the socio-political and economic implications of alternative risk reduction options. An integrated risk assessment could make clear, which receptors are at risk and which not. Communalities as well as differences will become clear and risk management options can be focussed without neglecting receptors and interactions between them. For example, restrictions on the presence of humans in fields during spraying and for some period thereafter reduce risks to humans, but not ecological receptors. Granular formulations of OPs are less risky to humans than sprays but are more risky to birds, because the latter ingest granules as grit. 2.6.1 Summary An integrated approach makes effective use of available resources for estimating risks. It also allows a balanced judgement on the risks to all organisms potentially at risk, showing communalities as well as differences and interactions.

2.7. Risk communication Risk communication involves risk managers, risk assessors, the general public and stakeholders. Risk questions and answers presented in an integrated way show communalities and differences between the various receptors and will highlight the interaction between risk reduction options for individual receptors. Simple or unnecessary solutions to parts of the problem are avoided. This will increase understanding of oftencomplex problems and support coherent decision making which is acceptable to all parties. A ban of a specific OP in spite of the absence of proven risks for professionals or environmental organisms will be more easily explained and accepted when it can be demonstrated that there is an aggregated risk to children exposed via food and via household applications. 2.7.1 Summary An integrated approach will increase the likelihood of understanding and acceptance of risk reduction measures in the risk communication stage.

40


Acknowledgements The contribution of all members of the IPCS Planning Group on Approaches to Integrated Risk Assessment is gratefully acknowledged. The authors would like to thank specifically Prof. Dr. C.J. van Leeuwen (University of Utrecht) for critically reviewing the manuscript.

2

41


Chapter 2


3

European Union System for the Evaluation of Substances (EUSES): the second version

Vermeire, T.1, Rikken, M.1, Attias, L.2, Boccardi, P.3, Boeije, G.4, de Bruijn, J.5, Brooke, D.6, Comber. M.7, Dolan, B.8, Fischer, S.9, Heinemeyer, G.10, Koch, V.11, Lijzen, J.1, Müller, B.12, Murray-Smith, R.13, Tadeo, J.14 National Institute of Public Health and the Environment, RIVM, P.O. Box 1, 3720 BA Bilthoven, The Netherlands 2 Istituto Superiore di Sanità, Rome, Italy 3 Ministero dell’ambiente e della tutela del territorio, Rome, Italy 4 CEFIC, Procter & Gamble, Strombeek-Bever, Belgium 5 European Chemicals Bureau, ECB, Joint Research Centre, Ispra, Italy 6 BRE Environment, Watford, UK 7 CONCAWE, Exxon Mobile, Brussels, Belgium 8 Pesticide Control Service, Dublin, Ireland 9 National Chemicals Inspectorate (KEMI), Solna, Sweden 10 Federal Institute for Risk Assessment, Berlin, Germany 11 CEFIC, Clariant GmbH, Sulzbach, Germany 12 Umweltbundesamt, UBA, Berlin, Germany 13 CEFIC, Astrazeneca Global SHE Operations, Brixham, UK 14 INIA, Madrid, Spain 1

Adapted from: Chemosphere 59:473-485 (2005)

Chapter 3


3. European Union System for the Evaluation of Substances (EUSES): the second version This publication presents a general overview of EUSES followed by a description of major changes in the assessment of the risks of chemicals to human health and the environment as implemented in the second version of the European Union System for the Evaluation of Substances, EUSES 2.0. EUSES is a harmonised quantitative risk assessment tool for chemicals. It is the PC- implementation of the technical guidelines developed within the framework of EU chemical legislation for industrial chemicals and biocides. As such, it is designed to support decision making by risk managers in government and industry and to assist scientific institutions in the risk assessment for these substances. The development of EUSES 2.0 is a coordinated project of the European Chemicals Bureau, EU Member States and the European chemical industry. Several model concepts, the technical background and the user interface of EUSES have been improved considerably. Major changes in the environmental assessment such as the implementation of emission scenario documents for industrial chemicals and biocides, the addition of the marine risk assessment, the enhancement of the regional model to include global scales, and improvements in the secondary poisoning and environmental effects modelling will be discussed. The update of the human risk assessment module in EUSES focuses on the risk characterisation for both threshold and non-threshold substances with, among others, the introduction of assessment factors. The performance of EUSES is illustrated in an example showing the human and environmental risk assessment of a sanitation disinfectant for private use.

44

European Union System for the Evaluation of Substances (EUSES)

3.1 Introduction EUSES, the European Union System for the Evaluation of Substances, is a harmonised quantitative risk assessment tool. It is the PC- implementation of the technical guidelines developed within the framework of EU chemical legislation for industrial chemicals and biocides. As such, it is designed to support decision making by risk managers in government and industry and to assist scientific institutions in the risk assessment for these substances. In 1967, the European Community adopted Directive 67/548/EEC on the classification, packaging and labelling of dangerous substances – the first of a growing number of Community directives aimed at protecting human health and the environment (EC, 1967). Since then, the concepts of risk assessment and risk management of substances gradually became firmly established in EU legislation. New substances were not on the EU market in the 10 years prior to 18 September 1981 and therefore not appearing in the European INventory of Existing Commercial chemical Substances (EINECS). With regard to new substances the so-called ‘Sixth amendment’ to Directive 67/548/EEC introduced a premarket testing, hazard assessment and notification procedure. The first article of the seventh amendment, Directive 92/32/EEC (EC, 1992) required an evaluation of the potential hazards and risks of notified substances on the basis of a specified data set. This Directive also required that principles be laid down for carrying out the risk assessment of new substances. On 20 July 1993, Commission Directive 93/67/EEC was adopted, which laid down these principles for new substances (EC, 1993a). A detailed package of Technical Guidance Documents (TGD; EC, 1993b) supported this Directive. Similar regulatory developments took place for existing substances (EC, 1993c; EC, 1994) and biocides (EC, 1998). In 1996, Technical Guidance Documents had been developed for the human and environmental risk assessment of new and existing substances (EC, 1996a). This TGD was supported by the PC-program EUSES 1.0, the European Union System for the Evaluation of Substances (EC, 1996b; Vermeire et al., 1997). Further to the adoption of the Biocides Directive and new scientific developments in risk assessment, the TGD was updated (EC, 2003a). The EU guidance supporting the risk assessment of biocides is the same as for new and existing substances with regard to the environment and the human health hazards. The human exposure assessment for biocidal products and the human risk characterisation have been elaborated in separate Technical Notes for Guidance (TNG; EC, 2002a, and EC, 2002b, respectively).

45

3


Chapter 3


The second version of EUSES, EUSES 2.0, not only incorporates changes in and additions to the EU-TGD of 2003, it also builds on the experience gained by many users since the release of the first version. Some bugs were reported in the user interface and many suggestions for improvement have been collected in a so-called EUSES Blacklist. In addition, several more formal evaluations of EUSES 1.0 have been performed (Jager, 1998a; Schwartz et al., 1998; Schwartz, 2000; Berding, 2000; Trapp and Schwartz, 2000). EUSES 2.0 (EC, 2004b) is a co-ordinated project of the European Chemicals Bureau, EU Member States and the European chemical industry. Background material for EUSES 2.0, in particular the EU-TGD (EC, 2003a), as well as EUSES 2.0 itself can be downloaded from the website of the European Chemicals Bureau (http://ecb.jrc.ec.europa.eu). A general description of the first version of EUSES 1.0, largely based on Vermeire et al., 1997, will be presented in Section 3.2. Subsequently, major changes in the risk assessment methodology incorporated in EUSES 2.0 with respect to EUSES 1.0 will be discussed. It takes the form of a concise summary of various changes in the environmental assessment and a more elaborate presentation of a recently added key element in the TGD, the human risk characterisation for both threshold and non-threshold substances.

3.2 The first version 3.2.1 Historical developments Since the early 1980s, along with the implementation of the European legislation on new chemicals, projects were initiated at national level to develop a more systematic approach towards the hazard and risk assessment of substances. This was in recognition of the fact that risk assessment of the many substances in use nowadays could only be performed if rapid, systematic and transparent approaches were available based on the latest scientific developments. In 1990, the EU Member States adopted a document describing common principles and a stepwise procedure for the environmental risk assessment of new substances. In The Netherlands, a risk assessment system was developed integrating risk assessment tools for new and existing substances and agricultural and non-agricultural pesticides. Several EC Member States took an active interest in this project. The risk assessment system was launched in 1994 as USES 1.0 (Uniform System for the Evaluation of Substances; RIVM et al., 1994; Vermeire et al., 1994). USES 1.0 was already much in line with the separate packages of TGDs for new and existing substances and also appeared to be useful as a risk assessment tool outside the European Union. As a next step, a project was initiated to develop an update of USES 1.0 which would be fully in line with the package of amalgamated TGDs

46


for new and existing substances (EC, 1996a). EUSES is a co-ordinated effort of EU Member States, the European Commission and its European Chemicals Bureau, and the European chemical industry. 3.2.2 Objectives EUSES has been developed for quantitative assessment of the risks posed by new and existing chemical substances to man and the environment. This assessment should be transparent to all users and easy to perform and EUSES is, therefore, well-documented and available as a userfriendly computer program. The risk assessment system should be attuned to current chemical management policies and in accordance with the principles laid down in the TGDs for new and existing substances of the European Commission. Risks to man pertain to consumers, workers and humans exposed through the environment. Risks to the environment include risks to sewage treatment plant populations of micro-organisms, aquatic, terrestrial and sediment ecosystems and populations of predators. EUSES was designed to support decision-making by risk managers in the government, scientific institutes and industry in the evaluation of new and existing chemical substances. On the basis of the results of the risk assessment process with EUSES, and taking other factors into account (e.g. political, social, economic, and engineering factors), risk managers may take decisions with respect to regulatory actions to be taken (EC, 1996a). In line with most assessment procedures, EUSES allows for tiered risk assessments of increasing complexity, requiring an increased amount of data. Using OECD terminology, EUSES can specifically be used in the initial, or screening, and intermediate, or refined, stages of assessment (OECD, 1989). On the basis of a screening with EUSES, it can be decided if more data need to be generated and if a more refined (i.e. intermediate) assessment is necessary. When dealing with (large) numbers of chemicals, this screening can be used to set priorities for data gathering or refined assessments. EUSES can also be applied for intermediate or refined assessments by allowing replacement of default values, estimated parameter values, or intermediate results by more accurately estimated values or by measured data. The system is not specifically designed for site-specific assessments, but adjustment of parameters may provide insight in specific local or regional situations. 3.2.3 General principles The assessment in EUSES is carried out in a stepwise procedure encompassing the following stages: 1. Exposure assessment; 2. Effects assessment, comprising hazard identification and dose-response assessment

47

3


Chapter 3


3. Risk characterisation. At the risk characterisation stage, this procedure will result in a quantitative comparison per substance of the outcome of the exposure assessment, a Predicted Environmental Concentration (PEC) or an estimated daily dose, and that of the effects assessment. The ratio of these outcomes is called the Risk Characterisation Ratio (RCR). The RCRs should be regarded as surrogate parameters for risk characterisation since they do not quantify the “incidence and severity” of adverse effects. The RCRs are used as indicators of the likelihood of adverse effects occurring since a better method for a more quantitative risk characterisation with general applicability is not available at the moment. The human sub-populations and ecological systems and populations considered to be protection goals in EUSES are shown in Figure 3.1. The risk assessment for man aims at such a level of protection that the likelihood of adverse effects occurring is low taking into account the nature of the potentially exposed population (including sensitive groups) and the severity of the effect(s). In the environmental risk assessment it is assumed that ecosystem sensitivity depends upon the most sensitive species with regard to toxicity and that protection of ecosystem structure also protects community function. The PNEC derived for each ecosystem is regarded as a concentration below which an unacceptable effect will most likely not occur (EC, 1996a). Risk assessment with EUSES will depart from a screening level in which so-called generic exposure scenarios are applied. It is assumed that substances are emitted in a standard environment with pre-defined environmental characteristics, agreed upon by the EU Member States. No measured data are used at this level apart from physico-chemical properties and (eco)toxicological data. The resulting screening-level risk assessment is in principle valid for all EU countries as required by the relevant EU regulations. Subsequently, in an iterative process, the risk assessment can be refined by:  Adapting any default parameter value for which this is considered necessary; and  Replacing intermediate results by: o The results of other models judged to be more suitable for the substance under investigation; and o Reliable and representative measured data. The output of EUSES will always show the result of the standard assessment, in addition to the results of refined assessments made. The exposure assessment in EUSES covers the whole life cycle of substances as well as their fate in all environmental compartments. As explained in more detail later, three spatial scales and two time scales are distinguished. The exposure assessment aims at ‘reasonable worst

48


case’ results by applying unfavourable, but not unrealistic standard exposure scenarios and, as much as possible, mean, median or typical parameter values. Human populations: -

Workers

-

Consumers

-

Man exposed via the environment

Ecological systems and populations: -

Micro-organisms in sewage treatment systems

-

Aquatic ecosystem

-

Terrestrial ecosystem

-

Sediment ecosystem

-

Top predators

3

Figure 3.1 Human populations and ecological systems and populations in EUSES. 3.2.4 System structure The main structure of EUSES is presented in Figure 3.2. In the input module, the basic data to run EUSES are entered in a format comparable to that of the EU/OECD Harmonised Electronic Data SET (HEDSET), the New Chemicals Database (NCD) and the International Uniform Chemical Information Database (IUCLID). Data gaps, especially regarding secondary chemical-specific data such as partition coefficients and bioconcentration factors, are filled by estimates, using generally agreed procedures, or default values. In the emission module, emission factors for all relevant life cycle stages are chosen from a database on the basis of the known properties, uses, functions and process characteristics of a substance. This database is filled with default values agreed upon by experts in this area and, where possible, based on industry category documents.

49


Chapter 3






    

 

    

 

   

 

 Figure 3.2: The main modules of EUSES

Dedicated local models and a regional multi-media model of the Mackay level III type (SimpleBox, Van de Meent, 1993; Mackay et al., 1992) can be found in the distribution module. Figure 3.3 shows the distribution routes considered in EUSES as well as the organisms to be protected. The local model comprises, among others, a sewage treatment plant (STP) simulation model (SimpleTreat; Struijs, 1996), an air distribution model (Van Jaarsveld, 1990), a surface water dilution and sorption model and a one-compartment soil module. Endpoints are concentrations in air, STP-influent and effluent, sewage sludge, surface water, sediment, soil and groundwater.





 

The exposure module produces intake levels for predators and for humans. To indicate the potential of a substance to cause adverse effects in organisms higher in the food chain, two example food chains are modelled: predating birds and mammals exposed through consumption of fish or earthworms. For humans, a distinction is made in exposure through consumer products, exposure at the workplace, and exposure through the environment by inhalation of outdoor air and consumption of food and drinking water. In the effects module, no-effect levels for relevant time scales are determined on the basis of results from single-species experiments for aquatic and terrestrial organisms, sewage treatment plant micro-organisms and top predators. Depending on the data available, assessment factors are applied. In case data on soil and sediment organisms are lacking, the 56 50


equilibrium partitioning method (Di Toro et al. 1991) is used, assuming these organisms are equally sensitive to the chemical in pore water as the aquatic organisms are to the chemical in surface water. The risk characterisation module compares the results of the exposure assessment of a substance with those of the effects assessment by calculating Risk Characterisation Ratios (RCRs) for the various groups to be protected. Finally, the output module presents input data, defaults changed, intermediate results, and final results in a suitable format. 



3











  



      

 

  







 









Figure 3.3: The distribution pathways and organisms to be protected in EUSES

 3.2.5 System dimensions

Three factors determine the dimensions of EUSES: the spatial scale, the time scale, and the ‘realism scale’, the latter being the degree of realism attained in the exposure assessment. Spatial scales For the risk assessment system, a distinction can be made in three spatial scales. On the personal scale, individual consumers or workers are considered, exposed directly to individual substances and preparations, and to substances embedded in a solid matrix. The local scale considers the protection targets in the vicinity of one large point source of the substance. This is not an actual site, but a hypothetical one with pre-defined, agreed environmental characteristics, the so-called ‘standard environment’. The regional scale assesses the risks 56 51

57


Chapter 3


to protection targets due to all releases in a larger region assuming the same environmental characteristics as the local standard environment. The local scale receives the background con centration from the regional scale. Time scales Local releases of industrial chemicals can either be continuous or discontinuous (e.g. in case of batch processing of substances). Depending on the release frequencies and durations, organisms with a relatively short lifespan may be exposed locally to toxic concentrations during a considerable length of time, even if average exposure levels are low. This will be relevant for sewage treatment plant (STP) micro-organisms and aquatic organisms. Therefore, for these organisms, the average exposure levels during emission episodes are assumed to be continuous. It follows from this assumption that the estimated environmental concentrations can be considered as estimates of long-term exposure levels, which can be compared to noeffect levels derived from long-term toxicity data. In case intermittent release is identified, only short-term effects are considered for the aquatic ecosystem and no-effect levels are derived from short-term toxicity data only. The exposure of terrestrial organisms is assumed not to be influenced by temporal fluctuations in emission rates due to slow dynamics of the soil compartment compared to air and water. In case of human beings, these fluctuations are of rather short-term compared to their life span and the time scale on which chronic effects are regarded. These protection goals are therefore assumed to be exposed to levels averaged over a longer period, and derived from average emission rates. Exposure of consumers and workers can be judged to be acute, sub-chronic or chronic, depending on the product and its use pattern. The results of the exposure assessment should be compared to experimental animal studies of adequate duration. Releases at the regional and continental scale are regarded as diffuse and continuous, leading to steady-state environmental concentrations. These steady-state levels can be considered as estimates of long-term average exposure levels. They can therefore be compared to no-effect levels derived from long-term toxicity data. The ‘realism scale’ A model can never give an exact representation of reality. This is, among others, due to the complexity of reality, and the limited of knowledge of it. Furthermore, the data available for a model are often incomplete and contain measurement errors. In risk assessment, we are typically confronted with this situation, as the available data are limited and mechanisms

52


often poorly understood. The values for nearly all parameters are therefore accompanied by a significant amount of uncertainty, not only resulting from limited scientific understanding, but also from natural variability in time and space. River flow rates, as an example, can be measured with reasonable accuracy. Nevertheless, variability in time can be a significant source of uncertainty in this parameter. Furthermore, as the standard assessment cannot be performed site-specific, the differences between locations will result in large spatial variability. A model system like EUSES can therefore only give an approximation of the potential risk of a substance.To avoid underestimation of potential risk, a worst-case approach might be followed by choosing a worst-case exposure scenario with the worst possible emission factors, model parameters, and environmental conditions. Such an accumulation of worst cases will, however, eventually lead to unrealistically high risk levels which are extremely unlikely to occur. The aim of EUSES is to perform a ‘reasonable worst-case’ risk assessment. The chosen standard exposure scenario represents an unfavourable, but not unrealistic, situation. For the model parameters, however, mean, median or typical values will be used in most cases. As an example, the human environmental exposure scenario on a local scale is a typical worst case since all food products are derived from the vicinity of a point source. In contrast, many model parameters such as environmental characteristics are median or typical values. A model is a simplified representation of reality. Therefore, model analysis is important to gain more insight into the model’s behaviour. With USES1.0, limited model analysis was performed: a framework for uncertainty analysis was defined, some example uncertainty analyses were performed and the validation status was discussed (Slob, 1994; Jager and Slob, 1995; Jager, 1995a and b)

3.3 What’s new in EUSES 2.0? Several model concepts, the technical background and the user interface of EUSES have been improved considerably. Table 3.1 presents the most important new or updated items of EUSES 2.0. The user interface has also been improved to enhance the user-friendliness of the program. For example, more options for parameter units are available, the input screens of the emission module are adapted and the menu items are renamed and regrouped, compliant to standard Windows menu styles.

53

3


Chapter 3


Table 3.1: Most important new or updated items of EUSES 2.0 Item Model: Temperature correction

EUSES 1.0

EUSES 2.0

No

Yes (vapour pressure, solubility, hydrolysis and biodegradation)

Koc-QSAR Number of use patterns Life-cycle stages

1 (predominantly hydrophobics) 10 max 5 (production, formulation, processing, private use, recovery)

Emission Scenario Documents Biocides emission scenarios Consumer exposure Worker exposure

No No 1 scenario 1 scenario: EASE 1.0

Biomagnification factor (top)predators Marine (salt water) assessment Regional distribution model

No No SimpleBox 2.0

Sediment effects assessment Standardised sediment and soil effects Multiple chronic endpoints Non-threshold effects Ref-MOS, MRR, AOEL, AER

No No No No No

19 (all TGD QSARs) 20 max (with titles) 6 (production, formulation, industrial use, private use, service life, waste treatment) Yes Yes 10 scenarios max (with titles) 10 scenarios max (with titles): EASE 2.0 Yes Yes SimpleBox 3.0 (incl. global scales, marine compartments) Yes Yes Yes Yes Yes

Program: Operating environment

16-bit, MS-Windows 3.x

Printing Batch command mode Import facilities

Built-in report generator No EXP (EUSES), HedSet, Snif

32-bit, MS-Windows 95, 98, 2000, NT, XP Export to Word® and Excel® Yes EXP (EUSES), Snif, ConsExpo

3.3.1 Release estimation Emission scenario documents (ESD) have been developed and described in the context of the TGD for new and existing substances and are available to be used for more in-depth release estimates. ESDs are based on more profound (as compared to default emission factors) studies of the environmental release of substances from different industries, which are classified in industrial categories. All documents describe environmental releases from specific use categories under an industrial category. The release information that is given in the ESDs should be used instead of the default emission factors that are given in Part II, Appendix 1, of the TGD. In EUSES 2.0 not all ESDs are available, because some documents are still under preparation. An overview of the documents for new and existing substances covered by EUSES 2.0 is given in Table 3.2.

54


Table 3.2: Emission Scenario Documents for new and existing chemicals present in EUSES 2.0 Industrial category 5 6 7 8 10 11 12 13 14

Use category

Description of industrial category

9, 15 9 10, 51 11, 14, 29, 35, 40, 49, 50 42 20, 22, 49, 53 2, 10, 31, 43, 47, 55 10 2, 10, 20, 48, 52

Personal and domestic: production volume > 1000 tonnes/year Public domain: production volume > 1000 tonnes/year Leather processing industry Metal extraction industry Photographic industry Polymers industry Pulp paper and cardboard Textile processing industry Paints, lacquer and varnished industry

Biocides are active substances or preparations intended to destroy or deter the action of harmful organisms. The risk assessment of biocides (use category 39) is comparable to that of new and existing substances. What differs mainly is the release estimation. Biocides have been divided into 23 product types according to the Biocidal Products Directive 98/8/EC (EC, 1998). The biocidal product types for which specific emission scenario documents were available up to the middle of 2003 are incorporated in EUSES 2.0 (Table 3.3). In principle, the risk assessment for biocides is carried out for the local scale only. Nevertheless, in the future a regional risk assessment for water will be possible for those biocidal product types that are based on a use tonnage of the substance (e.g. for the product type ‘disinfection of rooms, furniture and objects in the medical sector’). The life cycle stages of production and formulation are not specifically considered in the ESDs, as these are covered under the general chemical categories. In the beginning of an assessment with EUSES 2.0 a choice has to be made for new and existing substances or biocides. Therefore, two runs of EUSES 2.0 are required when a biocide must also be evaluated for the relevant production or formulation stages. For product type 12 and 13, for which no specific biocidal emission scenario is available, the emission scenario documents (ESDs) for new and existing substances are used instead.

55

3


Chapter 3


Table 3.3: Biocidal product types with emission scenarios in EUSES 2.0 Product type Description of product type 2 Private area and public health area disinfectants and other biocidal products: sanitary and medical sector 6 In-can preservatives: fluids used in paper production 7 Film preservatives: paper and cardboard 9 Fibre, leather, rubber and polymerised materials preservatives: textile and fabrics, leather and hides, paper and cardboard 12 Slimicides 13 Metalworking-fluid preservatives 22 Embalming and taxidermist fluids

3.3.2 Regional multi-media model In EUSES 2.0 steady-state exposure concentrations at the regional, continental, and global scales are calculated for all environmental compartments using the multi-media fate model SimpleBox 3.0. In contrast to the SimpleBox 2.0 version of EUSES 1.0, the version of EUSES 2.0 includes the following extra options (Den Hollander and Van de Meent, 2004): • Three global scales (moderate, arctic and tropic); • Sea-water compartment; • Temperature dependence of parameters; • Substance dependent mass transfer at air-soil and air-water interface. The EU risk assessment is extended to cover risks to the marine environment. Therefore, a marine compartment is implemented in EUSES 2.0. A simplified representation of the regional model, including a marine compartment, is presented in Figure 3.4. In EUSES 2.0 the regional marine assessment assumes that 99% of the input comes from rivers and that 1% is emitted directly from inland sources to the marine compartment without treatment. Continental rivers flow directly to the continental sea or to regional rivers. The marine local assessment is practically identical to the inland freshwater assessment, except for the use of a sewage treatment plant (STP) and the dilution factor. By default, direct discharges to the marine compartment are assumed to be untreated and hence bypass the STP module. In EUSES 2.0, the marine sewage treatment plant can be used optionally. Dilution in receiving marine surface water and sorption to suspended solids are taken into account. As proposed in the TGD (EC, 2003a), the local marine dilution is assumed greater than in a river, which becomes higher when tidal influences are taken into account. For the coastal zone a dilution factor of 100 is assumed that is related to a discharge volume of 2000 m3/d. Just as for fresh water this factor represents the dilution at the point of complete mixing of effluent and receiving water. For estuaries, which are influenced by currents and tidal movements, as a first approach it is assumed that either the inland or the marine risk assessment covers them. No specific estuary assessment is implemented in EUSES 2.0, but the user can adapt dilution factors if relevant data are available. In the TGD (1996a, 2003a) the area

56


fractions of natural soil and agricultural soil were exchanged. In EUSES 2.0 the correct area fractions of 27% and 60% of the land area are applied to natural soil and agricultural soil, respectively.

3

 Figure 3.4: Schematic representation of the regional model in EUSES 2.0



3.3.3 Secondary poisoning According to the TGD of 2003, a biomagnification factor (BMF) is added to the secondary poisoning concept. The BMF is defined as the relative concentration in a predatory animal compared to the concentration in its prey. The freshwater food chain is modelled with a  fish-eating predator, or predators feeding on other relevant species. The food chain for the  marine environment is additionally modelled with a top-predator preying on fish-eating predators (see figure 3.5). Measured BMF data are at present very limited and therefore  default values should be used; these vary from 1 to 10 (Chapter 3.8 Part II of the 2003 TGD). By establishing these factors it is assumed that a relationship exists between the  BMF, the BCF and the log Kow. A BMF is selected on the basis of a log Kow. If a BCF for fish is available, it is possible to use that as a trigger instead of log Kow. A BCF for other  relevant organisms may also be considered, such as bivalves for the marine assessment. EUSES 2.0 gives priority to a measured BCF over a trigger based on log Kow.

57


Chapter 3


 





 

 

 

 

 





 







Figure 3.5: Indicative food chains of EUSES 2.0 

                  

3.3.4 Environmental effects assessment For the environmental effects assessment, most of the changes are carried out in the aquatic and sediment compartment. The main changes are the addition of a marine compartment and sediment assessment factors. Like the freshwater assessment, the marine assessment is based on the three-taxa model. Compared to fresh waters, the greater diversity of taxa of marine waters is assumed to produce a broader distribution of species sensitivity. Based on this assumption, the assessment factors used to derive a PNEC (predicted no effect concentration) from toxicity data are higher for marine waters, to reflect the greater uncertainty in the extrapolation. The assessment factor can be lowered, when additional data for marine taxonomic groups are available such as echinoderms or molluscs. EUSES 1.0 only applied the equilibrium partitioning method for the effects assessment of sediment organisms. Consistent with the TGD of 2003, EUSES 2.0 will derive a PNEC for sediment dwelling organisms using specific sediment ecotoxicity data, by applying an appropriate assessment factor to the lowest available LC50, NOEC or EC10, depending on the type of data available. As for the aquatic compartment, the assessment factors for marine sediment are higher than those for fresh water sediment and can also be lowered

58

59


when additional marine data are available. In contrast to the other PNEC values, the PNEC for sediment is an open parameter, i.e. a parameter that can be overwritten by the user, to allow for expert judgement from the available data outside EUSES. When a substance is not analysed in whole sediment, but in a specific fraction of the sediment, the toxicity data can be corrected for the organic carbon content of the tested fraction. In EUSES the results can be corrected to standard sediment with an organic carbon content of 5% (or an organic matter content of 8.5%). The correction method used for sediment toxicity data is equal to the method used for soil toxicity data, the only difference being that soil toxicity data are corrected to the standard soil characteristic of an organic carbon content of 2% (or an organic matter content of 3.4%).

3

3.3.5 Human risk characterisation The human risk characterisation chapter in the TGD of 2003 was revised completely in specialised working groups for both threshold and non-threshold effects, respectively. Bottom line of this chapter is that the risk characterisation should be performed as quantitatively as possible5. 3.3.5.1 Threshold effects For threshold-based effects, the quantitative risk characterisation is carried out by calculating ‘Margins Of Safety’ (MOS). The MOS is the ratio of an effect or no-effect parameter value, e.g. an acute oral LD50, a subchronic, inhalatory No-ObservedAdverse-Effect Level (NOAEL) or Bench Mark Dose (BMD) and an exposure value of corresponding time scale and route of exposure. For each human subpopulation to be protected (workers, consumers, humans exposed via the environment), the risk characterisation should be performed for each relevant exposure scenario, for the time scales (acute, subchronic, chronic) and route(s) of exposure chosen, and for each relevant endpoint. Endpoints here are acute toxicity, repeated dose toxicity, non-genotoxic carcinogenicity and reproductive toxicity, including fertility, developmental toxicity and maternal toxicity. For some other relevant endpoints such as irritation, corrosion and sensitisation, data usually are insufficient to determine a threshold. The MOS should account for the various uncertainties and variabilities in the extrapolation from experimental data to the human situation (interspecies differences, intraspecies differences, differences in duration and in route of exposure, dose-response relationship), for the uncertainties in the available data set (adequacy of and confidence in available data set, 5

Changes in the TGD version discussed here relative to the REACH TGD are discussed in Section 7.2.

59

59


Chapter 3


nature of effect), and for the uncertainties in the exposure estimate of the exposure scenario under consideration. All aspects that can be dealt with quantitatively (as assessment factors, AFs) are combined to form the overall assessment factor or reference-MOS (RMOS). In judging the acceptability of the MOS, in a second step of the quantitative risk characterisation the MOS is compared to this RMOS. Although the interpretation of the ratio MOS versus RMOS (MRR) is outside the scope of EUSES 2.0, in general there will be concern if the MOS is well below the RMOS. EUSES will mark MRR-levels below one in red. If the MOS is well above the RMOS there will usually be no concern. If the MOS is in the range of the RMOS, a thorough evaluation of the aspects that can only be dealt with in a qualitative way, as well as the uncertainties in the exposure estimate, is required. Next to the MOS approach, for biocides the TNsG (Technical Notes for Guidance; EC, 2002b) requires the derivation of Acceptable Operator Exposure Levels (AOELs) and their comparison with corresponding exposure values. The AOEL is a health-based exposure limit for operators and bystanders. It relates to the internal (absorbed) dose available for systemic distribution from any route of absorption and is expressed as internal level (mg/kgbw/d). The AOEL is based on the highest level at which no adverse effect is observed in tests in the most sensitive relevant animal species, or, if appropriate data are available, in humans. As default-procedure, the AOEL is based on the NOAEL (or exceptionally, the LowestObserved-Adverse-Effect level, LOAEL) from an oral short-term toxicity study (28- or 90-day study), which is to be converted to an internal dose by correction for systemic bioavailability. To translate this internal N(L)OAEL into an AOEL, assessment factors accounting for uncertainties in the extrapolation from experimental data to the human situation have to be applied: hence, the internal N(L)OAEL is to be divided by the overall assessment factor or RMOS. For risk characterisation the AOEL is compared to the internal exposure values of corresponding time scale. Although the interpretation of the ratio AOEL versus internal exposure (AER) is outside the scope of EUSES 2.0, in general there will be concern if the internal exposure is well above the AOEL. EUSES will mark AER-levels below one in red. If the internal exposure is well below the AOEL there will usually be no concern. EUSES 2.0 can be used to perform the above described risk characterisation. Substance specific values for the AFs are preferred as input for EUSES 2.0 above defaults. Various default values for assessment factors have been proposed in the updated TGD: see Table 3.4. These factors are based on recent knowledge on the distributions of extrapolation factors (USEPA, 1993; BAUA, 1998; Kalberlah and Schneider, 1998, IPCS, 1999; Vermeire et al., 1999a and 2001b; Kalberlah et al, 1999; DK-EPA, 2001, ECETOC, 2003; Schneider et al., 2004). The default assessment factors can be found in the EUSES 2.0 background report as well as in the Help-function.

60


3.3.5.2 Non-threshold effects For non-threshold-based effects (e.g. genotoxicity, genotoxic carcinogenicity), the MOS approach is not applicable, and an AOEL or NOAEL generally cannot be set. Two methods of risk characterisation are described in the updated TGD and incorporated in EUSES 2.0: 1. Determination of the lifetime cancer risk. The T25 derived (i.e. the chronic dose rate that will give 25% of animals tumours at a specific tissue site after correction for spontaneous incidence within the standard life time), possibly converted to the appropriate route of human exposure, by default is first extrapolated to an equivalent human dose (HT25) applying allometric scaling factors for the interspecies conversion and, if deemed necessary, a route-to-route assessment factor (Table 3.4). The resulting HT25 and the estimated lifetime daily exposure are then used to derive the lifetime cancer risk (LR). This lifetime cancer risk is to be evaluated against existing risk management criteria. 2. MOE-approach The Margin of Exposure (MOE) approach is equivalent to the MOS-approach for threshold substances. The MOE is the ratio of the selected dose descriptor, T25 or BMD05 (Bench Mark Dose: lower confidence limit of the dose that produces a response of 5%), possibly converted to the appropriate route of human exposure, and the estimated lifetime daily exposure. Next, the MOE should be compared to the reference MOE (RMOE). Like the reference MOS, the reference MOE should account for the various uncertainties and variabilities in the extrapolation from experimental data to the human situation (Table 3.4), and for the uncertainties in the exposure estimate of the exposure scenario under consideration. Additionally, a specific assessment factor should be used for low risk extrapolation since, contrary to the risk assessment for threshold effects, for non-threshold effects a dose without effect cannot be derived. Therefore, a target margin between the high risk related to the dose descriptor and a very low risk for the population exposed needs to be set. The magnitude of the low risk extrapolation factor is policy-driven. EUSES 2.0 applies by default a low risk extrapolation factor of 250,000. This factor is derived from linear extrapolation of the LR of 25:100, associated with the T25, to a default low reference risk level of 1:1,000,000. All aspects that can be dealt with quantitatively (as assessment factors) are combined to form the overall assessment factor RMOE. In judging the acceptability of the MOE, in a second step of the quantitative risk characterisation the MOE is compared to this RMOE. Although the interpretation of the ratio MOE versus RMOE (MRR) is outside the scope of EUSES, in general there will be concern if the MOE is well below the RMOE. In this case EUSES 2.0 marks the MRR in red. If the MOE is well above the RMOE there will usually be no concern. If the MOE is in the range of the RMOE, a thorough evaluation of the aspects that can only be dealt with in a qualitative way, as well as the uncertainties in the exposure estimate, is required. 61

3


Chapter 3


Table 3.4: Default human assessment factors Assessment factor Interspecies Intraspecies Exposure duration

Route-to-route extrapolation Dose-response a

 correction for metabolic differences  remaining differences  worker  general population  subacute to sub/semi-chronic  sub/semi-chronic to chronic  subacute to chronic difference between human and experimental animal exposure route issues related to reliability of the dose-response, incl. LOAEL/ NAEL extrapolation and severity of effect

Default value ASa,b 2.5 5 10 3 2 6 1 1

AS = allometric scaling factor: 4 for rat, 7 for mouse, 3 for guinea pig, 2.4 for rabbit, 2 for monkey

and 1.4 for dog. b

AS = 1 in case of inhalation, diet study or local effects

3.4 Example: EUCIDE Since the major new elements of EUSES 2.0 are expressed in the risk assessment for biocides, a hypothetical biocide case is presented. The assumed properties and characteristics of the active substance in this biocide, EUCIDE, are shown in Table 3.5. EUCIDE is used as a sanitation disinfectant for private use. It is a ready-to-use spray formulation for use on surfaces that come into contact with food, drinks and their raw materials.

62


Table 3.5: Properties of EUCIDE, EUSES input Parameter Product identification General name of a.i. Chemical class of a.i. Product type of biocide Use

Value

Unit

Concentration of a.i. Density of formulation

EUCIDE non-hydrophobics PT 2.2 disinfectant private use sanitary sector 0.17 0.70

g.l-1 g.cm-3

Physical properties. Molecular weight Melting point Boiling point Vapour pressure at 25 ºC Octanol-water partition coefficient Water solubility at 25 ºC

100 -18 101 0.44 -0.33 1E+6

g.mol-1 ºC ºC hPa log10 mg.l-1

Fate Biodegradability Half-life for hydrolysis at 12 ºC Half life for photolysis 12 ºC

readily biodegradable 193.6 19.6

d d

Ecotoxicological properties EC50 for micro-organisms in a STP LC50 for fish EC50 for algae LC50 additional taxonomic group NOEC for fish NOEC for algae NOEC for additional taxonomic group LC50 in avian dietary study NOEC via feed of birds

51 32 0.51 7.5 1.3 200 0.78 5000 2500

mg.l-1 mg.l-1 mg.l-1 mg.l-1 mg.l-1 mg.l-1 mg.l-1 mg.kgbw-1.d-1 mg.kgfeed-1

Human-toxicological properties NOAEL, gavage, 28 days, rats NOEC, diet, 90 days, rats NOAEL, inhalation, 28 days, rats NOAEL, dermal, 90 days, rats

5 55 0.25 100

mg.kgbw-1.d-1 mg.kgfeed-1 mg.m-3 mg.kgbw-1.d-1

3

3.4.1. Environmental risk assessment The life cycle stages formulation and private use are applied for calculating the releases of EUCIDE to the environment. The formulation stage is based on the Emission Scenario Document (ESD) for ‘personal/domestic use and public domain’ (industrial category 5 and 6), with emission fractions to water and air that are applicable to washing liquids. The emission scenario for biocidal product type 2, ‘use of disinfectants in the sanitary sector’, is

63


Chapter 3


used to calculate the release at private use, which also includes releases at waste treatment. For this emission scenario, application data are used that are supplied by the notifier. There was no information available to use the alternative scenario for this biocidal product type that is based on a use tonnage. The emission scenario parameters for both life cycle stages are shown in Table 3.6. Table 3.6: The emission scenario parameters for formulation and private Parameter description Regional tonnage of substance (t/y) Fraction of chemical in formulation (%) Tonnage of formulated product (t/y) Fraction of tonnage released to air Fraction of tonnage released to waste water Fraction of main local source Number of emission days Daily consumption per inhabitant (ml/d) Active substance in product (mg/l) Penetration factor of disinfectant (-) Local emission to air during episode (kg/d) Local emission to waste water during episode (kg/d)

Formulation 1 1.7 58.8 2.10-5 9.10-4 1 118 Not applicable Not applicable Not applicable 1.69.10-4 7.63.10-3

1)

s s o o o o o

o o

Private use 0 1 365 5 170 0.5 0 4.25.10-3

1)

o o o s d o

1) Classification of the parameter: s; user input data set, o; output from EUSES calculation, d; standard default value

The exposure assessment for biocides can be carried out for the local scale only (see section 3.3.1). The effects assessment is based on toxicity data for micro-organisms, freshwater organisms and birds (secondary poisoning). Marine toxicity data are based on freshwater data, using marine specific assessment factors (TGD). The sediment and soil toxicity data are derived via equilibrium partitioning from the freshwater data (TGD). An overview of the effect data is presented in Table 3.5. The results of the risk assessment are shown in Table 3.7. It can be concluded that there is no risk for the environment, which is based on the PEC/PNEC ratios for all applicable compartments. It must be noted that regional background data are not taken into account. Nevertheless, it is not expected that the environmental risk ratios will be higher than 1, when regional data is included in the assessment. The highest risk is calculated for the marine environment with a PEC/PNEC ratio of 0.2. This ratio is about a factor of 100 higher than calculated for the freshwater environment. The calculated concentrations in the freshwater and marine environment are about the same. Therefore, the difference is mainly caused by the higher assessment factor used for the marine effects assessment

64


Table 3.7: PEC/PNEC-ratios of the local environmental risk assessment End point Freshwater Marine water Freshwater sediment Marine water sediment Soil Sewage treatment plant Fish Marine fish Marine top-predators Worm

Formulation 2.5.10-3 0.19 2.5.10-3 0.19 1.3.10-3 9.5.10-4 2.2.10-5 1.5.10-5 3.1.10-6 4.7.10-6

Private use 1.4.10-3 0.11 1.4.10-3 0.11 7.2.10-4 5.3.10-4 3.5.10-5 2.5.10-5 5.2.10-6 2.6.10-6

3.4.2. Human risk assessment Humans are assumed to be exposed to the active substance in the product by inhalation during and after spraying and through dermal contact of the hands with the droplets in the spray. At first, the simple realistic worst case EUSES 2.0 consumer exposure scenarios are used: the Constant Concentration scenario for inhalation and the Fixed Volume model for dermal contact (OECD, 1993). The constant concentration model calculates the exposure by assuming that all the active substance in the biocide is immediately released into the air and homogeneously mixed. After that the concentration is taken to be a constant in the time, since there is no ventilation included in the calculation. In the Fixed Volume model dermal uptake is calculated from a well mixed, fixed volume of product contacting a certain area of the skin. The finite volume of the product sets the maximal amount of substance that can be taken up. The models are applied to users spraying the formulation on a kitchen-working top. The spray is used for 10 minutes and the user stays in the room for another 50 minutes. The scenario parameters and the results are shown in Table 3.8. The risk characterisation is based on the comparison of the estimated Margin of Safety with a reference value (RMOS). In addition, a comparison is made between the AOEL and the estimated internal exposure. In this example, the MOS/RMOS and the AOEL/internal exposure ratios have the same value. This is because the basis for the derivation of the RMOS and the AOEL is the same inhalatory NOAEL, default assessment factors only are used for the derivation of the RMOS (Table 3.8) and absorption factors are all assumed to be 100%.

65

3


Chapter 3


Table 3.8: Private use scenario, EUSES screening models* Parameter INPUT Scenario for inhalation Scenario for dermal contact Time scale Number of events Duration of use per event Duration of exposure per event Amount of product used - Volume of product used - Surface area kitchen table top - density of formulationn Breathing space (room volume) Respirable fraction Absorbed inhalation/dermal fraction Inhalation rate user Body weight user

Value

Unit

constant concentration model fixed volume model chronic 1 10 60 175 50 5 0.7 5 1 1 20 70

d-1 min min g cm3.m-2 m2 g.cm-3 m3 m3.d-1 kg

5.95 0.0708 0.0277 0.248 0.0985

mg.m-3 mg.kgbw-1.d-1 mg.kgbw-1.d-1 mg.m-3 mg.kgbw-1.d-1

OUTPUT Exposure Concentration in air of room Inhalatory intake Potential dermal uptake Annual average inhalation exposure Total chronic external exposure Risk characterisation inhalation MOS RMOS MOS/RMOS AOEL AOEL/internal exposure Risk characterisation dermal MOS RMOS MOS/RMOS AOEL AOEL/internal exposure Risk characterisation total exposure MOS RMOS MOS/RMOS AOEL AOEL/internal exposure

1.01 150 0.0134 0.00167 0.0672 3610 600 6.02 0.167 6.02 50.8 600 0.0846 0.00833 0.0846

* Input values based on EUSES defaults, the product prescription and data from Bremmer et al., 2002.

Based on these risk characterisation ratios ( 1 for dermal exposure) it may be concluded that there is a risk for inhalation exposure to EUCIDE. In view of the screening type of risk assessment it can be decided to refine the exposure assessment first. The cloud model of CONSEXPO 3.0 (Van Veen, 2001) is subsequently used in interaction with EUSES 2.0. EUSES 2.0 provides a link with

66


CONSEXPO 3.0 and the results of CONSEXPO can be transferred back to EUSES 2.0. Whereas the constant concentration model of EUSES assumes immediate evaporation and homogeneous mixing across the room without ventilation, the cloud model assumes exposure of the user to the chemical in the cloud composed of both spray droplets and vapour from the target area in a ventilated room. The evaporation rate is derived using the surface of the droplets and the target area and taking into account that the concentration in the formulation decreases. The scenario parameters and the results are shown in Table 3.9. Based on the ratio of the MOS and the RMOS and the ratio of the internal exposure estimate and the AOEL, it can now be concluded that there is still a risk from inhalatory exposure to EUCIDE as a sanitary disinfectant under the conditions prescribed.

3

Table 3.9: Private use scenario, CONSEXPO model and EUSES * Parameter INPUT Scenario for inhalation Time scale Number of events Duration of use per event Duration of exposure per event Emission rate formulation (Amount of product used Weight fraction of substance Surface area kitchen table top Airborne fraction Droplet size Cloud radius Release height Ventilation rate room Breathing space (room volume) Respirable fraction Absorbed inhalation fraction Inhalation rate user

Value

Unit

cloud model chronic 1 10 60 0.291 175 0.24 5 0.15 15 20 30 12.5 5 0.1 1 20

d-1 min min g.sec-1 g) g.kg-1 m2 µm cm cm m3.h-1 m3 m3.d-1

Exposure Concentration in air of room Inhalatory intake Annual average inhalation exposure

1.669 0.00201 0.07036

mg.m-3 mg.kgbw-1.d-1 mg.m-3

Risk characterisation inhalation MOS RMOS MOS/RMOS AOEL AOEL/internal exposure

3.55 150 0.0237 0.00167 0.0237

OUTPUT

* Input values based on EUSES defaults, the product prescription and Bremmer et al., 2002.

67


Chapter 3


3.5 Discussion This synopsis on new features of EUSES shows that significant improvements and additions have been made both scientifically and with regard to the user interface. This is done on the basis of numerous comments received from users of EUSES 1.0 and from risk assessment and modelling experts. Acknowledgements should go to all of them. This interaction should continue and be encouraged. In this way EUSES will be a living harmonised risk assessment tool. The launching of a new version is also a good moment to reflect on the validation status of the models used. EUSES 1.0 has been the object of several validation projects that have been mentioned in the introduction. These have been scientific validations largely concentrating on the reliability, accuracy and usefulness of the individual models in EUSES within the specified field of use as well as on the quality of the documentation. It was noted by Jager et al. (1998a) that validation activities for individual models are seldom directly applicable to EUSES 1.0, since it is a generic instrument, using fixed standard scenarios and averaged-out variations in time and space. Measured data used for validation are often nonrepresentative for the standard scenarios used. As a consequence, the degree of deviation from reality can be substantial, especially with regard to release estimation, environmental effects assessment and workplace exposure assessment. Nevertheless, the models used are state-of-the-art for substance risk assessment and EUSES provides conservative estimates for the standard scenarios based on limited data requirements. Schwartz et al. (1998) concluded that the EUSES 1.0 software basically fulfils the postulated quality criteria, but was lacking in transparency due to high complexity, low modularity and incomplete documentation. The performance of the model was characterised as a good compromise between complexity and practicability. Both publications stress the limited usefulness of EUSES for chemicals outside the domain of the persistent, non-dissociating substances of intermediate lipophilicity. Most of these conclusions will still apply to the second version of EUSES, but the transparency has improved since documentation is available now on all models. The emission module has been improved significantly by adding the available emission scenario documents and by improvements in the user interface. The complexity of EUSES is in one way unavoidable since it covers the whole risk assessment chain from releases and exposure to effects and risks for both humans and the environment in an integrated way. Moreover, new elements in the updated TGD of 2003 had to be incorporated. On the other hand the user friendliness could be improved further in the future. Useful ideas have been proposed, the realisation of which is limited at present by the available resources. Under the new EU chemical legislation REACH (Registration,

68


Evaluation and Authorisation of CHemicals; EC, 2003c) the number of users with limited knowledge of risk assessment will increase and simple tools should be available for a first tier assessment, targeted towards the needs of producers, importers and downstream users. Although it is already possible to choose specific parts of the risk assessment (e.g. environmental and/or human, local and/or regional, sewage treatment on or off, consumer exposure assessment only and/or worker exposure assessment only), the modular structure could be extended for instance by allowing running of submodels or allowing simplified input via an interface file. The Help function could be improved further and extra pop-up fields for user comments added. Apart from improvements in user friendliness, model improvement should have continuous attention. EUSES should incorporate major scientific developments provided these are accepted by the research and regulatory community. Models to be considered could, among others, stem from the European Chemical Industry’s Long-range Research Initiative (CEFIC, 2004) or the Global Net on Consumer Exposure Modelling (IHCP, 2004). Some areas where improvements are considered needed are: • Emission estimation: the current emission module should be adapted to answer the needs of users: it should incorporate relevant ‘Emission Scenario Documents’’ adopted by OECD or EU and be adapted to the ‘exposure scenario’ concept being developed under the REACH-legislation. • Exposure of humans through the environment: the models used for this estimation need improvement, especially with regard to bioaccumulation in meat and milk and the exposure via drinking water (Rikken and Lijzen, 2004). • Consumer exposure assessment: development, validation, data storage and dissemination of exposure factors and validation and harmonisation of consumer exposure models. • Uncertainty analysis: incorporation of the possibility to perform sensitivity and uncertainty analysis could provide more quantitative insight into the range of possible outcomes and therefore better inform both risk assessors and risk managers (Jager et al., 2001a; Vermeire et al., 2001a). The above mentioned increase in users with limited knowledge of risk assessment also triggers a warning. Although simplifications can be made as specified above, a certain amount of expertise is definitely required for users of risk assessment tools like EUSES. Users should at least be able to make a quality check on the input data, to understand the risk assessment process and to evaluate and communicate the final results.

69

3


Chapter 3


3.6 Acknowledgements The EUSES projects have been dependent on and stimulated by the input of experts from several national and international organisations. Scientists from the Netherlands National Institute of Public Health and the Environment (RIVM) were responsible for the scientific aspects of the development of EUSES. The UK Health and Safety Executive (HSE) developed the workers exposure module in EUSES. The efforts of all persons involved are gratefully acknowledged. The development of EUSES 1.0 was commissioned by the Netherlands Ministry of Housing, Physical Planning and the Environment (VROM), the European Commission, the UK Department of the Environment (DoE), the Danish EPA, the German Umweltbundesamt (UBA), and the Swedish National Chemicals Inspectorate (KEMI). The EUSES 2.0 project was commissioned by the European Chemicals Bureau of the JRC Institute of Health and Consumer Protection.

70

4 Opportunities for a probabilistic risk assessment of chemicals in the European Union

Tjalling Jager, Theo Vermeire, Mathieu Rikken, Paul van der Poel RIVM, The Netherlands

Published in: Chemosphere 43:257-264 (2001)

Chapter 4


4. Opportunities for a probabilistic risk assessment of chemicals in the European Union. In the EU risk assessment of industrial substances, it is current practice to characterise risk using a deterministic quotient of the exposure concentration, or the dose, and a no-effect level. A sense of uncertainty is tackled by introducing worstcase assumptions in the methodology. Since this procedure leads to an assessment with an unknown degree of conservatism, it is advisable to deal quantitatively with uncertainties. This paper discusses the advantages and possibilities of a probabilistic risk assessment framework, illustrated with an example calculation. Furthermore, representatives of EU Member States and the chemical industry were interviewed to find out their views on applying uncertainty analysis to risk assessment of industrial chemicals.

72

Opportunities for a probabilistic risk assessment of chemicals in the European Union

4.1 Introduction Chemical risk assessment aims to protect humans and the environment from the possible adverse effects of substances. In the European Union, the methods for risk assessment of new and existing chemicals are harmonised between the Member States and laid down in Technical Guidance Documents (TGD) (EC, 1996a). The TGDs, in turn, are implemented in a computerised decision-support system: the European Union System for the Evaluation of Substances (EUSES) (EC, 1996b; Vermeire et al., 1997). The assessment should protect all ecosystems and human populations from unacceptable risks but, for pragmatic reasons, several groups are explicitly defined for which a risk estimate is made; the socalled protection targets (Table 4.1). The measure of risk that is used in this framework is a point estimate; the ‘risk characterisation ratio’ (RCR), which compares the results of exposure and effects assessment. For environmental endpoints the RCR is the PEC/PNEC ratio (quotient of the Predicted Environmental Concentration and the Predicted No-Effect Level for an endpoint) and for human populations a MOS (Margin Of Safety, quotient of a toxicity measure -e.g. an NOAEL- and the exposure level). Table 4.1: Protection targets in the TGD/EUSES. Human populations Workers Consumers Humans exposed via the environment

Ecological systems and populations Micro-organisms in sewage treatment systems Aquatic ecosystems Terrestrial ecosystems Sediment ecosystems Predators via fish and earthworms

These RCRs are, however, not true risk levels. Risk is generally defined in the scientific community as the magnitude of an ‘impact’ times the probability that this impact will actually occur. Unfortunately, we cannot usually define a risk in this strict sense in chemical risk assessment because the probabilities are not routinely quantified, but first of all, because impacts are not properly defined. We can only indicate how much a certain ‘noeffect level’ (the PNEC or an NOAEL) is exceeded at a certain exposure level (PEC or a dose). Because the dose-response relationship for the protection targets is unknown, the absolute magnitude of the RCR cannot be interpreted and chemicals cannot be properly compared on this basis. Point estimates like these ratios may be efficient in a first stage to focus on the most important contaminants and emission sources (Bartell, 1996) but the disadvantages are numerous. It is impossible to determine where the point estimate lies in the range of possibilities, the point estimates give a false sense of accuracy and ignore variability in the population (see detailed discussion in Thompson and Graham, 1996).

73

4


Chapter 4


The minimum data set on which a risk assessment can be based is legally prescribed as the so-called EC Base Set. Apart from information on production volume and usage of the chemical, this set contains results of standard tests on physico-chemical data, biodegradation and toxicity. Clearly, the results based on such a limited data set must be accompanied by a fair amount of uncertainty caused by measurement errors, lack of knowledge, model assumptions and also natural variability. Up till now, risk assessors have reacted upon this sense of uncertainty by introducing worst-case assumptions in the methodology. This is a potentially dangerous situation as a multiplication of worst cases could eventually lead to unrealistic assessments and this is neither transparent, nor efficient when it is inducing unnecessary further testing or risk reduction measures. At least from a scientific point of view, it is advisable to quantify this uncertainty and take it explicitly into account in the decision-making process. Quantitative uncertainty analysis is a tool that can be used to tackle the propagation of uncertainties in a more scientific manner. There are numerous studies were the power of quantitative uncertainty analysis is demonstrated in the chemical risk assessment domain (e.g. McKone and Ryan, 1989; De Nijs and De Greef, 1992; Traas et al., 1996; Copeland et al., 1993). Nevertheless, the application of uncertainty analysis to decision making is far from routine as virtually all decisions are still based on point estimates of exposure and effects. One of the reasons for the reluctance of regulators to accept probabilistic risk assessment is the lack of proper guidance and policy (Finley et al., 1994) although regulators in the U.S. have already started addressing these issues (see e.g. US EPA, 1997). Of course, there is always a discomfort in risk assessment when the scientific process meets the legal one; decision makers are usually not statisticians and may feel ill at home with probability distributions. Instead of focusing on the statistical technicalities for uncertainty analysis, due attention should be paid to transparency, presentation and interpretation of uncertain end results to allow the risk managers to make informed decisions. The purpose of this paper is to discuss how chemical risk management can benefit from a probabilistic risk assessment. The work presented in this paper is described in more detail in two reports (Jager et al., 1997; Jager, 1998b).

4.2 Choice of uncertainties to include in the assessment The interpretation of probabilistic model results hinges strongly on the uncertainties that are accounted for. Using the same model, two scientists may get very different distributions in the end, depending on the choices and assumptions they make with respect to input distributions. Therefore, these assumptions must be clear to the risk manager to allow proper interpretation and decision making. The possible sources of uncertainty should be

74


identified and a selection must be made with regard to the uncertainties to be included in the assessment. One source of uncertainty is the uncertainty caused by lack of knowledge. This type of uncertainty can (in theory at least) be diminished by further research. An example is emission estimation. When these estimates are solely based on the use pattern of the chemical and the production volume (as is generally the case for new chemicals), the uncertainty is large but can be decreased by monitoring or site-specific calculations. Another obvious source of uncertainty is the natural variability in time or space. The world around us is inherently variable which will also influence the risk for the protection targets. As an example: the flow rate of a river will vary between different rivers but also according to seasonal influences. It is important to note that this variability cannot be reduced by further research, it can only be characterised more accurately. Risk assessments in the framework of the TGDs are (initially) not performed for an existing location but for a so-called standard environment by using a “reasonable worst-case” exposure scenario. The uncertainty or variability in the choices made for this scenario is extremely difficult to quantify (e.g. the variability in the exposure location for inhalation by humans). The influence of uncertainty and variability should, however, not be mixed but must be considered separately (Hoffman and Hammonds, 1994). There are more sources of uncertainty in model results. A model is a simplification of reality and the simplifications themselves also constitute a source of uncertainty. Furthermore, the decisions about the system and situation to be modelled (e.g., the selection of the protection targets as shown in Table 4.1) can be considered a source of uncertainty. These are extremely difficult to quantify and are usually ignored in quantitative analyses. This illustrates that it is of utmost importance that the user of the end results of a probabilistic risk assessment is aware of the uncertainties that are excluded and how the others are included.

4.3 Integrating uncertainty in Risk characterisation As stated in the introduction, RCRs are not estimators of risk but only indicate whether a certain no-effect level is exceeded. As inheritance from the methods for setting ‘safe’ environmental quality standards, we are left with a tendency to aim for conservative noeffect levels rather than quantifying actual impacts. The impact of any exceedance of this PNEC is unknown as a dose-response relationship is lacking. A risk-based comparison of chemicals (priority setting) is even further hampered as the uncertainty in the risk quotient is unknown. The sources and magnitude of uncertainties in the model results will differ between chemicals. They depend on the properties of the chemical, which means that some substances can be assessed with greater confidence than others. This is illustrated in Figure

75

4


Chapter 4

4.1: chemical A and B share the same median PEC/PNEC ratio (lower than 1) but their probability distribution is very different. This implies that a chemical with a PEC/PNEC ratio of 100 is not necessarily more risky than a chemical with a ratio of 10. Nevertheless, in the risk characterisation for new substances, decisions are based on the absolute threshold values of 10, 100 or 1000 of the PEC/PNEC ratio (EC, 1996a). For human effects assessment, there is no generally accepted extrapolation to no-effect levels in the EU6. The available effects data from a mammalian laboratory test or a human epidemiological study are used directly in comparison with the exposure level. The resulting Margin Of Safety (MOS) is evaluated on the basis of expert judgement. Although this process leaves a lot of freedom for interpretation, it lacks transparency. The TGD furthermore provides no quantitative guidance in the interpretation of the MOS and this procedure may thus lead to inconsistency in the risk assessments of individual Member States (Vermeire et al., 1999a).

Probability density


A B

-4

-2

0 2 log PEC / PNEC

4

Figure 4.1: Probability distributions of two hypothetical chemicals with the same median PEC/PNEC ratio

6 Changes in the TGD version discussed here relative to the REACH TGD are discussed in Section 7.2.

76


What is lacking in chemical risk assessment is a dose-response relationship for the protection targets. If such a relation could be established, we would be able to attach an effect level to a certain exposure level. Uncertainty about the exposure will thus result in a probability distribution of effect levels, in other words: a risk in the strict sense. For human populations, the type of effect that can be expected is usually evident from animal studies (although this does not necessarily mean that the same effect occurs in humans or that it is the most sensitive effect) but there is a lack of reliable dose-effect information. For ecosystems, our scientific knowledge is still too limited to predict the nature and the extent of the impacts that chemicals may have (Power and McCarthy, 1997). Although the effect of a chemical on a single species can be tested in the laboratory, the impact in the field cannot simply be extrapolated from these tests. In the field situation, organisms will interact with others in an often complex manner. Clearly, if true risks are to be characterised, the effects assessment is the critical stage. The following options are available to revise the effects assessment in order to deal with uncertainties in the risk characterisation stage. These options result in different interpretations of the final ‘distribution of risk’ that the risk managers are confronted with (schematically drawn in Figure 4.2). In order of preference (in view of the definition of ‘risk’): A. Establish a dose-effect relationship for human populations and ecosystems. The result of the risk characterisation stage will be a probability distribution of effects. Decisions can be based on an acceptable level of effects. An example of such an approach for ecosystems is the use of the variability in sensitivity of (laboratory) test species as a measure of the variability of all species in the ecosystem (Aldenberg and Slob, 1993). In that way, for each exposure level, a fraction of species exposed above its NOEC can be calculated (Klepper et al., 1998). This approach is however not broadly accepted as representation of ecosystem impact. B. Revise the assessment factors in the effects assessment to yield a median, or most likely, PNEC instead of a conservative estimate and attach uncertainty to these factors (e.g. instead of a factor of 1000 use an assessment factor of 100 with a factor of 10 uncertainty). The result of the risk characterisation will be a probability distribution of PEC/PNEC ratios (Bartell, 1996). This approach will not quantify true risks but at least acknowledges the uncertainty in the effects assessment. However, some research is needed to quantify proper distributions for the assessment factors (e.g. by analysing existing databases). C. Leave the effects assessment as it is now. In that case, only uncertainty in the exposure estimate can be quantified. The result of the risk characterisation will be a probability that PEC exceeds a fixed, worst-case PNEC. This option has the advantage that it

77

4


Chapter 4

is easy to perform and is more acceptable for the decision makers (Jager, 1998). However, one is still confronted with an ill-defined and conservative PNEC whose uncertainty can be considerable. A Acceptable Effects

Effects

probability density


B 1

PEC / PNEC C PNEC

PEC

Figure 4.2: Options for uncertainty in risk characterisation

4.4 Example calculation In this section, an example calculation for a new chemical (a chemical intermediate) with EUSES is given to illustrate the way probabilistic methods will influence the risk assessment. Monte Carlo analysis was applied which is a simple and straightforward method but requires a precise characterisation of parameter distributions, even when the underlying empirical information is actually insufficient. This characterisation is usually done using ‘a combination of professional judgement, limited empirical information, and blind faith’ (Moore, 1996). The difficulty of selecting appropriate parameter distributions may be another reason for lack of acceptance of probabilistic risk assessment among

78


regulators (Finley and Paustenbach, 1994). The final distribution of the risk estimate depends strongly upon the selected input distributions but the validity or appropriateness of a given distribution is difficult to judge for a risk manager. Clearly, there is a need for a default set of agreed parameter distributions for chemical risk assessment. For the sake of simplicity and transparency, we restricted the example to the uncertainty in chemical-specific parameters; uncertainties in the scenarios are excluded. For the chemical-specific parameters, uncertainty distributions were characterised as shown in Table 4.2 (specific information in Jager et al., 1997). By default, log-normal distributions are selected as there are strong theoretical and empirical considerations to assume this distribution a priori for many physical entities (Slob, 1987; Seiler and Alvarez, 1996). The log-normal distribution is less suited for parameters that have an upper or lower limit like release fractions or absorption factors for human exposure. In those cases, triangular or uniform distributions were used as a rough estimate of uncertainty. The characteristics of the environment are defined as “scenario properties”, thereby ignoring any uncertainty or variability in them. In this way, the resulting distribution indicates a risk given the worstcase exposure scenario. In many cases, this distinction between chemical-specific and scenario properties is not completely straightforward: many parameters both depend upon properties of the chemical as well as on the environment (e.g. biodegradation rates). Table 4.2: Derivation of parameter distributions for the example uncertainty analysis Group of parameters Physico-chemical data Release estimation Partition coefficients Degradation rates Distribution models Bioconcentration factors Absorption factors Assessment factors

Source data from handbooks expert judgement data from open literature (usually QSARs) expert judgement expert judgement data from open literature (usually QSARs) expert judgement Literature, databases and expert judgement

Type of distribution log-normal triangular log-normal log-normal log-normal log-normal uniform log-normal

As example of the derivation of a parameter distribution, the estimation of the organiccarbon normalised partition coefficient (Koc) is illustrated in Figure 4.3. The Koc is estimated from the octanol-water partition coefficient (Kow) with a log-linear regression (Sabljic et al., 1995). The original data (the training set) were used to quantify the uncertainty in the estimate from the residuals. The standard deviation of the residuals is used to define the best fitting log-normal distribution. The residuals increase with increasing Kow and therefore a distinction was made between the uncertainty for log Kow less than 4 and more than 4 (geometric standard deviation of 1.8 and 3.9, respectively).

79

4


Chapter 4 㜀

㄀

㘀㔀

　

㐀

爀攀猀椀搀甀愀氀猀

氀漀最䬀漀挀


㌀㈀

ⴀ㄀

㄀　

ⴀ㈀　

㄀

㈀

㌀

㐀

㔀

㘀

㜀

㠀

氀漀最䬀漀眀

　

㄀

㈀

㌀

㐀

㔀

㘀

㜀

㠀

氀漀最䬀漀眀

Figure 4.3: QSAR regression for Koc as function of Kow (left figure) and the residuals from this regression (right figure) For now, only a risk assessment at the local scale (in the vicinity of a point source) is examined since this scale is usually dominant for new chemicals. In this case option B from Figure 4.2 is worked out (distributions are assigned to the assessment factors as well as the exposure factors). All calculations were performed with an Excel spreadsheet of EUSES; Monte Carlo sampling was performed with Crystal Ball using 1000 runs and Latin Hypercube sampling. The results of the uncertainty analysis are presented graphically as cumulative probability distributions in Figure 4.4 and are summarised in Table 4.3. These distributions must be interpreted as follows: the probability that the RCR of a sensitive human is below 1 is 65% (the line of the RCR sensitive human crosses the line log RCR=0 at the 65th percentile). This implies that there is 35% probability that the RCR exceeds 1. Furthermore, one can be 90% sure that the RCR is lower than 10. Of course, all these probabilities are conditional chances: they represent the probability given the standard exposure scenario and given our assumptions of parameter distributions. Table 4.3: Deterministic EUSES risk estimate and the percentiles of the risk distribution shown in Figure 4.4 Water

a

EUSES RCR

413

50% 80% 90% 95%

37 190 523 1140

p(RCR>1) Location of EUSES RCR

97% 88%

Soil

Fish-eater Worm-eater Deterministic PEC/PNEC ratio 286 1.17 5.68 Percentiles of PEC/PNEC distribution 15 5.2 2.5 73 56 31 189 176 95 401 430 230 Probabilities 92% 74% 64% 93% 28% 60%

Human health a

0.39 3.5 10 32 36% -

EUSES estimates a MOS of 23 which, given the nature of the NOAEL used, gives reason for

concern. 80


100%

Cumulative percentage

90% 80%

RCR water

70%

RCR soil RCR fish-eater

60%

RCR worm-eater

50%

RCR sens. human

40%

RCR=1

30%

90th perc.

20%

80th perc.

10% 0% -5

-4

-3

-2

-1

0

1

2

3

4

5

log RCR

Figure 4.4: Cumulative probability distribution for the RCR. Clearly, the degree of uncertainty (the width of the distribution) differs between the endpoints. For the aquatic and terrestrial ecosystem, the distribution spans approximately three orders of magnitude (5-95th percentile) whereas for humans, this is nearly four orders of magnitude. This is to be expected as the exposure estimation for the aquatic and terrestrial endpoints covers less calculation steps. Sensitivity analysis reveals that the combined assessment factors dominate the total uncertainty in all endpoints, followed by emission data and, for predators, the bioconcentration factors. The median risk estimate is generally much lower than the deterministic PEC/PNEC ratio of EUSES. Only for predators the EUSES estimate is comparable or lower than the median of the distribution. This is caused by the fact that the proposed uncertain extrapolation of the laboratory toxicity data includes lab-field difference in caloric content of the food and metabolic rates (this is more stringent than the factors proposed in the TGD). The risk estimate produced by EUSES for the other end-points lies at the 88-93th percentile of the distribution. For humans, an RCR is calculated using empirical distributions for the assessment factors (Vermeire et al., 1999a) to arrive at a No-Effect Level for a sensitive human individual. In this way, the risk to humans can be directly compared to the other endpoint From Figure 4.4 it can be concluded that there is reason for concern for this chemical. Of course, this can also be concluded based on the EUSES RCRs in Table 4.3 but the distributions provide more insight in the degree of uncertainty and risk (e.g. in the probability that PEC will exceed PNEC). For all endpoints the 65th percentile of the risk quotient exceeds 1 (at least, using the standard exposure scenario). Clearly, the aquatic ecosystem is the most sensitive endpoint. For this endpoint, the main part of the uncertainty 81

4


Chapter 4


is caused by the combined assessment factors (62% of total uncertainty) although the release estimation is also important (in total 28% of total uncertainty). Further toxicity testing will therefore be most effective in reducing the uncertainty in the risk assessment (although it is unlikely that this is sufficient to get the RCR below 1).

4.5 Interviews with end-users The people involved in chemical risk assessment and management in the EU are hesitant to support probabilistic methods. For this reason, interviews were conducted with nine representatives of Member States and one from chemical industry to investigate whether a probabilistic framework is feasible and what type of further studies are necessary in this respect. The interviews were conducted in 1998 during an EU Technical Meeting on Existing Chemicals in Arona, Italy (full description in Jager, 1998). Several general conclusions could be drawn from all these interviews: • Implicitly, uncertainties are taken into account. Most risk assessors believe they have a pretty good feel when the point is reached where they can take a confident decision. • Currently, uncertainty analysis is not seen as a high-priority issue. A more accurate release estimation is currently the top priority as this is the stage where the largest uncertainties are expected. • It may not be necessary to base decisions on probability distributions but the distributions can be used to indicate the significance of the deterministic quotient. This may be a more feasible first step than a full-blown probabilistic risk assessment. • The main perceived disadvantages of uncertainty analysis are the time consumption and the fact that it suggests a higher degree of accuracy than you actually have: you still have the same data basis but the probability distributions suggest a considerable degree of scientific knowledge. • Parameter distributions must be accurately quantified. There seems to be little confidence in ‘expert-derived’ or ‘not-unreasonable’ parameter distributions. • Not only must the uncertainty in input parameters, but also natural variability be addressed. • Quantifying the uncertainties in the effects assessment has little support because the data basis and degree of knowledge are perceived as insufficient. Furthermore, it is felt that further testing can easily reduce these uncertainties, when necessary. • The chemical industry seems particularly interested in quantifying uncertainties to make risk assessment more realistic. • Despite their concerns, most representatives see uncertainty analysis as an interesting future development. Some example risk assessments would be useful, especially when it can be demonstrated how it will affect decision making.

82


Summarising, there seems to be a guarded interest in uncertainty analysis among most of the people interviewed although it does not receive high priority at this moment. There seems to be a gap between the scientist and the risk managers. In the scientific community, uncertainty analysis is broadly accepted as a necessity when presenting model results (just like confidence intervals when presenting calculations on experimental data). The risk manager, however, has to deal with the legal aspects and a decision must be reached within a certain time frame. A series of probability distributions, although very scientific, does not seem to be an obvious help in this process. The best way to proceed with the work on uncertainty analysis is to try to bring together these two fields: it must be demonstrated how risk management can benefit from the extra work needed in performing and understanding probabilities. A few probabilistic risk assessments can be performed as example and the results can be compared to the current TGD approach. At this moment, it is important to keep an uncertainty analysis simple and transparent and probably restrict it to the exposure assessment. Even though the example calculation indicates that uncertainties in the effects assessment may well dominate the total distribution, the support for this approach is very limited. Furthermore, some work is required to study how variability can be taken into account in a transparent manner.

4.6 Conclusions From a scientific point of view, there is sufficient argumentation to strive for a probabilistic framework in chemical risk assessment. In a probabilistic assessment one can include all the available information and postpone the necessary conservatism to the last moment (e.g. by taking a high percentile of the risk distribution). This is preferable to compromising the transparency by applying worst-case assumptions throughout the assessment. The example demonstrates that a probabilistic assessment is relatively easy to make and offers more relevant information. Especially the possibility to quantify the main sources of uncertainty offers a powerful tool for risk assessors to ask for specific further information from the notifier. However, all this information is also potentially confusing. Before uncertainties can be taken explicitly into account in risk management decisions, a high degree of transparency in the methods is needed to facilitate interpretation: the decision maker needs at least a basic understanding of the assumptions underlying the risk distribution. Furthermore, if true risks are to be quantified, the effects assessment is the critical stage and needs further consideration. It should be noted that an important contribution of uncertainty and variability is ignored in these examples: the exposure scenario. This scenario is currently worst case and thus influences the interpretation of the risk distribution. As reflected in the interviews, insight

83

4


Chapter 4


in the influence of this variability is needed. Unfortunately, variability in the exposure scenario is rather difficult to quantify and must be treated separately from uncertainty due to lack of knowledge. A simple possibility to visualise the effects of variability is by performing assessments with alternative, equally plausible, scenarios (leading to alternative distributions for an endpoint). Interviews revealed that the end users of the results of risk assessment are not very enthusiastic about these methods as they provide more work and potentially give rise to more discussion. Nevertheless, the interviewed feel that it is a future development that cannot be ignored but there is a need for transparent examples to demonstrate the effect of probabilistic risk assessment on decision making. Interestingly, there is little support for quantifying uncertainties in the effects assessment although the example indicates this may well dominate the total uncertainty. It is our opinion that probabilistic risk assessment is preferable to deterministic quotients to address the complex problems of chemical risk assessment and risk management. Nevertheless, it takes time to change the conventional and well-accepted approaches and risk managers as well as risk assessors need to familiarise themselves with probabilistic methods and distributions of risk. Perhaps it is possible to use a deterministic and a probabilistic approach side-by-side in a decision-support system. In this way, risk assessors and risk managers can compare the outcomes and grow more accustomed to dealing with probability distributions.

84

5 A probabilistic human health risk assessment for environmental exposure to dibutylphthalate

Theo Vermeire1, Tjalling Jager1, Geert Janssen1, Peter Bos2, Moniek Pieters1 RIVM, The Netherlands TNO - Nutrition and Food Research, The Netherlands

1 2

Published in: Human and Ecol. Risk Assessment 7: 1663-1679 (2001)

Chapter 5


5. A probabilistic human health risk assessment for environmental exposure to dibutylphthalate In the European Union, Directive 92/32/EC and EC Council Regulation (EC) 793/93 require the risk assessment of industrial chemicals. In this framework, it is agreed to characterise the level of ‘risk’ by means of the deterministic quotient of exposure and effects parameters. Decision-makers require that the uncertainty in the risk assessment be accounted for as explicitly as possible. Therefore, this paper intends to show the advantages and possibilities of a probabilistic human health risk assessment of an industrial chemical, dibutylphthalate (DBP). The risk assessment is based on non-cancer endpoints assumed to have a threshold for toxicity. This example risk assessment shows that a probabilistic risk assessment in the EU framework covering both the exposure and the effects assessment is feasible with currently available techniques. It shows the possibility of comparing the various uncertainties involved in a typical risk assessment, including the uncertainty in the exposure estimate, the uncertainty in the effect parameter, and the uncertainty in assessment factors used in the extrapolation from experimental animals to sensitive human beings. The analysis done did not confirm the reasonable worst case character of the deterministic EU-assessment of DBP. Sensitivity analysis revealed the extrapolation procedure in the human effects assessment to be the main source of uncertainty. Since the probabilistic approach allows determination of the range of possible outcomes and their likelihood, it better informs both risk assessors and risk managers.

86

A probabilistic human health risk assessment for environmental exposure to dibutylphthalate

5.1 Introduction and background In the European Union, Directive 92/32/EC (EC 1992) and EC Council Regulation (EC) 793/93 (EC, 1993c) require the risk assessment of all new industrial chemicals and prioritised existing industrial chemicals, respectively. Risks for both human health and the environment need to be addressed. Principles for how to conduct such risk assessments have been established by the European Commission (EC, 1993a; EC, 1994) and are described in detailed Technical Guidance Documents (EU-TGD, EC, 1996a) and the software package EUSES (EC, 1996b; Vermeire et al., 1997). This risk assessment process proceeds along a causal chain following the substance from its origin to the place where it is available to populations and may exert adverse effects. The exposures can be identified as acute (short-term), semi-chronic (intermediate) or chronic (long-term). Populations considered are consumers and workers, both exposed directly during or following use of chemicals, and humans exposed to chemicals following transport to and distribution in the environment. In this framework, the level of ‘risk’ is characterised by means of the quotient of exposure and effects parameters. This risk characterisation is deterministic: the quotient is a point estimate. Following the terminology of the EU-TGD and EUSES, these quotients are called Risk Characterisation Ratios (RCR), which for human health risk characterisation is the Margin of Safety (MOS), i.e., the ratio of the NOAEL (No-Observed-Adverse-Effect Level) and the estimated or measured exposure. The level of concern depends on the size of the MOS considering the uncertainties involved. The risk assessment should be ‘reasonable worst case’, defined as ‘reasonably unfavourable but not unrealistic’ (EU-TGD). Uncertainties are an inherent part of any risk assessment. From a scientific viewpoint, the uncertainty in the risk assessment should be accounted for explicitly in decision making. Examples of the application of probabilistic models to estimate human exposure (e.g., McKone, 1994; Hoover, 1999), cancer risks (e.g., Park and Hawkins, 1993) and non-cancer risks (e.g., Davis et al., 1998; Slob and Pieters, 1998; Swartout et al., 1998) are available. Comparisons between the results of probabilistic models and deterministic models have also been published (e.g., Copeland et al., 1993). However, probabilistic risk assessments covering both exposure and effects assessment as well as risk characterisation and application in decision-making are rare. The advantages and possibilities of probabilistic risk assessment in the legal framework have been extensively discussed earlier with regard to environmental (Jager et al., 2001a; Schwartz, 2000) and human health (Vermeire et al., 1999a) risk assessment. Vermeire et al.(1999a) proposed a stepwise approach in implementing probabilistic techniques in human risk assessment, first replacing fixed default assessment factors by distributions thereof, followed by the introduction of doseresponse modelling techniques which allow the derivation of the distribution of benchmark

87

5


Chapter 5


doses for each critical effect. Additional data such as substance specific information on interspecies and intraspecies differences on toxicokinetics or toxicodynamics will reduce uncertainties and thus narrow down the probabilistic distributions (IPCS, 2000). The implementation of probabilistic risk assessment in the legal framework, however, needs further explanation and more research (Vermeire et al., 1999b). It must be shown how risk management can benefit from the extra work needed in performing and understanding probabilities. The best way to demonstrate this is by performing a few probabilistic risk assessments as examples and comparing the results to the current deterministic EU-TGD approach. At this moment, it is clear that the uncertainty analysis must be kept as simple and transparent as possible to allow interpretation by non-statisticians. In this paper, the probabilistic human health risk assessment of dibutylphthalate (DBP) is presented. DBP is a chemical for which a draft risk assessment has been prepared by The Netherlands in the existing chemicals programme of the EU (EC, 1999). DBP is a highproduction volume chemical that is used mainly as plasticizer in resins and polymers such as polyvinyl chloride. In view of the widespread use of DBP in consumer products and its presence in foodstuffs and indoor air it is a potential human health concern. A large and evaluated set of experimental studies is available for DBP. As much as possible, the parameter distributions were based on the data available for the compound. In other cases, default distributions were derived from data on a representative set of chemicals. Full information regarding the uncertainty analysis is given in (Jager et al., 2000). A parallel example for the probabilistic environmental risk assessment of DBP is presented in Jager et al. (2001b).

5.2 Methods 5.2.1 Exposure Assessment 5.2.1.1 Model Description The exposure assessment presented here is based on the EU draft risk assessment report of May 3, 1999, using emission data for 1998 (EC 1999). The deterministic Predicted Environmental Concentrations (PECs) and Total Human Doses in this document (Table 5.1) have not been subject to change since.

88


Table 5.1: Predicted Environmental Concentrations (PECs) and Total Human Dose in the EU risk assessment report for DBP (EC, 1999) Local scale Production PEC-air (µg.m-3) PEC-surface water (µg.l-1) PEC- grassland soil (mg.kg-1) PEC-porewater agric. soil (µg.l-1) PEC-porewater grassl. soil (µg.l-1) PEC-groundwater (µg.l-1) Total Human Dose (µg.kgbw-1.d-1)

0.02 6 0.0007 0.006 0.006 0.006 0.8

Regional scale Formulation of adhesive 0.3 8.9 0.6 13.6 5.4 13.6 36

Processing as a softener 2.4 2.8 0.15 3.2 1.4 3.2 93

0.006 1.0 0.01 0.08 0.08 0.36

The EU-TGD method for the estimation of the PECs of DBP for the ‘Standard European Environment’ as well as the estimation of the PEC-distributions have been described by Jager et al. (2001b and a, respectively). The ‘Standard European environment’ is described by pre-defined environmental characteristics agreed upon by the EU Member States. The estimation of the Total Human Dose is based on the estimated PECs in air, surface water, groundwater, grassland soil and agricultural soil for each of the relevant industry-use category combinations and life cycle stages (production, formulation and processing) at the local scale. Figure 5.1 summarises the exposure pathways considered (EC, 1996a; EC, 1996b; Vermeire et al., 1997). In the following, data will be shown for processing of DBP as a softener in polymers. In comparison withother life cycle stages, processing leads to the highest Total Human Dose (Jager et al., 2000). This estimation of the Total Human Dose takes into account: • Bioconcentration in fish from surface water (Veith et al., 1979); • Bioconcentration in crops from contaminated air and agricultural soil pore water (Trapp and Matthies, 1995); • Bioconcentration in meat and milk products from cattle consuming contaminated drinking water, grass and soil and inhaling contaminated air (Travis and Arms, 1988); • Purification of drinking water prepared from surface water, using purification factors according to Hrubec and Toet (1992). Groundwater, or pore water under agricultural soil, is assumed to be consumed without purification; • Calculation of the Total Human Dose by multiplication of the concentrations in food, drinking water and air with food intake factors, drinking water consumption and respiratory volume, respectively. Subsequently, all intakes are summed, taking account of differences in absorption through the oral and inhalatory route of exposure. Default values for the intake factors and absorption rates have been used as agreed upon by the EU Member States (EC, 1996a).

89

5


  

Chapter 5


 

 



 

 







  

   

Figure 5.1. Aggregated environmental exposure pathways for man (EC, 1996)

5.2.1.2 Uncertainty Analysis Before  an uncertainty analysis of exposure can be performed, it must be decided which uncertainties should be analysed and in what way (the underlying assumptions). This is a crucial step, as the interpretation of the resulting probabilistic risk measure will depend on it. For the purpose of these calculations, probability distributions were assigned to ‘chemical-specific’ and not to ‘environmental’ or ‘scenario-dependent’ parameters (e.g., dilution factors, food intake factors). The chemical-related parameters may include: actual use level, physico-chemical properties, parameters related to behaviour in wastewater treatment plants, toxicity and bioconcentration values. (Jager et al., 1997; Jager et al., 2001b). When the assessment is performed for a hypothetical, generic consensus scenario such as the ‘Standard European environment’, it is impossible to characterise environmental variability in a transparent manner. Therefore, though environmental variability may be important, it is not covered in the probability distributions used in this analysis. Another source of uncertainty that is not addressed here is the fundamental uncertainty in the model concepts.



The risk assessment was performed with the EUSES 1.00 equations programmed in Microsoft Excel 97. EUSES 1.00 is the PC-implementation of the EU-TGD. The uncertainty analysis was performed using Crystal Ball version 4.0c, an add-in for Excel. Sampling was performed according to the Latin-Hypercube option, 2000 runs

 

90






were sufficient to obtain smooth distributions. Crystal Ball contains a large gallery of distributions including lognormal, triangular and uniform (the assumed distributions for the sample calculations). Table 5.2 shows the distributions applied in the uncertainty analysis for the environmental exposure assessment of human beings. Table 5.2: Parameter distributions for the calculation of the Total Human Dose for processing of DBP as a softener in polymers Parameter1 PEC-air PEC-surface water PEC-grassland soil PEC-pore water grassland PEC-porewater agricultural land PEC-groundwater BCF-fish TSCF Leaf conductance BAF-meat BAF-milk Purification factor Respirable fraction Bioavailability inhalation Bioavailability oral 1

Type of distribution2 L (k= 4.64) L (k= 5.85) L (k= 8.62 ) L (k= 11.59) L (k= 8.51) L (k= 8.51) L (k=834, M=131) T (0-1) L (k=1.4) L (k=64) L (k=36) T (M=0.15, range 0-0.65) U (0-1) U (0-0.75) U (0-1)

References: Jager et al. (2001a) for PECs and BCF-fish and Jager et al. (1997) for all

5

other parameters. 2

L = Lognormal. Defined by a geometric mean (GM) and an uncertainty factor k (95% of the parameter values is within a factor k from the geometric mean). This factor k is related to the standard deviation on log scale (slog X) by: or k = 10 ^ (1.96slogX) where b denotes the base of the logarithm used to calculate the standard deviation and sln X = ln GSD (GSD = geometric standard deviation). U = Uniform. Defined by an upper and lower limit. T = Triangular. Defined by an upper and lower limit and a mode (M). Note that this mode can differ from the median or the average in case the triangular distribution is skewed.

5.2.2 Effects Assessment DBP is not considered a genotoxic carcinogen and therefore the risk assessment is based on non-cancer endpoints assumed to have a threshold for toxicity. According to the EU draft risk assessment report (EC, 1999) the lowest NOAEL of DBP was observed in a two-generation reproduction test in rats with a continuous breeding protocol for both male and female animals (Wine et al., 1997). At the lowest dose-level of 0.1% (equivalent to 1000 mg.kgfood-1) in the diet (52 mg.kgbw-1.d-1 for males and 80 mg.kgbw-1.d-1 for females) a reduced number of live pups per litter and decreased pup weights were seen in the absence of maternal toxicity. The effects

91


Chapter 5


were reproducible among the litters, but the greatest effects were observed in litter 3. The lowest dose level of 52 mg.kgbw-1.d-1 was chosen as the NOAEL in the EU risk assessment7 As an alternative to the NOAEL-approach Crump (1984) introduced the Benchmarkapproach, which is based on dose-response modelling. Crump defined the benchmark dose as the lower 95% confidence limit on the dose corresponding to a small increase in a particular effect, typically 1 to 10% over the background level. Slob and Pieters (1998) proposed to find the complete uncertainty distribution of the Critical Effect Dose (CED). For this, the dose-response data of the reproduction study were re-analysed by fitting a dose-response model to the data, defining a Critical Effect Size (CES) and deriving the associated CED from the fitted regression model. The best fitting regression model was selected from a family of nested models by applying likelihood ratio tests. For continuous endpoints the CES is defined as a specified change in response level at some dose relative to the response level in controls. It is difficult to indicate a CES for each toxicological endpoint as no consensus exists (yet) on this issue. In this example the distribution of the CED05 is derived, which represents the dose associated with a 5% change in the endpoint analysed. The uncertainty in the estimate of the CED is assessed by a bootstrap method: Slob and Pieters (1998) use bootstrapping to generate a large number of new data sets from the original data, each time with the same number of data points per dose group as animals observed in the real experiment. For each new data set the CED dose is re-estimated. Taking all these CEDs together results in an uncertainty distribution from which any desired confidence interval can be derived. In this report the 5th and 95th percentiles are presented (i.e. 90% confidence intervals). Note that the 5th percentile can be considered as the benchmark dose as originally defined by Crump (1984). 5.2.3 Risk Characterisation The interpretation of the MOS between the exposure estimate and the NOAEL or LOAEL should be based on a consideration of all variability and uncertainty involved in the extrapolation from experimental animal data to HLVs such as the ADI and RfD. Two approaches prevail: 1. Application of substance-specific assessment factors (IPCS, 2000), and, in the absence of sufficient substance specific data: 2. Application of default assessment factors.

7 In the final report in 2003, after re-evaluation, this dose level was judged to be a Lowest-Observed-Effect Level (LOAEL).

92


The data on DBP are insufficient to derive substance-specific assessment factors and therefore default factors have to be used. In the standard procedure for deriving HLVs, various default assessment factors, each considered to be worst case estimates, are multiplied to obtain an overall assessment factor. However, multiplication of assessment factors implies compounding conservatism: the more extrapolation steps are taken into account, the higher the level of conservatism. Compounding conservatism can be avoided by the use of Monte Carlo sampling to combine probability density functions. In this method each assessment factor is considered uncertain and characterised as a random variable with a lognormal distribution. The assumption of lognormality has been justified in Vermeire et al. (1999a). Propagation of the uncertainty can be evaluated using Monte Carlo simulation yielding a distribution of the overall assessment factor. This method requires characterisation of the distribution of each assessment factor. Distributions of assessment factors have been proposed by Baird et al. (1996), Kalberlah and Schneider (1998), Price et al. (1997), Slob and Pieters (1998), Swartout et al. (1998), Vermeire et al., (1999a) and Rennen et al. (1999). As a first approach we assume that all factors are independent. Table 5.3 shows the distributions used in this study. Combination of the distributions in Table 5.3 results in a distribution of the overall assessment factor. Table 5.3. Default distributions for assessment factors. Factor Interspecies

GM 1*

GSD 4.5

Remark database derived

Intraspecies-general population

1 + 3**

1.6

theoretical, based on factor 10

Time factor: semi-chronic to chronic

2

3.5

database derived

Reference Vermeire et al. 1999a Rennen et al. 1999 Slob and Pieters 1998

5

Vermeire et al. 1999a Groeneveld et al. 1998

* This factor needs to be multiplied by an allometric scaling factor based on differences in caloric demand (mouse 7; rat 4; guinea pig 3; rabbit 2.4; monkey 1.9; dog 1.5)(Kalberlah et al. 1998; Vermeire et al. 1999a). A factor of 4 is used in the present evaluation. ** The whole distribution is increased by one (shifted to the right) since by definition the intraspecies factor cannot be smaller than unity (Slob and Pieters 1998)

For extrapolation of data from animal studies to humans, account should be taken of species-specific differences between animals and humans. These interspecies differences can be divided in differences in metabolic size and remaining species-specific differences in kinetics and dynamics. To account for differences in metabolic size, scaling on the basis of caloric demand (the 0.75 power of body weight) is considered appropriate (Vermeire et al., 1999a).

93


Chapter 5


The distribution of the CED for the experimental animal and the default distributions for the assessment factors in Table 5.3 can be used to derive the distribution of the CED for sensitive human beings (Slob and Pieters, 1998). Figure 5.2 illustrates this approach.

 Figure 5.2. The probabilistic derivation of the CED for sensitive humans, which can be considered an estimate of the true No-Adverse-Effect Level (NAEL) (Slob and Pieters, 1998). EF indicates distributions of assessment factors

 

In the final risk characterisation step, the probability that the estimated human exposure exceeds the CED for sensitive human beings is estimated by Monte Carlo sampling. In each run a Total Human Dose, drawn randomly from the probability distribution of the Total Human Dose, is divided by a CED, drawn randomly from the distribution of the CED. In this publication, these ratios are called Risk Characterisation Ratios. These RCRs, however, should not be confused with the MOS (NOAELs versus exposure) in the EU assessments. 94




5.3 Results 5.3.1 Exposure Assessment The distribution of the Total Human Dose for processing of DBP, as derived from the distributions in Table 5.2, is assumed to be lognormal. It has a GM of 0.17 mg.kgbw-1.d-1 and a GSD of 2.68. This distribution is shown in Fig. 5.3. The point estimate, based on the EU risk assessment for DBP is 0.093 mg.kgbw-1.d-1 (Table 5.1). Sensitivity analysis shows that 96% of the total uncertainty is attributable to the uncertainty in the exposure through leaf crops.  



 













 



5

   











       

Figure 5.3. Cumulative probability plot of the Total Human Dose of DBP for processing. The deterministic estimate is shown (max DOSE = 93µg.kgbw-1.d-1) as well as the HLVs derived using defaults of 10 (AF 1000 = 52 µg.kgbw-1.d-1) and using the 5th percentile of the CED-human (CED 5% = 18 µg.kgbw-1.d-1) 5.3.2 Effects Assessment The Benchmark approach (Slob and Pieters, 1998) was applied to the data of the twogeneration reproduction study. The best fitting regression model is presented in Figure 5.4. Assuming that a 5% reduction of live pups is toxicologically relevant, a CED of 281 mg.kgfood-1.d-1 was derived with a 90% confidence interval of 224-362 mg.kgfood-1.d-1 (Figure 5.5). 95


 Chapter 5

         



 























 






Figure 5.4. Live pups of litter 3 versus dose (g.kgfood-1.d-1). Triangles denote the geometric means. The horizontal dashed line indicates a CES of 5%. The vertical dashed line indicates the associated CED

 

96







    
















      



















    









 







 

5 



Figure 5.5. Uncertainty distribution around CED-animal (mg.kgfood-1.d-1) of DBP (processing) for live pups of litter 3 obtained by 500 bootstrap.





97






Chapter 5

5.3.3 Risk Characterisation The extrapolation to the sensitive human is carried out by combining the distribution of the CED with distributions of assessment factors presented in Table 5.3 in a probabilistic manner (Slob and Pieters 1998, Vermeire et al. 1999a). This cumulative distribution is characterised by a GM of 8.2 mg.kgfood-1.d-1 and a 5% lower confidence limit of 0.37 mg.kgfood-1.d-1 (equivalent to 0.018 mg.kgbw-1.d-1). Any value in this distribution could, by choice, be considered a Human Limit Value and be compared with any point on the distribution of the exposure estimate (Figure 5.3). The resulting distribution of the RCR of the Total Human Dose and the Human Limit Value (CED-human) is subsequently derived by Monte Carlo sampling from the distribution of the Total Human Dose and the CED-human. The result is shown in Figure 5.6. Sensitivity analysis shows that 80% of the uncertainty is attributable to the assessment factors and 20 % to input parameter uncertainty. The deterministic estimate of this RCR would be based on the deterministic exposure estimate of 0.093 mg.kgbw-1.d-1 and the deterministic HLV of 0.052 mg.kgbw-1.d-1, the latter being the NOAEL divided by the overall assessment factor of 1000. The deterministic RCR would then be 1.8, based on the NOAEL.    


 













    











 

 

Figure 5.6: Cumulative probability plot of the RCR for processing of DBP, i.e., the Total Human Dose divided by the CED, for sensitive human beings. The deterministic estimate based on the NOAEL and an assessment factor of 1000 is also shown [RCR (AF1000)].

   

98


5.4 Discussion This example risk assessment shows that a probabilistic risk assessment in the EU framework covering both the exposure and the effects assessment is feasible with currently available techniques. A similar analysis has been performed with a less data-rich new chemical, applying more default distributions (Jager et al., 2000). The example shows the possibility of comparing the various uncertainties involved in typical risk assessment, including the uncertainty in the exposure estimate, the uncertainty in the toxicological starting point (benchmark approach), and the uncertainty in assessment factors. The benchmark-dose concept takes into account the information on the dose-response and the uncertainties in the estimation of the ‘true’ experimental threshold in the animal, depending on the quality of the particular study from which the data are used. The use of assessment factors for LOAEL-NOAEL extrapolation and for the ‘steepness’ of a dose-effect curve is completely redundant applying the benchmark approach. Although the probabilistic approach is less straightforward than the NOAEL-approach and requires some statistical experience in fitting mathematical dose-response models to data, it allows determination of the range of possible outcomes and their likelihood. It therefore better informs both risk assessors and risk managers than the deterministic approach. The degree of conservatism in the deterministic risk assessment can be evaluated against the probabilistic distribution. The following should, however, be noted (Vermeire et al., 1999a): • The default distributions of the assessment factors are partly based on historical NOAEL-data with different systemic endpoints and with often unknown quality and errors and, and partly on theoretical assumptions. One obvious discussion point, considering the critical endpoint for DBP, is a possible change in the interspecies default distribution if based on NOAELs for reprotoxic endpoints only. Another point of debate may be the intraspecies default distribution which in this example pertains to the unborn and therefore may be different from the general distribution used here; • Consensus needs to be reached yet on the CES for different toxicological endpoints that may be relevant. The final risk distribution can be presented in a variety of ways. In this publication a choice was made for a cumulative form (Figure 5.6). Another option is to base decisions on a combination of an upper percentile (e.g., 95th percentile) and a central tendency estimate (mean or median).

99

5


Chapter 5


The present example only takes into account uncertainty in chemical-specific input parameters and is valid only for the ‘Standard European Environment’ as defined in the EU-TGD. Environmental variability is not considered, but may be significant. One could take environmental variability into account by repeating the approach used for different exposure scenarios (e.g., a best case, an average case and a worst case scenario). This approach necessitates a discussion on relevant exposure scenarios and on a clear-cut separation in chemical-specific and scenario parameters. This approach is considered more transparent to both risk assessors and risk managers than a formal two-dimensional uncertainty analysis in which variability and uncertainty are propagated separately as shown by Hoffman and Hammonds (1994). The RCR distribution spans 3-4 orders of magnitude (90% confidence interval, Figure 5.6). According to the sensitivity analysis – a powerful tool for efficient decisions on further research - the uncertainty due to the extrapolation from experimental data to the sensitive human being explains about 80% of this variability and variation in input parameters for the exposure estimation the remaining 20%. Since the probability that the CED is exceeded in sensitive human beings is considerable (approximately 35%, Figure 5.6), this huge variation in the RCR may be problematic. In such cases the generation of data which would allow the derivation of chemical-specific assessment factors (IPCS, 2000) or relevant human effects data could reduce the uncertainty in the outcome of the risk assessment considerably. Further investigations into the variation in the distributions for the assessment factors and possibilities for refinement are explored. Comparing the results of the probabilistic approach with those of the deterministic approach as applied in the EU risk assessment report the following can be observed: • The deterministic estimate of the Total Human Dose, expected to be realistic worst case, is quite low (Figure 5.3). According to the uncertainty analysis performed the probability is only approximately 20% that the Total Human Dose is lower than the deterministic estimate of 93 µg.kgbw-1.d-1. Air is the main source of exposure for plants, which is the main source for human exposure to DBP. The deterministic PEC for air is at the 75th percentile of the distribution (Jager et al., 2001b). This discrepancy is caused by the choice of the distributions for water solubility and vapour pressure in the uncertainty analysis. The medians of these parameters differ from those in the dossier in such a way that a much lower air-water partition coefficient results (more than a factor of 10 with the median parameters values) leading to higher values in plants than the deterministic value in the EU risk assessment report. It should be noted that the median value of our plant-water partition coefficient is at the top of the experimental range reported in a validation study (Polder et al., 1998) and may be unrealistic as it is

100


•

•

•

extremely dependent on small changes in physico-chemical properties. Nevertheless, this stresses the importance of uncertainty analysis since the choice of vapour pressure and water solubility is so important! Compared to the total distribution derived for the RCR (Figure 5.6), the deterministic RCR of the Total Human Dose and the NOAEL/1000 is not conservative, i.e., between the 60th and 70th percentiles of the distribution. Without the uncertainty analysis, it would be unknown whether the RCR would really represent a reasonable worst case, the target specified in the EU-TGD. It has also been demonstrated that the main source of uncertainty is the compounding of worst case assessment factors of 10 in the effects assessment and not the estimation of the Total Human Dose. As pointed out earlier, more information on the main sources of the variability in the RCR can be obtained from the sensitivity analysis, allowing better-informed decisions on further research. In the EU risk assessment report, the MOS for processing of DBP is 562, which is the ratio of the NOAEL of 52 mg.kgbw-1.d-1 and the Total Human Dose of 93 µg.kgbw-1.d-1. The final conclusion “no concern for man indirectly exposed via the environment” may be challenged considering the ‘Minimal MOS’ of 1000 based on the current use of default assessment factors of 10 and the fact that at the NOAEL used still a slight effect was observed. The probabilistic analysis shows that the conclusion drawn may indeed be challenged. The application of a factor of 10 for the semi-chronic to chronic extrapolation may be considered unnecessary for the reprotoxic effect observed, but is justifiable since humans should be protected against other adverse effects as well. The semi-chronic, oral NOAEL for general toxicity was 152 mg.kgbw-1.d-1. The degree of conservatism in the deterministic risk assessment can be evaluated against the probabilistic distribution of the RCR. Presently, a deterministic RCR below 1 would indicate ‘no concern’ to the risk manager. A risk manager would probably think differently if he knows that the probability of exceeding this value still is 35%. Note, however, that the RCR, even a probabilistic one, is not a true risk level, since the relationship between the RCR and the toxicological impact is unknown. In this particular example we have estimated the probability that in sensitive human beings the Total Human Dose will lead to adverse reproductive effects.

It is noted that this example of a probabilistic risk assessment is based on a draft EU risk assessment for DBP. The final EU risk assessment of DBP may differ from this draft. In conclusion: since the probabilistic approach allows determination of the range of possible outcomes and their likelihood, it better informs both risk assessors and risk managers. The degree of conservatism in the deterministic risk assessment can be evaluated against this distribution. Hence, despite some of the drawbacks mentioned above, it is

101

5


Chapter 5


believed that the method presented provides a more transparent risk assessment than the deterministic approach. Decisions can be taken with explicit knowledge of the uncertainties involved and without a false sense of accuracy. The probabilistic approach makes better use of existing information. This information can be drawn from the database of the chemical considered and from data of other chemicals as shown in this example in the application of default distributions of assessment factors.

Acknowledgements This publication is based on an investigation performed by order of the DirectorateGeneral for Environmental Protection, Directorate of Chemicals, Radiation and Safety of the Ministry of Housing, Spatial planning and the Environment, within the framework of project M/679102 ‘Risk assessment methodology’. The authors gratefully acknowledge the contribution of H.A. den Hollander, P.van der Poel, and M.G.J. Rikken to the environmental and human exposure estimation in this RIVM-project.

102

6 Assessment factors for human health risk assessment: a discussion paper

Theo Vermeire1, Hanzen Stevenson2, Moniek Pieters1, Monique Rennen2, Wout Slob1, Betty Hakkert2 RIVM, The Netherlands TNO - Nutrition and Food Research, The Netherlands

1 2

Published in: Critical Reviews in Toxicology 29: 439-490 (1999)

Chapter 6


6. Assessment factors for human health risk assessment: a discussion paper The general goal of this discussion paper is to contribute towards further harmonisation of the human health risk assessment. First, it discusses the development of a formal, harmonised set of assessment factors. The status quo with regard to assessment factors is reviewed: i.e. the type of factors to be identified, the range of values assigned as well as the presence or absence of a scientific basis for these values. Options are presented for a set of default values and probabilistic distributions for assessment factors based on the state of the art. Methods of combining default values or probabilistic distributions of assessment factors are also described. Secondly, the effect parameter, the No-Observed-Adverse-Effect Level (NOAEL), is discussed. This NOAEL as selected from the toxicological database may be a poor substitute for the unknown, true No-Adverse-Effect level (NAEL). New developments are presented with regard to the estimation of the NAEL. The already widely discussed Benchmark Dose concept can be extended to obtain an uncertainty distribution of the Critical Effect Dose (CED). This CED-distribution can be combined with estimated uncertainty distributions for assessment factors. In this way the full distribution of the Human Limit Value will be derived and not only a point estimate, whereas information on dose-response relations is taken account of. Finally, a strategy is proposed for implementation of the new developments into human health risk assessments.

104







 



    

     

Assessment factors for human health risk assessment: a discussion paper

ummary of expressions S 

Geometric mean of the lognormal distribution

 1 n GM = exp( ∑ ln Xi ))        n   i=1 Sample variance of log-entities  1 n 2  (ln Xi − ln GM ) sln2 X =  ∑  i =1          n − 1       Geometric standard deviation   GSD = exp( sln X )       95th percentile P 0.95            n: number of observations Xi  : lognormally distributed ith observation (e.g. NOAEL) s  :  sample standard deviation of lognormally distributed X lnX   z : 95th percentile of the standard normal distribution 0.95  6.1 Introduction 6.1.1 TGD approach towards human risk characterisation The general goal of this report is to contribute towards the further harmonisation of the human health risk assessment. Although much of the contents of this report is applicable to the human health risk assessment of chemical substances in general, it concentrates on the assessment of new and existing substances within the scope of the European Union legislation. In the European Union, prior to REACH, Directive 92/32/EC (EC, 1992) and EC Council Regulation (EC) 793/93 (EC, 1993c) required the risk assessment of new and existing substances, respectively. Principles for this risk assessment have been laid down (EC, 1993a: EC, 1994), supported by a detailed package of Technical Guidance Documents (TGD; EC, 1996a) and the software implementation EUSES (EC, 1996b; Vermeire et al., 1997) 8. This risk assessment process proceeds along a causal chain following the substance 8

Changes in the TGD version discussed here relative to the REACH TGD are discussed in Section 7.2.

105

6


Chapter 6


from its origin to the place where it is available to organisms and may exert adverse effects. The exposures can be identified as acute, semi-chronic or chronic. Populations considered are consumers, workers and man exposed through the environment. The following summarises important aspects of the human effects assessment and risk characterisation from the TGD (EC, 1996a). In the human risk assessment an attempt is made to identify the hazards of the substances and to relate them to exposure. For those substances for which a threshold for toxicity is assumed to exist, a No-Observed-AdverseEffect Level (NOAEL), has to be derived or, if this is not possible, a Lowest-ObservedAdverse-Effect Level (LOAEL). The NOAEL is the highest concentration or amount of a substance, found by experiment or observation, which causes no detectable adverse alteration of morphology, functional capacity, growth, development, or life span of the target organisms under defined conditions of exposure WHO, 1979). Unless a N(L)OAEL value is available from human data, the N(L)OAEL values are those derived from animal studies. In the risk assessment for man, the risk is characterised by comparing estimated (or measured) concentrations in air or on skin or total daily intakes to the results of the effects assessment. This analysis is made separately for each population potentially exposed and for each effect. Where the exposure estimate is higher than or equal to the N(L)OAEL, the TGD (EC, 1996a) qualifies the substance as “of concern”9. If the exposure estimate is less than the N(L)OAEL, the risk assessor is asked to decide whether the magnitude by which the N(L)OAEL exceeds the estimated exposure, the “Margin of Safety”, is of concern. The TGD recommends the following parameters to be considered in assessing the Margin of Safety: • The uncertainty arising, among other factors, from the variability in the experimental data and intra- and interspecies variation; • The nature and severity of the effect; • The human population to which the quantitative and/or qualitative information on exposure applies; • The differences in exposure (route, duration, frequency and pattern); • The dose-response relationship observed; • The overall confidence in the database. Further, the TGD states that expert judgement is required to weigh these individual parameters on a case-by-case basis. Transparency is required, meaning that all decisions taken and their rationale should be described carefully. The TGD refers to several relevant publications which may support this expert judgement. 9 If it is concluded that a substance is “of concern”, it is considered likely that adverse effects can be expected in human populations from known or reasonably forseeable use. In that case further data have to be requested or risk reduction recommendations made.

106


6.1.2 Methods using assessment factors Another widely used approach is the explicit derivation of a Human Limit Value (HLV)10 for man from experimental or epidemiological toxicity data by dividing the NOAEL or LOAEL by an assessment factor. This overall assessment factor is a multiple of several factors (called safety factors, uncertainty factors, extrapolation factors, adjustment factors, conversion factors; see definitions below), accounting for inter- and intraspecies variations, differences in exposure time scales, the nature of the adverse effects, the adequacy of the database etc. This approach is described and discussed extensively in the scientific literature. Pertinent reviews are produced by Dourson (1996), the Dutch Health Council (1985), McColl (1989), IPCS (1994), ECETOC (1995) and Stevenson et al. (1995a).

The consistent use of terminology is a prerequisite for harmonisation discussions. Reviewing the available literature on this subject, the differences in terminology with regard to these factors is striking. Therefore, before turning to a comparison of the two approaches in human effects assessment and risk characterisation as introduced above, it is essential to present the definitions as they will be used throughout this report, acknowledging that the use of another set of definitions may be equally justified (Box 6.1): Box 6.1 Definitions Assessment factor

general term to cover all factors designated as safety factor,

uncertainty factor, extrapolation factor, adjustment factor,

conversion factor, etc. and the composite thereof.

Extrapolation factor

database-derived factor, used in the extrapolation from

experimental or epidemiological toxicity data to a health

based recommended exposure level for man, which takes

into account uncertainty due to inter- and intraspecies

variability,variability in exposure duration, variability in

nature and severity of effects, including the dose-response,

and variability in the adequacy of the database.

6

Other terminology will not be used in this report or explicitly explained, unless the term is a quote (Section 6.2).

10

HLV: general term covering various limit values such as the ADI, RfD, PNAEL and HBROV (see Section 6.2.2.)

107


Chapter 6


Within the scope of the EU legislation on risk assessment for new and existing substances two approaches for the application of assessment factors in human health risk assessment have been proposed: one by a Task Force of the European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC, 1995) and one by TNO Nutrition and Food Research Institute (Stevenson et al., 1995a; Hakkert et al., 1996). TNO Nutrition and Food Research Institute also compared both approaches (Stevenson et al., 1995b). A harmonised WHOscheme for the derivation of guidance values for health-based exposure limits (ADI) has been adopted by task groups of the International Programme on Chemical Safety (IPCS, 1994). 6.1.3 A comparison In this section the Margin of Safety approach (MOS-approach) of the TGD (EC, 1996a) is compared to the assessment factor approach (AF-approach). Note that this TGD does not provide any quantitative guidance on the size of the Margin of Safety. This subject clearly was not yet ready for EU-wide harmonisation at the time of the development of the TGD of 1996 given the time constraints and the gaps in opinions to bridge. The AF-approach can be used either to derive a limit value or to indicate the minimum size of the Margin of Safety. Aspects to be covered in the comparison of the MOS-approach and the AF-approach are: variability, transparency, consistency and acceptability. Variability The MOS-approach is inherently more variable than the AF-approach. The risk characterisation on the basis of a MOS is heavily dependent on expert judgement, whereas assessment factors need to be applied on the basis of more or less fixed criteria. The more elaborate these criteria are, the less degrees of freedom are left for expert judgement. Expert judgement, however, always plays a role in the application of assessment factors, but in a more formalised way. The variability in the outcome of expert judgement was illustrated by Dourson and Lu (1995) who compared two sets of 65 risk assessments developed by WHO and US-EPA. ADI and RfD values were within a 3-fold range for 38 sets, a 3 to 30fold range for 20 sets, a 30 to 300-fold range for 6 sets and in one case differed 700-fold. It is noted that at least part of these differences will be due to the different year of evaluation and thus, different data evaluated. Transparency Transparency in this context indicates how clear the choices made in the human risk characterisation stage are to all stakeholders. The high variability of the MOSapproach should force experts to explain elaborately their way of thinking in each risk

108


characterisation. Very often risk assessors fail to do so and therefore transparency is not easily achieved. Although the application of assessment factors also needs to be made transparent, the burden on the expert is less since certain criteria are already explicit. Consistency Consistency in this context denotes the logical agreement of risk characterisation decisions made for different substances in similar situations. Unless the application of the MOSapproach is accompanied by careful documentation of decisions and criteria applied by experts, it will be very difficult to maintain consistency. “Institutional memory” is a great asset here, but no guarantee to full consistency over longer periods of time. Clearly, the demand for careful documentation also applies to the AF-approach, but in this case many decisions to be made and criteria are already explicit and only need to be referred to. Problems to maintain consistency are increased when more expert groups are involved sharing the burden of the work such as within the scope of the EU risk assessment for new and existing substances. Acceptability • Acceptability to the risk assessor One could argue that a highly variable risk characterisation system such as the MOSapproach may be more acceptable than an inherently more rigid AF-approach because it allows each party involved to keep its own standards high and therefore keeps everyone satisfied. However, from the above it is also clear that in a co-operative risk assessment scheme such as that of the EU for new and existing substances, it will be virtually impossible to maintain consistency across all substances and to be transparent at the same time. •

Acceptability to risk managers In this situation risk managers are confronted with outcomes of risk assessments that are difficult to classify and to base risk reduction strategies upon. Would the risk manager have the luxury of a second opinion, chances are high she or he would be confronted with differing risk statements for the same exposure scenario. The lack of guidance in the TGD on the acceptable size of the Margin of Safety leads to lengthy discussions for each priority substance with respect to the acceptability of risks to man. As a consequence, the risk manager is left in doubt.

It is concluded that decision criteria for human risk characterisation need to be made explicit. This could be achieved by establishing a formal, harmonised set of default assessment factors accompanied by elaborate guidance. This was also one of the

109

6


Chapter 6


conclusions of a Workshop held under the auspices of the European Chemicals Bureau for an exchange of experience within the scope of risk assessment for new substances under Directive 92/32/EEC (Vollmer et al., 1996). Several prerequisites can be formulated for a harmonised default set of assessment factors: • The default set should only be applied in the absence of data which permit a more substance-specific, scientific choice. The routine use of data driven assessment factors will increase the confidence in risk assessments and encourage mechanistic research (Dourson, 1996; Stevenson et al. 1995a); • In view of the possible differences in exposure scenarios, including groups at risk, and in the toxicological database the default set should allow for differentiation with regard to these differences; • Harmonisation should not hamper further developments, that is, should not be seen as standardisation for ever. It is too easy to argue that the problem of quantifying human health risks will be solved completely by such a harmonised set of assessment factors. Unless the criteria are very rigid - which is not desirable and probably not possible in view of the uncertainties involved - there will always remain opportunities for scientific debate on the choices to make. Furthermore, it is still extremely difficult to indicate to the risk manager what the risk of a threshold substance will be in terms of the number of people at risk, their geographical distribution and the nature of the effect(s), once the minimum Margin of Safety is not reached or a no-effect level exceeded. The risk manager needs the information on exposure and the adjusted dose-response curve in order to set priorities. The latter discussion, however, is outside the scope of this report. 6.1.4 Goals This report intends to be a contribution towards the development of a formal, harmonised set of assessment factors to be applied within the scope of the EU risk assessment for new and existing substances. This means that both acute, subchronic (covering both subacute and semi-chronic) and chronic exposure and occupational as well as non-occupational exposure need to be addressed. The report will first discuss the status quo with regard to the type of factors to be identified, the range of values assigned as well as the presence or absence of a scientific basis for these values (Section 6.2). Section 6.3 is a discussion on the options possible with regard to a set of default assessment factors based on the state

110


of the art. For each factor conclusions will be presented on the scientific basis, the most likely distribution and the assumptions made. Methods for combining assessment factors are presented as well. In Section 6.4 the Benchmark dose concept is discussed as well as methods of combining probability density functions for the Benchmark dose and the assessment factors to arrive at the probability density function of the HLV. The various methods presented are illustrated using an example substance in Section 6.5. Section 6.6 will summarises the major findings of this report and addresses recommendations. The report extensively refers to relevant earlier studies done in this area both inside and outside the EU framework. It is intended as a contribution towards further discussion within the EU framework in the first place but also explicitly taking into account the status quo and developments elsewhere. One harmonisation effort, currently underway, at the international level is the IPCS project “Harmonisation of approaches to the assessment of risk from exposure to chemicals” (Sonich-Mullin, 1995). This project considers qualitative and quantitative risk assessment methods as well as methods used for determining endpointspecific effects. The project, among others, considers the issue how uncertainty and variability are taken into account in risk assessment. The harmonised use of terminology is a subproject undertaken in collaboration with the OECD.

6.2 Review of applied assessment factors in human health risk assessment 6.2.1 Introduction This section presents an overview of published extrapolation methods based on the assessment factor approach for establishing Human Limit Values (HLVs). This overview is based on a report of Stevenson et al. (1995a). It is limited to a safety factor approach, thus to effects with threshold characteristics and is not meant to be exhaustive. Several terms are used for the factors introduced for the translation of NOAELs from experiments as described in the previous section. In this section, the terminology of the regarding authors is used when reference is made to the methods described. The term “assessment factor” is used when no reference is made to a specific term or method. The assessment factor can cover both extrapolation and uncertainty. This section is not intended to give a complete overview of all procedures described. Some alternative extrapolation methods (e.g. the Benchmark approach) are described in Section 6.3.

111

6


Chapter 6


6.2.2 Extrapolation methodology 6.2.2.1 Acceptable Daily Intake (ADI) Historically, the so-called safety factor approach (SF-approach) was introduced in the United States in the mid-1950s in response to the legislative guideline needs in the area of food additives (Food and Drug Administration (FDA)). This approach proposed that a safe level of food additives or contaminants can be derived from a chronic NOAEL (in mg/kg of diet) from animal studies divided by a 100-fold safety factor (ECETOC, 1995; Lehman and Fitzhugh, 1954). In a slightly modified form, this proposal was adopted by the Joint FAO/WHO Expert Committee on Food Additives (JECFA) and by the Joint Meeting of Experts on Pesticides Residues (JMPR) of the WHO/FAO in 1961: the safe level was called the Acceptable Daily Intake (ADI) and was expressed in mg/kg body weight per day. The procedure involved collecting all relevant data, ascertaining the completeness of the available dataset, determining the NOAEL using the most sensitive indicator of toxicity, and applying an appropriate safety factor to derive the ADI for humans. The ADI approach is now widely used as well as the comparable Tolerable Daily Intake (TDI) approach for contaminants. The ADI is defined as ‘the daily intake of a chemical which, during the entire lifetime, appears to be without appreciable risk on the basis of all known facts at the time’. The rationale for the 100-fold safety factor is reviewed by Dourson and Stara (1983). Initially, Lehman and Fitzhugh (1954) reasoned that the safety factor 100 accounts for several areas of uncertainty: • Intra (human) species variability; • Inter (animal to human) species variability; • Allowance for sensitive human populations due to illness as compared to healthy experimental animals; • Possible synergistic action of the many intentional and unintentional food additives or contaminants. Bigwood (1973) and Lu (1979) justified the 100-fold factor on the basis of: • Differences in body size of the laboratory animals versus that of human; • Differences in food requirements varying with age, sex, muscular expenditure, and environmental conditions within species; • Differences in water balance of exchange between the body and its environment among species; • Differences in susceptibility to the toxic effect of a given contaminant among species.

112


Vettorazzi (1980) justified the use of the 100-fold factor by: • Differences in susceptibility between animals and humans; • Variations in sensitivities in the human population; • The fact that the number of animals tested is small compared to the size of the human population that may be exposed; • The difficulty in estimating the human intake; • The possibility of synergistic action among chemicals within the human diet. It is apparent that the factor100 has no quantitative basis, and the choice of the value 100 is more or less arbitrary. It is, in several ways, justified retrospectively. Others have attempted to interpret this factor as the product of two uncertainty factors with default values of 10, one for intra- and one for interspecies variability. But, by either interpretation, the purpose of the safety factor is to allow for uncertainties in knowledge of the toxic response of a small number of rather homogeneous laboratory animals in establishing safe doses for a heterogeneous human population (Stevenson et al., 1995a). The safety factor should not be considered immutable. When setting the ADI, various test data and judgmental factors should be considered and are needed to be taken into account, e.g., adequacy of data base, nature of the effects, age-related effects, metabolic and pharmacokinetic data, and available human data. The overall safety factor ranges from 10 to greater than 1000, and the most commonly used factor is 100. The FDA recommends an additional factor 10 when estimating an ADI from short-term toxicity data. Comments Usually a safety factor of 100 is used for establishing ADIs. Retrospectively, some attempts have been made to support this factor. The procedures of the JECFA and JMPR do not generate a clear motivation for deviation from the factor 100. However, in some individual cases an expert explanation is given for the use of different factors. 6.2.2.2 Health Council of the Netherlands The Health Council of the Netherlands (1985) presented an approach for the establishment of HLVs for the general population based on the same safety factor 100. Two steps are considered: a real extrapolation step to compensate for differences in body size between the test species and human, and the application of safety factors to compensate for observation errors, possible species-specific differences in biological availability, and susceptibility between test species and human.

113

6


Chapter 6


Two ways of extrapolation were compared: 1. On basis of body weight: This method assumes that body weight is a good relative measure of factors determining the concentration in blood. 2. On basis of caloric demands equivalent to (body weight)0.75 : Because caloric demands (basal metabolism) varies with (body weight)0.75 interspecies adjustment factors can be calculated. If extrapolation on basis of caloric demands is chosen, the Health Council proposed a safety factor of 30 in combination with adjustment factors. They justified this safety factor as follows: • The safety factor for interspecies variation should be smaller, because the variation introduced by differences in body weight has been accounted for. • The safety factor can also be rationalised otherwise. The variations and errors (interspecies, intraspecies, observation errors) are independent lognormal variations. The intra- and interspecies variation are estimated to be 10: the factor for observation errors is estimated to be 3. The calculation of the safety factor is as follows:

log(totalvariation) = (log 10) 2 + (log 10) 2 + (log 3) 2 ≈ 1.4925 ≈ log 31.1 (6.2.1) The Health Council emphasized the role of experts in judging the quality of the data base, in formulating the toxicological starting-points, and in establishing the safety factors. The absence of relevant data should be taken into account. At present, the safety factor approach is still in use. However, the Health Council Committee believes that limits should be derived using a method which makes systematic use of data on the relationship between exposure and response (Health Council, 1996b). Comments This method is not commonly used in The Netherlands. An advantage of this method is the theoretical distinction between extrapolation and uncertainty. However, in practice, such a distinction is not always possible. Scaling on caloric demands is considered to be preferable above scaling on body weight, both on theoretical grounds and because of the similarity with setting respiratory HLVs from inhalation studies. This method is still based on the default factor of 100 used in the derivation of the ADI. The rationale for the formula is not given but lognormal uncertainty measures can indeed be summed logarithmically (Slob, 1994) (see also Section 6.3).

114


6.2.2.3 EPA: Reference Dose (RfD) In 1988, the US-EPA also adopted the ADI approach in its regulatory measures against environmental pollution; with a number of modifications, however. Instead of the terms ADI and safety factor (SF), the terms “reference dose” (RfD) and uncertainty factor (UF) are used, respectively. The RfD is derived from the NOAEL by consistent application of generally one order-of-magnitude UFs that reflect the various types of data set used to estimate RfDs. UFs generally consist of: • A 10-fold factor to account for human variation in sensitivity; • A 10-fold factor to account for uncertainties in interspecies extrapolation; • A 10-fold factor to adjust for the use of the NOAEL obtained from a subchronic animal study rather than a chronic study; • A 10-fold factor to adjust for the use of the LOAEL in the absence of the NOAEL; • A 10-fold factor that considers the adequacy of the total database In addition, a modifying factor (MF) ranging from less than 1 to up to 10 is applied when the data base includes for example a very large number of animals per dose level (MF less than 1). 6.2.2.4 Method of Calabrese and Gilbert The modifications in uncertainty as used by the EPA, i.e. a factor 10 for interspecies and 10 for intraspecies variations were also proposed by Calabrese and Gilbert (1993). They suggested modifications of UFs by the lack of total independence of these factors. The interspecies UF is generally recognized as providing an extrapolation from the average animal to an average individual assuming that humans may be 10-fold more sensitive. The intraspecies UF assumes that most human responses fall within approximately a 10-fold range. Calabrese and Gilbert stated that, given this assumption, the application of a 10fold intraspecies UF should begin with the average human and extend to cover the higher risk segments of the population. Consequently, an UF of 5 would be expected to protect most humans. However, an UF 10 is indicated when the HLV is based on an occupational epidemiological study since this type of study does not consider the most sensitive humans. The factor used when a semi-chronic study is used as starting point, incorporates an agedependent factor which is comparable in some respects to the age-dependent factor in the intraspecies uncertainty factor. The age component in the intraspecies uncertainty factor concerns the age differential response over the entire lifetime span whereas the age differential of the less-than-lifetime uncertainty factor concerns only the age-related differences from the end of the study to the end of the normal life span. High susceptibility is not exclusive for young animals. In certain circumstances susceptibility may be greater

115

6


Chapter 6


in adulthood than in the young and may further increase in elderly animals. Assuming that age differences account for 50% of the intraspecies variation, they stated that this factor could be reasonably apportioned as 60% for prior to weaning and 40% for after weaning. If a 24-month rodent exposure accounts for 40% of age effects, then it would be reasonable to reduce the age component of the intraspecies variation by the proportion described to age. If the intraspecies factor is 5 as recommended, then this reduces the factor to 4. Table 6.1 provides the uncertainty factors recommended by Calabrese and Gilbert in the light of the above considerations. Table 6.1: Recommended modifications in uncertainty factors using the concept of independence and interdependence of uncertainty factors (Calabrese and Gilbert, 1993) Extrapolation step Animal to human Interindividual less-than-lifetime animal study animal study with normal experimental lifetime occupational epidemiological study environmental epidemiological study (normal lifespan) LOAEL instead of NOAEL Less-than-lifetime

Uncertainty factor 10 5 4 10 5 10 10

Comments: The notion of interdependence between the factors is considered a valuable one. However, it is recommended to examine the interdependence of the factors in more detail before applying the concept in risk assessment procedures. The assumption of total independence of assessment factors should be recognised as a “worst-case” approach. 6.2.2.5 Method of Renwick The approach proposed by Renwick (1991, 1993a, 1993b) is also based on the 100-fold factor used to derive ADIs. It attempts to give a scientific basis to the default values of 10 for the interspecies and inter-individual differences. Renwick has proposed the division of each of these UFs into subfactors to allow for separate evaluations of differences in toxicokinetics and toxicodynamics. The advantage to such a subdivision is that components of these UFs can be addressed where data are available (e.g., if data exist to show similar toxicokinetic handling of a given chemical between laboratory animals and humans, then only an interspecies extrapolation factor would be needed to account for differences in toxicodynamics). Renwick examined the relative magnitude of toxicokinetic and toxicodynamic variations between and within species in detail. He found that toxicokinetic differences were generally

116


greater than toxicodynamic differences resulting in the proposal that the 10-fold uncertainty factors (for inter- and intraspecies) should, by default, be subdivided into factors of 4 for kinetics and 2.5 for dynamics. Factors up to 10 should be applied to critical effects as teratogenicity and nongenotoxic carcinogenicity. In cases where a reversible lesion (e.g. hyperplasia) is believed to be a precursor for a severe irreversible change, the NOAEL for each lesion should be used to calculate an ADI, using appropriate factors. The lowest value should be chosen as ADI. The rationale for an extra factor is the potential seriousness of any unrecognised aspect that has not been taken into account in the safety evaluation. This factor is not data derived but based on scientific judgement. A factor for the adequacy of the overall database has been introduced to consider aspects other than the pivotal study and the determination of the NOAEL. A value of 1 assumes an adequate database consistent with national or international guidelines. Comments Renwick also uses the default factor of 100 as basis. The main feature of this approach is that kinetic and dynamic aspects are distinguished in inter- and intraspecies differences. This offers the possibility to incorporate mechanistic information on these aspects in the establishment of the factors as shown for pharmaceuticals by Naumann et al. (1997), provided sufficient data are available. It should be remarked that the proposed default values are derived from only limited data. Renwick also uses expert judgement for the derivation of HLVs, but with an attempt for transparency and clear motivations. The International Programme on Chemical Safety11 has adopted the principles set forth by Renwick, but has suggested that while the UF for interspecies extrapolation be subdivided unequally into 4-fold and 2.5-fold, the UF for intraspecies extrapolation should be split evenly (3.16-fold for both kinetics and dynamics) (IPCS, 1994). 6.2.2.6 Approach described by Lewis/Lynch/Nikiforov In 1990 Lewis and his colleagues (1990) undertook revision of the long established practices, with the goal of introducing flexibility such that both new information and expert judgement could be readily incorporated. The Lewis-Lynch-Nikiforov (LLN)-method, and its refinements, are extensions of established principles and procedures. LLN guides the data evaluator to adjust experimentally determined ‘no-effect’ (or ‘minimum effect’) levels from laboratory animal studies, while taking the following aspects into account: • Differences between laboratory animals and humans; • Differences between experimental conditions and actual or anticipated human exposures; • The sensitivity of the exposed human populations; • Weight of evidence indicating an actual human health hazard;

117

6


Chapter 6


• • •

Quality of the experimental information base; Uncertainties in extrapolating from animals to humans; Potency of the toxic agent.

If suitable human data are not available, the HLV is estimated from laboratory results, using the following algorithm:

HLV =

NOAELanimal [ S] [ I ][ R][Q1 ][Q2 ][Q3 ][U ][c]

(6.2.2)

The terms are described in Table 6.2. An aggregate adjustment of about 250 is typical and is approaching the practical maximum. The theoretical maximum adjustment value is 100,000. By application of the factors Q1-Q3 and U this method intends to separate scientific judgements from policy/value judgements. Factors for data quality should reflect the completeness and suitability of the available information. According to Lewis et al. (1990) there are three distinguishing features of the LLN-approach: • Careful discrimination among the adjustments; • Discrimination between ‘best estimates’ of the correct adjustments for [S], [I], and [R] and the overall uncertainty; • Securing scientific consensus on the adjustment values. Table 6.2: Adjustment factors of the Lewis/Lynch/Nikiforov model(Lewis et al., 1990) AFa

Description

Range of values

Most likely value

Default value

S

Scaling factor to account for known quantitative differences between species and between experimental conditions and those likely to be encountered by humans Intraspecies variability Interspecies extrapolation Degree of certainty that the critical effect observed in laboratory animals is relevant to humans Subchronic to chronic extrapolation LOAEL to NOAEL extrapolation Accounts for residual uncertainty in estimates of S, I, and R A non-scientific, judgmental “safety” factor

>0

NSb

1

1-10 > 0-10 0.1-1

1-3c NS

10 10 1

1-10 NS 1-10 1-10

1-3 2 NS ≤3

10 10 10 1

I R Q1 Q2 Q3 U c a

AF, Adjustment Factor

b

NS, Not Stated by authors

c

Most likely value based on study of high quality

118


Comments The LLN-method discriminates to a large extent and therefore, this method is valuable to give guidance for which factors one should account. However, some remarks can be made: • In practice it will not be possible to distinguish all these factors; • It is likely that the scaling factor S influences the adjustment for interspecies differences R. However, the authors give no guidance for the choice of R when a value different from 1 is introduced for S; • The factors I and R are not consistently distinguished. R is said to account for a possible wider range of susceptibility among individuals. However, this belongs to the factor I; • The value Q1 seems superfluous, because for the selection of the NOAEL it should be considered whether the critical effect is relevant for humans; • It is not clear how the value of residual uncertainty in the estimates of S, I, and R can be determined; • Introduction of a non-scientific ‘safety’ factor C is not in accordance with the establishment of an HLV. • One should be aware that some factors may not be independent of each other. 6.2.2.7 Method used in advising the Dutch Competent Authority for occupational risk assessment The method used by TNO in advising the Dutch Competent Authority regarding the risk assessment of workers for New and Existing Chemicals is developed using literature, e.g. as mentioned in the EU-Technical Guidance Document (EC, 1996a), supplemented with information from studies of Stevenson et al. (1995a, 1995b) and with a guidance document for setting Acceptable Operator Exposure Levels (AOEL) (Anonymous, 1995). The method is used for setting Health-Based-Occupational-Reference-Values (HBORVs). The HBORV is defined as the maximum amount of a substance to which a worker can be exposed without adverse health effects being expected. For the time being a starting point is that workers may be exposed predominantly, but not exclusively, by two routes: dermally and by inhalation. HBORVs are assessed for both routes separately and for every effect (if possible) as defined in the TGD (Hakkert et al., 1996). The hazard assessment serves as starting point for the derivation of an HBORV. To translate the selected NOAEL into the HBORV assessment factors, compensating for uncertainties inherent to extrapolation of experimental (animal) data to a given human situation and for uncertainties in the toxicological data base, have to be applied. The assessment factors must be derived considering the toxicity profile of the substance. If no conclusions can be drawn a default factor will be used. The default factors are presented in Table 6.3.

119

6


Chapter 6


Table 6.3: Assessment factors applied for the calculation of HBORVs (Hakkert et al., 1996) Aspect Interspecies differences - mouse - rat - rabbit - dog Intraspecies differences Differences between experimental conditions and exposure pattern for workers: - chronic to chronic exposure - subacute to semi-chronic exposure - semi-chronic to chronic exposure - other aspects Type of critical effect Dose-response curve Confidence of the database Route-to-route

Assessment factor (default value) 7a x 3 4a x 3 2.4a x 3 1.4a x 3 3c

1 10b 10b 1 1 1 1 No default: if no relevant data on toxicokinetics and metabolism are available, worst case assumptions with respect to absorption% have to be made.

a

this is a calculated adjustment factor, allowing for differences in basal metabolic rate

(proportional to the 0.75 power of body weight)

b

the actual factor applied is often lower, and is derived from the toxicological profile of the test

substance

c

a factor of 3 is used for workers, a factor of 10 for the general population

The overall factor is established by multiplication of the separate factors, unless the data indicate another method to be used. It is stated, that one should be aware that in practice, it will be possible to distinguish all above mentioned factors, and that some factors are not independent of each other. Therefore, straightforward multiplication may lead to unreasonable high factors. Discussion and weighing of individual factors is essential to establish a reliable and justifiable overall assessment factor (Hakkert et al., 1996). Comments This approach discriminates factors to a large extent in order to distinguish between the single adjustments and to separate best estimates from uncertainty. Discrimination forces to a rational choice and to greater transparency, and invites to apply scientific consideration. However, multiplication can result in an unrealistically high overall factor. Besides, in practice, it is not possible to distinguish all above mentioned factors. One should be aware that some factors are not independent of each other. The overall factor will be the result of a number of different considerations which have to be made transparent.

120


6.2.2.8 ECETOC approach The approach recommended by the ECETOC (1995) to derive a scientific estimate of a human no adverse effect level (which is referred to in their report as the Predicted NoAdverse-Effect Level (PNAEL)) distinguishes three stages. In each of the stages an estimate is made of the most likely value of the factor described. At the end of the process the factors are multiplied together and the resultant number is used to derive the human PNAEL. The three stages are: 1. Application of a scientifically derived ‘adjustment factor’ to the NOAEL/LOAEL of the critical effect established in the pivotal study (summarised in Table 6.4). If the database is inadequate then human PNAELs cannot be derived scientifically and the recommended scheme cannot be developed further. 2. Application of an ‘uncertainty factor’ to the PNAEL to take into account the degree of scientific uncertainty involved. The following degrees of confidence in the human PNAEL are suggested as a guide based on several different required conditions: • High degree of confidence: 1 • Medium degree of confidence: 1-2 • Low degree of confidence: larger uncertainty factor. 3. Application of a non-scientific ‘safety factor’ taking into account political, socioeconomic or risk perception factors. Non-scientific safety factors are intended to account for: • Political aspects; • Socio-economic aspects (cost-benefit considerations); • Risk perception factors (the nature of the effect may justify the use of an additional factor). • No additional factors should be used if conservative (default) values were used in the stages 1 and 2. An important feature of this approach is the need to establish the route and duration of exposure to which the PNAEL refers before attempting to derive factors, since these may vary for different routes or exposure duration. For each element of the approach, ranges and default values for the numerical factors involved are recommended. Comments Like the previous method, this approach discriminates factors to a large extent in order to distinguish between the single adjustments and to separate best estimates from uncertainty. This method also gives guidance for setting occupational and non-occupational limit values. Discrimination generates a rational choice and greater clarity, and invites applying scientific consideration. However, the ECETOC approach does not mention the establishment of the

121

6


Chapter 6


overall factor. Furthermore, ECETOC gives guidance to the extrapolation step. Although they mention that all discriminated aspects introduce uncertainties, they don’t give guidance how to account for this. Finally, it can be questioned whether a non-scientific factor should be discussed in a scientific risk assessment. Table 6.4: Adjustment factors (recommended default factors) for use in deriving human PNAELs from human or animal NOAELs/LOAELs (ECETOC, 1995) Element

Factor Additional information (default value)

1. Short-term repeated/ subchronic/ chronic extrapolation: - short-term to subchronic - subchronic to chronic

3 2-3

2. LOAEL to NOAEL extrapolation

3

3. Route-to-route extrapolation

no default

4. Interspecies extrapolation - oral route - inhalation route

4 1

Extent and severity of effects may justify the use of another (higher/ lower) factor. Conversion factors must be calculated for each individual situation, making appropriate assumptions about body weight, minute volume, and percentage absorption. Based on caloric demands (4: default for the rat (Body weight 0.250 kg)). For substances with local effects on case-by case basis. Concerning toxicodynamic aspects: factor >1 only when human is considered to be more sensitive than the most sensitive species otherwise no factor.

5. Intraspecies extrapolation - general population - occupational population

3 2

6.3 Quantification of assessment factors 6.3.1 Introduction As shown in Section 6.2, the approaches described in the previous sections share many of the same underlying assumptions, judgements on critical effect, and choices of assessment factors or margins of safety. The approaches typically rely on existing human epidemiological and/or animal laboratory data. Scientists review all toxicity data, judge what constitutes an adverse effect, and determine the critical effect. Subsequently, the appropriate assessment (safety, extrapolation, or uncertainty) factors are applied to the NOAEL or LOAEL for the critical effect to account for the lack of data and inherent uncertainty in the extrapolations. Alternatively, in the case of the MOS approach, the magnitude by which the NOAEL or LOAEL exceeds the estimated exposure will be considered in view of several uncertainty parameters (e.g. interspecies differences,

122


intraspecies differences). These parameters can be compared quantitatively. The description of typical assessment and modifying factors in the development of an HLV for the different approaches is summarised in Table 6.5. The ideal method for the establishment of HLVs should fulfil a number of requirements: • Factors for extrapolation should be based on scientific data. • Extrapolation includes: short-term to long-term, interspecies, intraspecies, differences in exposure conditions between experimental or observational and the human situation for which the HLV is developed, and route-to-route; • The system should give possibilities to differentiate for: severity of effects, doseresponse curve, and data on kinetics or dynamics; • Account should be made on the adequacy of the selected study and the completeness of the database; • The choice of the factors should be motivated to assure a consistent application. • The method should correct for worst case combination of assessment factors such as occurs with multiplication. In Table 6.5 attention is also paid to these requirements. Based on the evaluations made in the previous sections it is clear that only a few approaches rely on scientific data. The factors applied will depend on the selected pivotal study and the critical effect. All methods use experts for judgement of the underlying toxicological database (completeness, relevance, and adequacy of the studies) and the severity of effect. The precision of the selected NOAEL is determined mainly by the sensitivity and relevance of the toxicological endpoint, the group size studied, and the increment between doses. Therefore, the selected NOAEL may be a poor estimate of the true but unknown NAEL. Most approaches using assessment factors basically rely on the default 100-fold factor used to derive the ADIs. The example in Section 6.5 illustrates this approach. Other approaches discussed in this and the next section will also be illustrated in Section 6.5, using the same dataset. There are no scientific grounds for the factor 100. Some scientists interpret the absence of widespread effects in the exposed human population as evidence of the adequacy of this factor. Several attempts were made to justify the subdivision in a factor 10 for interspecies and a factor 10 for intraspecies. Some apply modifying factors, on the basis of body weight, caloric demands, confidence in database, lack of independence of factors, variations in exposure circumstances, or differences between species in toxicokinetics and toxicodynamics. It is concluded that a scientific justification for the size of the factors used for intra- and interspecies differences is lacking. Several studies have been described

123

6


Chapter 6


concerning interspecies scaling, based on either a mechanistic approach or an empirical approach. The incorporation of this knowledge in a procedure for the establishment of HLVs deserves further study. The LLN, TNO, and ECETOC-method, discriminate factors to a large extent in order to distinguish between the single adjustments and to separate best estimates from uncertainty. Discrimination enhances the clarity, generates a rational choice, and necessitates an substantiation of the adjustment factors. However, multiplication can result in an unrealistically high overall factor. ECETOC does not mention the uncertainty and the establishment of the overall factor. In the TNO approach it is mentioned that weighing of the individual factors is essential because the overall factor is the result of a number of different considerations. The LLN-method introduces residual factors which do not belong to the establishment of HLVs or that cannot be quantified. Calabrese and Gilbert (1993) indicated that the applied assessment factors for a.o. inter- and intraspecies variation are not fully independent. Therefore, they proposed a modifying factor to incorporate the assumption of the lack of total independence of assessment factors in the establishment of the overall assessment factor. The choice of the applied assessment factors is seldom motivated. Transparency is essential to assure a consistent application. In addition, if more detailed information on a specific situation becomes available the assessment of an HLV may be refined. Therefore, the remaining uncertainties should be clarified. An approach which lends much more credibility to the use of the assessment factors is the investigation of the probabilistic nature of assessment factors Dourson, 1996; Baird et al., 1996; Price et al., 1997; Swartout et al., 1998; Slob and Pieters, 1998) (see also Section 6.4). The expression of the probability of the numerical value of each uncertainty factor can be based on the actual toxicity data on groups of chemicals for which HLVs have been developed. The most likely distribution for each assessment factor will be lognormal. For HLVs that have more than one area of uncertainty, the respective individual distributions can be multiplied using Monte Carlo techniques to develop an overall distribution reflecting total uncertainty which is then applied to the NOAEL or LOAEL of the pivotal study to develop a probabilistic HLV. In the following, specific attention will be paid towards the scientific information on which assessment factors are based and the quantification of the overall assessment factor. The possibilities for the application of the probabilistic approach for the derivation of a more rationalised assessment factor is further investigated in Sections 6.3 and 6.4.

124


Table 6.5: Evaluation of default assessment factors used or suggested for the establishment of HLVs Assessment factors

ADI HC JECFA JMPR

EPA (RfD)

Calabrese Renwick LLN & Gilbert IPCS

10

10

1-10

4-10

-

-

-

-

Interspecies toxicokinetics toxicodynamics oral route inhalation route

10

NDI)

1-10

10

Duration of exposure subacute/subchronic subchronic/chronic other aspects

10 H)

LOAEL to NAEL

-

-

1-10

L)

-

Route-to-route

-

-

-

L)

Type of critical effect -

-

-

Dose-response curve

-

-

Database adequacy

-

Non-scientific factor Modifying factor

Intraspecies non-occupational - toxicokinetics - toxicodynamics occupational

TNO

MOS ECETOC

- C)

+ 0-2.5E) 0-4.0E) -

+ 1-10F) +

10

+ 0-2.5 0-4.0

+ 1-10F)

+

AG)x 3 3 -

4 1

-

+

+ 1-10F) 1-10F) 1

ND

+ 3 2-3

2-10F)

ND

ND

2-3

-

-

ND

ND

ND

L)

-

-

1

ND

-

-

L)

-

-

1

ND

-

-

1-10

L)

1-10

+

1

ND

1: high 1-2:medium ND: low

-

-

-

L)

-

1F) -10 -

ND

+

-

-

>0-10

L)

1-10

+

+

-

+

Motivation for choice of factors

-

-

L)

+/-

+/-

+/-

?

+/-

Overall factor

other mult. (see 6.2.2)

L)

mult.

mult.

mult.

-

mult.

10

1-10

10

1-10

F)

mult.

+

method accounts for

-

method does not account for

K)

3

10 2

ND

+

6

mult. multiplication of different factors to establish the overall assessment factor ND

no default value, based on a case-by-case determination (expert judgement and scientific

information)

E)

IPCS recommended that the interindividual toxicokinetic and toxicodynamic default values

should be 3.16 and 3.16.

F)

default factor

125


Chapter 6


G)

this is a calculated adjustment factor, allowing for the differences in metabolic size (mouse: 7, rat: 4, rabbit: 2.4, dog: 1.4)

H)

the additional assessment factor for duration of exposure for establishment of the ADI has been recommended by FDA.

scaling on body weight or caloric demands

K)

nonspecific factor to cover unusual uncertainties or dose adjustments not covered by the

standard factors

I)

L)

Calbrese and Gilbert do not describe a full method for the derivation of HLVs

6.3.2 Quantification of assessment factors: use of historical data 6.3.2.1 Introduction In this section it will be shown how and to which extent assessment factors can be quantified on the basis of historical data (i.e. NOAELs). The uncertainty in the value of the assessment factors obtained will be taken into account by describing their entire distribution (see also Section 6.4). Possible default values for the assessment factors are derived from the higher end of these distributions for comparison with current worst-case default values. The data presented in Sections 6.3.2.2 and 6.3.2.3 confirm that assuming a lognormal distribution for assessment factors is reasonable, based on the observation that both NOAELs and ratios of NOAELs appear to be lognormally distributed. Apart from this empirical evidence, also theoretical considerations, given by the (multiplicative analogue of the) central limit theorem, favour the choice of a lognormal distribution (Slob, 1994). In addition, it can be easily shown that ratios of lognormal distributions are again lognormal (based on the fact that sums of normal distributions are again normal). 6.3.2.2. Interspecies extrapolation For extrapolation of data from animal studies to humans account should be taken of species-specific differences between animals and humans. These interspecies differences can be divided in differences in metabolic size and remaining species-specific differences. To account for differences in metabolic size three methods are used in practice: extrapolation based on body weight, surface area, and caloric demand (see Box 6.2). These methods can be described by an allometric equation: for that purpose body weight has to be raised to the power 1, 0.67, and 0.75, respectively.

126


Box 6.2: Scaling according to caloric demand Scaling according to caloric demand means that a comparable dose rate in milligram per kilogram body weight dose for the average person (70 kg) is equal to the rat (0.25 kg) dose rate divided by an interspecies factor which is equal to 70/0.25 to the power 0.25 (= 4). In formulae: Equivalent doses: mgrat/mghuman = (kgrat/kghuman)0.75 Expressed as dose rates: mgrat.kgrat-1/mghuman.kghuman-1= (kgrat/kghuman)0.75 . (kgrat/kghuman)-1 Or: dose ratehuman = dose raterat/(kghuman/kgrat)0.25 = dose raterat/4

Based on theoretical grounds, scaling on the basis of surface area or caloric demand can be considered more appropriate compared to extrapolation based on body weight (Van de Gevel and Hakkert, 1997). Experimental work did not answer the question which of these two methods is the most correct. However, based on theoretical grounds, the Health Council of the Netherlands (1985), TNO (Hakkert et al., 1996) and Kalberlah and Schneider (1998) consider the interspecies extrapolation based on caloric demands (the 0.75 power of body weight) as preferable above scaling on body weight. An allometric exponent of 0.67 seems to describe better intraspecies relations (Feldman and McMahon, 1983). To express the dose in mg/kg body weight (to the power 1) assessment factors are calculated. The size of these factors are e.g. 7 for mice (25 g), 4 for rats (250 g), and 1.4 for dogs (15 kg), etc. for the extrapolation from the test species to humans (see Box 6.2). For inhalation NOAELs for systemic effects no correction is made for differences in metabolic size, because extrapolation is already based on toxicological equivalence of a concentration of a substance in the air; animals and humans breath at a rate depending on their caloric requirements (Hakkert et al., 1996). To account for the remaining interspecies uncertainties usually a default factor is used. In theory, the remaining uncertainty could be assessed by comparing NOAELs in test animals with estimates of human NOAELs. However, in practice, such an assessment must rely on data from studies derived experimentally for the same substance in different animal species because human data are lacking. The degree of remaining interspecies uncertainty

127

6


Chapter 6


may be obtained by examining the differences (ratios) of the NOAELs established for the same substance in different species. The actual uncertainty in extrapolating from animals to humans is likely to be at least as large as the uncertainty in extrapolating among mice, rats, and dogs. An example of such an approach will be shown in the following paragraphs. For the purpose of assessing the remaining interspecies uncertainty, data from the TNO database (including pesticide dossiers, existing chemical dossiers, IPCS Environmental Health Criteria documents, JMPR evaluations, and public literature), including 184 substances tested in different species and via different routes, were analysed. NOAELs were selected from studies with mice, rats, and dogs exposed to the same test substance via the same route and with the same duration of exposure. In order to increase the comparability of the different animal experiments with respect to their duration of exposure, two categories of exposure duration were defined: semi-(chronic) and subacute. Further refinement by matching for test duration in the latter category did not affect the results of this analysis significantly. The definition of the categories is species specific, partly depending on their maximum lifetime. For mouse, rat, and dog the categories are summarised in Table 6.6. Table 6.6: Exposure duration categories for different species Exposure duration

mouse (days)

rat (days)

dog (days)

Subacute (Semi-)chronic

21-50 90-730

21-50 90-730

28-90 365-730

If within one category (same exposure duration, same test substance, and same species) more NOAELs were available, the lowest NOAEL has been used for the selection. NOAELs based on carcinogenicity have been left out of consideration. The oral NOAELs were adjusted to account for differences in metabolic size (as described above). In order to increase the comparability of the derived factors to the actual uncertainty (animal to human), the ratios were calculated by dividing the NOAELs derived in the smaller animal by the NOAEL derived in the larger animal. The following ratios were calculated: NOAELmice/NOAELrat, NOAELmice/NOAELdog, and NOAELrat/NOAELdog. The ratios (both adjusted and unadjusted for metabolic size) were evaluated by examining their distributions. Table 6.7 presents the number of ratios (N), the geometric means (GM), the geometric standard deviations (GSD), and the 90 and 95 percentiles of the distributions of the ratios. Percentiles are calculated from the GM and the GSD as shown before Section 6.1.

128


With respect to dermal toxicity, insufficient relevant data were available. For respiratory toxicity data only the NOAELmice - NOAELrat ratios were analysed: with respect to the other ratios insufficient data were available to be statistically analysed. Table 6.7: Distribution parameters derived from the NOAEL ratios Ratio NOAELrat / NOAELdog (oral, unadjusted) NOAELrat / NOAELdog (oral, adjusted) NOAELmouse / NOAELrat (oral, unadjusted) NOAELmouse / NOAELrat (oral, adjusted) NOAELmouse / NOAELdog (oral, unadjusted) NOAELmouse / NOAELdog (oral, adjusted) NOAELmouse / NOAELrat (respiratory)

N GM GSD P90 P95

N 63 63 67 67 40 40 21

GM 1.3 0.5 4.2 2.4 6.4 1.3 3.1

GSD 5.1 5.1 5.7 5.7 6.1 6.1 7.8

P90 10.4 3.6 39.3 22.5 64.7 12.9 43.6

P95 18.8 6.6 73.9 42.2 124.6 24.9 91.8

= number of ratios = geometric mean = geometric standard deviation = 90th percentile = 95th percentile

These data suggest that the distribution of the ratios can be described well by a lognormal distribution. If the interspecies differences would depend only on the differences in metabolic size, and if NOAELs were perfect estimates of the true no-effect levels (which they clearly are not), the geometric mean and the geometric standard deviation of the ratio distributions would be unity. The geometric means of the ratios of adjusted NOAELs for mouse/rats and mouse/dogs, but not for rat/dogs, are closer to one than the means of the unadjusted NOAELs which lends some support to the idea of accounting for the differences in metabolic size (scaling based on caloric demands). As an approximation of the remaining uncertainty in the extrapolation from animals to humans the mean of the distribution parameters will be used; a geometric mean of approximately 1 and a geometric standard deviation of 6. It should be noted that it is possible that the NOAELs were established based on different critical effects. Other differences not corrected for exist, such as in strain and the purity of the substance. And finally, the - probably large - noise in the NOAELs itself will blow up the dispersion of the observed distributions. In reality, the variation of the distributions will therefore be smaller. Other distributions have been proposed. Baird et al. (1996) proposed a distribution based on an analysis which is comparable to the one above, but used scaling on the basis of surface area (GMrat = 5.8 as scaling factor and GSD = 4.9). Price et al. (1997), Swartout et al. (1998) and Slob and Pieters (1998) proposed distributions considered to be consistent with the current use of the worst case default factor of 10. 129

6


Chapter 6


In summary. based on theoretical grounds, and supported by the analysis given in Table 6.7, scaling on the basis of caloric demand to adjust oral NOAELs for metabolic size can be considered more appropriate compared to extrapolation based on body weight. According to the example analysis shown, the assessment factor accounting for the remaining (i.e. after metabolic scaling) uncertainty in the extrapolation from animals to humans may be characterised as approximately lognormally distributed with a geometric mean of about 1 and a geometric standard deviation of 6 (Figure 6.1). Based on this distribution, default values for the 90-, 95- and 99-percentiles can be calculated to be 10, 19 and 65, respectively.

  Figure.6.1: Distribution of the interspecies assessment factor, adjusted for metabolic size  (rat-human). In this distribution the often applied default factor of 12 (adjustment for  metabolic size 4, remaining uncertainty 3) coincides with the 73th percentile   6.3.2.3 Intraspecies extrapolation  The response of humans to exposure of xenobiotic compounds may vary due to a number  of biological factors, such as age, sex, genetic composition and nutritional status. To   account for interindividual human variation a factor of ten for the extrapolation from the  average to the sensitive human being is generally assumed to be appropriate for deriving  HLVs. In an initial attempt to document the extent of human interindividual variation,     130        


Calabrese (1985) found considerable differences among human subjects in their capacity to metabolize foreign substances and reached the conclusion that in most cases a factor of ten would be sufficient to protect the majority (up to 80-95%) of the human population against adverse health effects. This conclusion was based on the observation that the vast majority of responses assessed seemed to fall clearly within a factor of 10. Exceptions may be due to e.g. increased susceptibility due to serious illness. Also genetic polymorphisms of metabolising phase I and phase II enzymes may cause a large variation in human responses (Daly et al., 1993). In only a few publications an attempt was made to investigate the human interindividual variation by data analysis. Hattis et al. (1987) investigated the total variation in pharmacokinetic behaviour of 49 pharmaceuticals in healthy adults. Depending on the pharmacokinetic parameter studied (elimination half-life, area under the curve, peak concentration), a tenfold difference in these pharmacokinetic parameters would correspond to 2.5 to 9 standard deviations in populations of normal healthy adults. Reanalysis of the data of Hattis et al. (1987) showed that for the half-life time the variation between individuals was quite small. Defining the intraspecies factor as the ratio of the P50 and P05 resulted in a factor of 1.4 (Schaddelee, 1997). Although from this analysis it appears that a factor of ten will be sufficient for pharmacokinetic variation, the real median to sensitive human variability is underestimated, since one should take into account that (i) variation also exist in pharmacodynamics and (ii) that only data of healthy volunteers were available. Renwick (1991, 1993a) analysed interindividual differences of healthy volunteers and patients by comparing the maximum and mean values of pharmacokinetic parameters and the minimum and mean values of pharmacodynamic parameters. Based on this analysis he proposed to subdivide the factor of ten into a factor of four for pharmacokinetic differences and a factor of 2.5 for pharmacodynamic differences. Re-analysis of the Renwick data by using distributions instead of ratios max/mean and min/mean gave comparable results (Schaddelee, 1997). The results of Renwick’s analysis have been adopted by the IPCS (1994). However, rather than using default factors which comply with a more or less arbitrarily chosen default factor of ten, it would be better to use toxicity profile-derived pharmacokinetic and pharmacodynamic factors. This would require further and better data analysis. Data availability, however, may be a problem here. By performing data analysis one should take experimental errors into account, becaue these may be substantial. The latter has not been considered in Renwick’s analysis.

131

6


Chapter 6


Based on an analysis of the available human data, Kalberlah and Schneider (1998) proposed an intraspecies factor of 25 for the general population, composed of a factor of 8 accounting for toxicokinetic variation and enzyme polymorphisms, and a factor of 3 accounting for toxicodynamic variation. For workers this factor is reduced and a total factor of 5 is considered to account for both inter- and intraspecies variation (after adjustment for differences in metabolic size). They claim that only in a few cases the sensitivity of special groups at risk will exceed these ranges. As the authors admit, it can be noted that this proposal is based on an overall impression based on several substance-specific examples. The combined factor for workers accounting for both inter- and intraspecies variation is not adequately explained. Several probabilistic distributions have been proposed. Baird et al. (1996) proposed a distribution on the basis of acute toxicity data on heterogeneity in rats and on the basis of assumptions on the unknown difference in heterogeneity between rats and humans (GM = 2.7 and GSD = 2.3 with rats and humans equally heterogeneous and GM = 5.3 and GSD = 2.1 with humans 1.5 more heterogeneous than rats). Price et al. (1997), Swartout et al. (1998), and Slob and Pieters (1998) proposed distributions considered to be consistent with the current use of the worst-case default factor of 10. In summary, no proposal for a database-derived distribution of the intraspecies factor can be made. Therefore for the time being it can be considered appropriate to remain consistent with the traditional default value of 10 and to assume that this value protects the majority of the general human population. For workers one could remain consistent with the traditional default value of 3. 6.3.2.4 Subchronic to chronic extrapolation When only subchronic (subacute and semi-chronic) data are available an extra assessment factor (usually ten) is currently used to extrapolate to chronic exposure. For the distribution of the extrapolation factor several studies comparing NOAELs from chronic and subchronic studies appear relevant (Kalberlah and Schneider, 1998; Weil and McCollister, 1963; Rulis and Hattan, 1985; Kramer et al., 1996; Pieters et al., 1998; Nessel et al., 1995). Tables 6.8 and 6.9 summarise the oral data.

132


Table 6.8: Semi-chronic to chronic oral NOAEL-ratiosa N

GM

GSD

P90

P95

33 41 20 149 23

2.2 1.0 1.9 1.7 2.0

9 11 20 21 22

2.4 1.7 2.0 1.7 2.5

Chronic exposure period 2 years not specified > 200 days 1-2 years 2 years

Species

Reference

8.7 2.5 12.0 29 5.1

Semi-chronic exposure period 30-210 days not specified < 200 days 10-26 weeks 90 days

2.3 1.7 3.0 5.6 1.8

6.4 2.0 8.0 15.4 4.2

rats rats, dogsb various various rodentsc

Weil and McCollister, 1963 McNamara, 1976 Rulis and Hattan, 1985 Pieters et al., 1998 Nessel et al., 1995

1.3 1.8 2.4 1.7 1.9

3.4 3.6 6.1 3.3 5.7

3.7 4.5 8.4 4.1 7.2

90 days 90 days 90 days 90 days 90 days

1-2 years 1-2 years 1-2 years 2 years 2 years

mice rats mice + rats mice rats

d d d d d

N = number of ratios, GM = geometric mean, , GSD = geometric standard deviation,

a

P90 = 90-percentile, P95 = 95-percentile (percentiles calculated from the GM and GSD) b

39 rat pairs, 2 dog pairs

c

matched pairs

d

Industry data from 13 agrochemicals, Kalberlah and Schneider, 1998

e

Data from the US National Toxicology Program

Table 6.9: Subacute to chronic oral NOAEL-ratiosa N

GM

GSD

71 20 26

4.1 3.1 3.9

4.4 1.9 2.2

P90

P95

Subacute exposure period

Chronic exposure period

Species

Reference

27 7.0 10.7

46 8.9 14.3

3-6 weeks 14 days 14 days

1-2 years 2 years 2 years

various mice rats

Kramer et al., 1996 Kalberlah and Schneider, 1998b Kalberlah and Schneider, 1998c

N = number of ratios, GM = geometric mean, , GSD = geometric standard deviation, P90 = 90-percentile, P95 = 95-percentile (percentiles calculated from the GM and GSD) b Industry data from 13 agrochemicals c Data from the US National Toxicology Program a

6

These studies assessed the ratios of observed NOAELs from subchronic versus chronic oral tests using historical data for a sample of various compounds. It is very likely that the databases used in these studies overlap each other significantly. In three studies the ratios have not been matched for the species concerned (Rulis and Hattan, 1985; Kramer et al., 1996; Pieters et al., 1998). The NOAELs in these studies have not been normalized for any differences in basal metabolic rate between the test species used for the subchronic test and that used for the chronic test. No doubt there will also be differences in the interpretation of the tests available (if interpretation is done at all in the case secondary sources have been used).

133


Chapter 6


It should be noted that subchronic toxicological studies usually have smaller sample sizes compared to chronic studies (typically twice as small). Therefore, it may be expected beforehand that NOAELs from subchronic studies will tend to be larger than NOAELs from chronic studies, even if the true dose-response relationships in both studies were identical. Thus, the geometric mean ratios for the NOAELs assessed in the studies mentioned most likely overestimate the median of the distribution of the EFsubchronic. Which distribution can now be thought to approach reality best? The main dichotomy in the meta-studies performed is in the way the available database has been analysed: several authors have computed ratios regardless the species of both (the lowest) NOAELs, whereas others have done so for (the lowest) NOAELs of the same species only. In the latter case lower mean ratios may be expected. . However, as a result of the reduced number of ratios available, the estimate of the variation may be poor. In the daily practice of risk assessment an HLV can be derived from a subchronic test applying an extra assessment factor. In such cases the question is what, given the lowest subchronic NOAEL, the value of the chronic NOAEL would have been, if a chronic test had been performed that would be acceptable to derive an HLV. Ideally, this chronic test should have been carried out using the most appropriate animal model. It follows that the most relevant NOAEL-ratios are those based on the same exposure period and the same species (in order to exclude interspecies variation) and the most relevant distributions of NOAEL-ratios are those that include a sufficient number of matched pairs of NOAELs of various species. Unfortunately, the available distributions (Tables 6.8 and 6.9) are not “the most relevant” since these are based on rather variable exposure periods for the semi-chronic NOAELs, include interspecies variation (no matching for species), for which no correction was made, and in several cases use rather old data. Differences in endpoints were not considered. The distributions obtained from NOAELs of various species therefore will probably be too wide, whereas the distributions obtained from NOAELs of one species and more strict criteria with regard to exposure period and overall study design might be too narrow. In summary, it does not seem appropriate to rely on one particular meta study since the selection criteria used for different distributions all have advantages and disadvantages. All individual distributions have to be taken into account. The geometric means of the oral semi-chronic to chronic ratios were similar in all these studies, i.e. approximately 2, whereas the GSD ranges from 1.3 to 5.6 as a result of use of different species, variable size of the database, and criteria for data selection. Selecting an appropriate distribution by average is not recommended since the data sets are dependent. Based on all data together a GSD of 4 is considered a reasonable approximation of the real standard deviation (Figure 6.2). This is lower than the standard deviation derived from the large database of Pieters

134



         


et al. (1998). However, in the latter database, interspecies differences might have played a role, since the NOAEL-ratios were not matched for species. The proposal is higher than could be assumed on the basis of studies with strict selection criteria. Based on the proposed distribution, default values for the 90, 95 and 99 percentiles can be calculated to be 12, 20 and 50, respectively. Other distributions have been proposed. Baird et al. (1996) proposed a distribution based on two pooled datasets with a total of 51 NOAEL-ratios of both oral and inhalation studies (GM = 2.1 and GSD = 2.5). Swartout et al. (1998) and Slob and Pieters (1998) assume distributions considered to be consistent with the current use of the worst case default factor of 10. The geometric means of the oral subacute-to-chronic ratios were significantly higher than of the semi-chronic-to-chronic ratios and show less consistency. Based on the data available it seems reasonable to approximate their real distribution with a geometric mean of about 4 and a geometric standard deviation of 4. Based on this distribution, default values for the 90, 95 and 99 percentiles can be calculated to be 24, 39 and 101, respectively.

6

  Figure 6.2: Distribution of semi-chronic to chronic assessment factor. In this distribution  the often applied default factor of 10 coincides with the 88th percentile   135        


Chapter 6


Whether the distributions also apply to inhalatory and dermal subchronic-chronic ratios is questionable. It might be possible that the influence of exposure period on the toxicological effect depends on the route of exposure. Further analysis of results from dermal and inhalatory studies must give more insight in this uncertainty. 6.3.2.5 Dose-response curve The dose-response curve is not used to derive a NOAEL. Apart from ensuring that the number and spacing of data points is adequate to provide a reasonable estimate of the NOAEL, all other data points are ignored. In theory, the steeper the slope of the curve, the smaller the assessment factor can be and vice versa. The threshold dose can be under- or overestimated, respectively. There is no scientific basis for any value of a default factor to account for uncertainty in the noael, nor any distribution. Alternative methods such as the Benchmark-dose concept do take the shape of the dose-response curve into account. The extrapolation from the LOAEL to the NOAEL may be regarded as part of the doseresponse analysis. LOAEL to NOAEL The use of historical LOAEL/NOAEL-ratios to estimate a NOAEL from a LOAEL (Dourson and Stara, 1983; Pieters et al., 1998) is questionable. Doses in toxicological tests are usually spaced in fixed intervals and the observed distribution of LOAEL/NOAEL ratios therefore primarily reflects the historical frequency of use of various dose spacing (Baird et al., 1996). There is no guarantee whatsoever that extrapolation of a LOAEL with any factor will yield an estimate of the NOAEL. Therefore this factor can only be assigned using expert judgement in which the shape of the dose-response curve and the magnitude of the effect at the LOAEL is taken into account. 6.3.2.6 Route-to-route extrapolation In the case relevant data are lacking on exposure routes of interest, route-to-route extrapolation is used in the risk assessment. The currently applied route-to-route extrapolation methodology is an easy, straightforward way to determine a dermal or inhalation NAEL based on an oral NOAEL. To account for differences between routes of exposure, data on absorption or acute toxicity are often used. However, these methodologies are not validated and are based on broad assumptions. A study was performed that was aimed at an evaluation of route-to-route extrapolation on the basis of (estimates of) absorption or acute toxicity data (Wilschut et al., 1998). By using experimental repeated-dose toxicity data, it was tried to establish a factor to account for other factors than bioavailability which are generally not taken into account in route-toroute extrapolation. 136


Data were primarily gathered on dermal and respiratory repeated-dose toxicity. An extrapolation factor, defined as the factor that is applied in route-to-route extrapolation to account for differences in the expression of systemic toxicity between exposure routes, was determined for each substance by using data on absorption and acute toxicity data. As experimental data on absorption are often not available, default values for absorption were also used to determine an extrapolation factor. Despite a rather large overall database, it was remarkable that relatively few data could be used for the evaluation. Therefore, conversions were performed to include data that initially were considered less suitable for data analysis: interspecies extrapolation based on caloric demands was introduced, and a factor of 3 was applied in case a LOAEL instead of NOAEL was available. The choice of NOAELs for different exposure routes known for a substance, suitable for analysis was primarily based on the same effect. However, this criteria could not be maintained. It appeared that for oral to respiratory route-to-route extrapolation (n=28), the predicted NAEL was often higher than the observed NOAEL. So, the substance was considered less toxic after extrapolation when compared with experimental observations. Based on the 95th percentile of the log-normal distribution of the ratios between the predicted NAEL and the observed NOAEL, uncertainty factors ranging from 75 to 201 for the different extrapolation methodologies were found. For oral to dermal route-to-route extrapolation (n=25), the predicted dermal NAEL was often lower than the observed NOAEL, e.g., the substance was often considered to be more toxic after extrapolation when compared with experimental observations. Uncertainty factors ranging from 2.7 to 35 were found for the different extrapolation methodologies. Given the implications of the use of these uncertainty factors, (inter)national discussion on the results of the study of Wilschut et al. (1998) would be of value for optimal tuning in view of future application. As a result of such discussions more data may become available. It should be noted that the reliability of the data is questionable as the influence of the several assumptions made in order to derive comparable data on the ratio of the predicted NAEL and the NOAEL is unknown. For both extrapolations, the results were hardly influenced by the assumptions made on absorption, indicating that other factors may be important in route-to-route extrapolation, and/or the reliability of the estimates of absorption used in this study was poor.

137

6


Chapter 6


Given these findings, it was concluded by Wilschut et al. (1998) that the development of scientifically based principles and procedures for route-to-route extrapolation appears to be a difficult task without the availability of adequate experimental data. In this study, scientific justification for the application of route-to-route extrapolation was not derived. As only a limited number of data on toxicity after repeated dermal and inhalatory exposure were found after an extensive search, it is doubtful whether such experimental data indeed do exist. Insight in the reliability of route-to-route extrapolation methodologies may then only be obtained if more suitable experimental data become available. In summary, several options can be considered to deal with the issue of route-to-route extrapolation: 1. Assessment factors are used to account for uncertainties in the route-to-route extrapolation; 2. Repeated-dose toxicity studies with exposure routes relevant for the risk characterisation are performed until validated and reliable route-to-route extrapolation methodologies are available. The role of PBPK modelling should also be further investigated here. The choice between these options is a regulatory one and depends on the desired degree of reliability of the risk assessment. At this moment, the application of route-to-route extrapolation heavily depends on expert judgement. 6.3.2.7 Type of critical effect The type of critical effect should be taken into account. Assessment factors may be applied by expert judgement depending on each individual case. By default it can be assumed that no extra correction is necessary. 6.3.2.8 Confidence in the database The size, quality, completeness, and consistency of the database should be considered. The schemes available indicate that the assessment factor should be higher than unity in the case one is less confident about the database and may run up to 100. This assessment factor can only be assigned on the basis of expert judgement, preferably made transparent through the application of a set of criteria. It may be argued that a database necessitating very high assessment factors are probably inadequate for the risk assessment altogether. ECETOC proposed to distinguish between a high, medium and low degree of confidence. The default (high confidence) is 1. A medium degree of confidence would warrant a higher assessment factor, running up to 10 for a low confidence database.

138


Criteria should be developed to make expert decisions transparent before implementing this approach. 6.3.3 Combining assessment factors In the standard procedure for deriving acceptable limit values, various assessment factors are multiplied to obtain an overall assessment factor. However, multiplication of assessment factors implies a piling up of worst case assumptions: the probability of simultaneous occurrence of worst case situations for the same chemical will be smaller than that of a single worst case situation to occur. Therefore, the more extrapolation steps are taken into account, the higher the level of conservatism. The piling-up of worst-case assumptions can be avoided by using probability distributions (Baird et al., 1996; Price et al., 1996; Swartout et al., 1998; Slob and Pieters, 1998). In this method each assessment factor is considered uncertain and characterised as a random variable with a distribution. Propagation of the uncertainty can be evaluated using Monte Carlo simulation yielding a distribution of the overall assessment factor. This method requires characterisation of the distribution of each assessment factor, as was attempted in this section, and of possible correlations between them. As a first approach it can be assumed that all factors are independent.

6.4 New concepts in deriving Human Limit Values 6.4.1 Introduction In the standard procedure for deriving Human Limit Values (HLVs), such as ADI, TDI, RfD, or HBORV from animal study data, the NOAEL is divided by a number of assessment factors according to Equation 6.4.1:

ADI , TDI , RfD =

NOAEL AF1 . AF2 . AF3 ......

(6.4.1)

The assessment factors are assumed to be independent from each other (see also Section 6.3.1). Because of this multiplication the standard method for deriving HLVs is generally considered to be conservative. Indeed, when each individual assessment factor by itself is regarded to reflect a worst case situation, their product, i.e. the overall assessment factor, will tend to be overly conservative. However, the degree of conservatism in the limit value in any particular assessment is unknown, which hampers risk managers to appraise possible health risks against other (e.g. economic) interests (See some attempts to address this problem: for example, the US-EPA adds a quantitative description of the uncertainty factors, and confidence descriptions for each of its RfDs and RfCs).

139

6


Chapter 6


On the other hand, the uncertainty in the numerator, the No-Observed-Adverse-Effect Level (NOAEL) as an estimate of the “true” No-Adverse-Effect Level (NAELtrue) in the animal is completely ignored. The NOAEL is defined as the highest dose level at which no statistically significant effects occur for all endpoints that are considered toxicologically relevant. This NOAEL is not the same as the NAELtrue. Suppose there is a (true, but unknown) threshold dose below which the substance does not evoke any adverse effects. Depending on the study design, the NOAEL resulting from a statistical analysis of the data can be lower or higher than this dose. The potential deviation of the NOAEL from the NAELtrue cannot be quantified. The latter uncertainty may be substantial and ignoring it may introduce an anticonservative or an additional conservative element in the derivation of acceptable exposure limits. This section will further examine the uncertainties present in both the numerator and the denominator of equation 4.1. To this end first a conceptual framework will be presented as proposed out by by Slob and Pieters (1998). In subsequent sections this concept will be operationalised. The practical implications of this operationalisation is shown in an example in Section 6.5.3.3. 6.4.2 The concept Assume there is a true No-Adverse -Effect Level in the sensitive human, or NAELsens.human. As this NAELsens.human usually is derived from animal data, extrapolation factors have to be applied. To that end, the true factor (EFtrue) is defined as an alternative for the assessment factor (AF, see Equation 6.4.1). The EFtrue, interspec is defined as the ratio between the ‘true’, but unknown, NAEL in the animal (NAELtrue,animal) and the ‘true’, but unknown, NAEL of the average human (NAELtrue,human). Any particular compound has its own EFtrue,interspec.

EFtrue,interspec ≡

NAELtrue,animal NAELtrue,human

(6.4.2)

Similarly, the intraspecies EFtrue is defined as

EFtrue,intraspec ≡

NAELtrue,human NAELtrue,sens.human

(6.4.3)

NAELtrue,animal EFtrue,interspec EFtrue,intraspec

(6.4.4)

Clearly, for a particular compound we have

NAELtrue,sens.human =

Although Equation 6.4.4 has the same appearance as the standard equation (6.4.1), it fundamentally differs in interpretation: all entities in (6.4.4) refer to true but unknown values. 140


For the operationalisation of this concept, the question therefore is how to estimate the NAELanimal and the EFs and the uncertainty distribution associated to each of them. The next section will deal with the best approximation of the distribution of the NAELanimal. With regard to the EFs is can be argued that, alhough the value of the EFs are unknown for specific compounds, the extrapolation factors for the universe of all compounds must have a specific distribution. One might be able to estimate that distribution from historical data (e.g. from drugs). Ideally this should be done on the basis of ratios of the best approximations of the NAELtrue. More crude estimates of the distributions of EFs can be obtained on the basis of NOAELs as was done in the previous section. It was argued that the database-derived distributions thus obtained are wider than would be obtained on the basis of the NAELtrue. 6.4.3 Estimation of the No-Adverse-Effect Level in the animal 6.4.3.1 The NOAEL and the true NAEL The numerator in Equation 6.4.1, the NOAEL, is defined as the highest dose level at which no statistically significant effects occur (for all endpoints that are considered toxicologically relevant). As pointed out in the introduction to this section, it is important to keep in mind that the NOAEL is not the same as the NAELtrue and that, although the NOAEL could be considered an estimate of the true threshold dose, the quality (precision) of the estimate cannot be assessed. Other objections against the use of the NOAEL have been discussed extensively elsewhere (McColl, 1989; Crump, 1984; Beck et al., 1993). In general, there is a call for consideration of the dose-response relationship as a whole. One of the alternatives proposed has been the Benchmark approach (Crump, 1984; US EPA, 1995). 6.4.3.2 The Benchmark dose concept In the Benchmark approach a regression function fitted on response data is used to estimate the dose at which adverse effects start to arise. Using regression models for describing the dose-effect relation has two advantages. Firstly, a ‘modifying factor’ to account for the steepness of a dose-effect curve is redundant and secondly, there is no need to extrapolate a LOAEL to a NOAEL. The latter may be considered as a major advantage since there is no scientific justification for the use of an assessment factor for LOAEL-NOAEL extrapolation.

141

6


Chapter 6


In the Benchmark concept one needs to postulate a critical effect size (CES) below which there is no reason for concern. The CES for a critical endpoint is defined as: CES = value of effect-size below which there is no reason for concern and the associated Critical Effect Dose (CED) as: CED = dose at which the average animal shows the (postulated) critical-effect-size defined for a particular endpoint. A drawback of using dose-effect curves for the evaluation of toxicity is that current toxicological and biological knowledge does not provide sufficient basis to unequivocally establish the breaking point between non-adverse and adverse effect size for most endpoints. Since a single ‘universal’ CES does not seem a realistic option, a value must be chosen for each separate endpoint. A wide-spread implementation and acceptance of the value of CES for each of the (most relevant) endpoints would require international consensus on this issue. The quantification of effect-size for continuous, quantal and ordinal data is discussed in Slob and Pieters (1998). The CED is referred to here as a true, unknown value, which we can only estimate with a certain degree of precision, if data for the endpoint of concern are available. The true NoAdverse-Effect-Level (NAELtrue) may then be defined as the lowest CED of all endpoints: NAELtrue = minimum of all CEDs. The definition of the NAELtrue refers to a true, unknown value, which can only be estimated with a certain degree of precision, if data are available for the endpoint involved. The definition of the NAEL refers not only to an unknown, but also to a rather theoretical value, since it is unknown to what endpoint it is associated. In practice, one can never be sure whether information on all relevant endpoints for the compound studied is present. Furthermore, the lowest CED in two situations (e.g., animal versus human) may not refer to the same endpoints. For example, rats may be most sensitive to endpoint A, but humans to endpoint B. EFs, as discussed in the previous section, therefore should be applied and can best be approximated by the ratio of CEDs per endpoint. The variation of EFs between all endpoints and all substances should preferably be expressed by the distribution of these CED-ratios rather than by the distribution of NOAEL-ratios. Unfortunately, quantitative knowledge on these distributions of CED-ratios, or Benchmark doses, is scarce. A drawback on the use of dose-response modelling from a practical point of view, is that

142


most toxicity data are not suitable for curve-fitting procedures (Crump. 1984; Beck et al., 1993; Woutersen et al., 1997). A typical study design as agreed on in, for example ,OECD guidelines considers three dose groups and a control. Ideally, more dose groups should be used with each dose group comprising less animals. See Slob and Pieters (1997) and Kavlock et al. (1996) for elaboration on this issue. 6.4.3.3 The probabilistic approach towards the CED When for a particular endpoint data are available that allow for fitting a regression function, the CED may be estimated. Depending on the quality of the data, this estimate has a certain degree of imprecision. To take this into account, Crump (1984) proposed to calculate the lower 95% confidence limit of the estimated CED. Slob and Pieters (1998) proposed to find the complete uncertainty distribution of this estimate by bootstrapping: once a regression model has been fitted, Monte Carlo sampling is used to generate a large number of new data sets from this regression model, each time with the same number of data points per dose group as observed animals in the real experiment. For each generated data set the CED is reestimated. Taking all these CEDs together, results in the required distribution. 6.4.4 The probabilistic approach towards the HLV Because for each EF a certain distribution over all endpoints and substances is assumed it is possible to extrapolate any CED from one situation to the other. Thus, instead of choosing a single (most sensitive) endpoint from the animal data, each CED distribution that is associated to a relevant endpoint is extrapolated to the distribution of the associated CED in the sensitive human (CEDsens.human) by probabilistic combination with the distributions of each EF. This results in a series of distributions for CEDsens.human, each related to another endpoint. Then this complete set of distributions can be considered as a basis for deriving a HLV, for example by choosing the lowest of each distribution’s first percentile. It is noted that the assumption of complete independence of the various distributions of EFs is also applied here. It has been argued that this worst-case assumption may not be valid (Calabrese and Gilbert, 1993). In the case correlations can be demonstrated and quantified the method can allow for these by introducing correlation coefficients. Figure 6.3. illustrates the proposed approach.

143

6



        Chapter 6   ·· · ·· · · ·

CES

· · ··

· · · ·· · ··

· · · ·· · · · ·

· · · · · ·

dose

CEDanimal NAELsens. human

= EFinterspec

EFintraspec

. . .

 Figure 6.3: The probabilistic determination of a Human Limit Value 6.4.5 Conclusions The approach as discussed above differs from the Benchmark approach (Crump, 1984) in various ways. Crump introduced the “Benchmark Dose Level (BMDL)”, defined as the lower 95% confidence limit of the CED, as a starting point for extrapolation to the (sensitive) human. By dividing the BMDL by assessment factors for interspecies and intraspecies variation (default values of ten), a HLV can be derived. Instead of this, it is first of all proposed to use the entire distribution of the CED instead of the lower 95% confidence limit of the Critical Effect Dose (CED). Secondly, it is proposed to combine this CED distribution with distributions of extrapolation factors in a probabilistic way. The result of the probabilistic combination of distributions is in the form of an assessment

144



    




distribution, so that the degree of conservatism is quantifiable in any particular assessment. As a matter of fact, this approach allows for deriving a HLV as a function of an a priori chosen degree of conservatism. In addition, the approach allows for estimating the lower and upper bounds for possible health effects in the sensitive population at a given exposure level.

6.5 An example 6.5.1 Introduction The different approaches discussed in previous sections will now be applied to an example substance EXA. EXA has an oral NOAEL of 400 mg.kgbw-1.d-1. This NOAEL was derived in a 3-month test (semi-chronic) in rats. This example will only include extrapolation from experimental animals to average humans (AF1), from average humans to sensitive humans (AF2) and from a semi-chronic toxicity test to a chronic test (AF3). In all approaches the 95th percentiles (P95) of distributions of assessment factors are chosen, in concordance with the widespread use of 5% as the level of significance in statistical tests (including those applied to toxicological data in assessing the NOAEL). Disclaimer: These example calculations are presented with the purpose of showing the procedures in each of four approaches. It should be recognized that no general conclusions can be attached to the quantitative outcome for this particular substance.

The different approaches applied are: 1. Multiplication of assessment factors (traditional approach using default factors of 10); 2. Probabilistic multiplication of database-derived assessment factors, proposed in Section 6.3.2; 3. Determination of the distribution of a Critical Effect Dose and, combining this with distributions of database-derived assessment factors, the probabilistic estimation of the HLV as discussed in Section 6.4. In all examples, the following equation is used to derive the HLV:

HLV =

NOAEL AFtot

(6.5.1)

145

6


Chapter 6


The overall assessment factor AFtot is calculated from the following formula:

AFtot = AF1. AF2 ...... AFn

(6.5.2)

6.5.2 Input data In an OECD-test, groups of 20 rats of both sexes received through their diet doses of 0, 400, 1200 or 4800 mg.kgbw-1.d-1 for 90 days. Increased mortality was observed in female rats at the highest dose. The main other effects observed were on body weight, lactate dehydrogenase levels and histopathology of the urinary bladder (mucosal hyperplasia). The mean effects data are shown in Table 6.10, but individual data for both sexes combined were used for modelling. For the purpose of this example all methods discussed will concentrate on the effect on lactate dehydrogenase levels. Table 6.10: Results of the semi-chronic test of EXA1 Dose (mg.kgbw-1.d-1) 0 400 1200 4800 1

Survival m 19/20 20/20 20/20 19/20

f 19/20 20/20 18/20 15/20*

mean body weight (g) m f

mean LDH level2 incidence of UBH (bb/ml) m f m f

480 480 453* 346*

1893 2075 2442* 4637*

264 266 274 252*

1427 1584 1971* 3866*

0/20 0/20 2/203 15/204

LDH = lactate dehydrogenase, UBH = urinary bladder mucosal hyperplasia, m = males, f = females

2

LDH-levels were determined in 10 rats/sex/dose.

3

very slight

4

very slight (3m, 9f), slight (3m, 3f), moderate (6m, 1f), marked (3m, 0f)

* = statistically significant (t-test)

6.5.3 The derivation of the HLV 6.5.3.1 Multiplication of assessment factors (traditional approach) The default assessment factors used traditionally take on values of 10: 10 for AF1 = interspecies factor 10 for AF2 = intraspecies factor 10 for AF3 = factor for duration of exposure. The HLV will be 400/(10.10.10) = 0.4 mg.kgbw-1.d-1 The overall assessment factor is 1000.

146

0/20 0/20 3/203 13/204


Comment: When sufficient data on a substance are available, default values may be replaced by toxicity profile derived (data derived) assessment factors. An example of such an approach is shown in Annex I. 6.5.3.2 Probabilistic multiplication of assessment factors The same formulae apply as stated above, but now some of the assessment factors may have a distribution of values, as proposed in Section 6.3. The propagation of the uncertainty in each assessment factor can be performed by Monte Carlo simulation. This analysis is performed by sampling randomly from the distributions specified for each assessment factor and combining these values to a distribution for the overall assessment factor. It is assumed here that the factors are not correlated. The data relevant to the Monte Carlo simulation for EXA are shown in Table 6.11. Propagation of the uncertainty was investigated using 10,000 runs. (Note that in this case, where lognormal distributions are multiplied, the propagation of uncertainty can also be calculated exactly, see e.g. Slob (1994); Monte Carlo analysis is required however, as soon as the distributions are not exactly lognormal, e.g. when a shifted lognormal distribution for AF2 were applied, having a lower bound of unity instead of zero). The histograms of the distributions in Table 6.11 are shown in Figures 6.1 (interspecies factor) and 6.2 (semi-chronic to chronic factor). The distribution of the overall assessment factor AF1.AF2.AF3 is shown in Figure 6.4. This assessment factor AFtot has a median of 80 and a P95 of 3300. Based on the P95 the HLV will be 400/3300 = 0.12 mg.kgbw-1.d-1 Table 6.11: Input for the Monte Carlo simulation for EXA Assessment factor

Distribution

Geometric mean 4 1

Geometric standard deviation 6

AF1: interspecies, kinetics interspecies, residual

discrete value lognormal

AF2: intraspecies

discrete value

10

-

AF3: duration of exposure semi-chronic to chronic

lognormal

2

4

6

147


Chapter 6




  Figure.6.4: Distribution of AFtot: the median is 80 and the GSD is 9.6: it follows that P87 is  1000, P90 is 1452, P95 is 3303 and P99 is 15424

  Comment: This method partly characterises the probabilistic uncertainty in the HLV as   derived from experimental data. The uncertainty in the experimental NOAEL itself is  ignored.   6.5.3.3 Probabilistic estimation of the HLV  In  this example the probabilistic approach as discussed in Section 6.4 will be applied to the example substance including the distribution in the NAEL, estimated from the dose  response curve.   The dose-response curves fitted to the male and female LDH data are shown in Figure  6.5. No difference in sensitivity between males and female rats is apparent and for both  sexes a CEDanimal of 967 mg.kgbw-1.d-1 is derived ata CES of 20%. The CES of 20% is     chosen by expert judgement and therefore is a matter of debate. The associated uncertainty  distribution of the CED at a CES of 20%, obtained by 1000 bootstrap runs (see Slob and  Pieters, 1998), is shown in Figure 6.6. This distribution is combined by Mont Carlo analysis    148






with the (distributions of) the assessment factors AF1, AF2 and AF3, as shown in Table 6.11. This results in the distribution of the CED for the sensitive human population, CEDsens.human, which is shown in Figure 6.7. Taking the HLV to be the P05 of the CEDsens.human, the HLV is 0.25 mg.kgbw-1.d-1.







 



   Figure 6.5: Regression curves for male (triangles) and female (plusses) rats exposed for 3 months to  EXA. The parameters a1 and a2 represent the LDH control levels for males and  females, respectively; var denotes the residual variance of LDH (on log-scale) with respect  to the fitted function, and lik denotes the log-likelihood value associated with the fitted function

 



149

6



    Chapter 6   

   Figure 6.6: Uncertainty distribution for the CEDanimal for EXA at a CES of 20%. L05 is 5% lower confidence limit, L95 is 95% upper confidence limit

150





 


  Figure 6.7: Distribution of the dose level of EXA that would evoke a 20% adverse effect in    5th percentile of this distribution (L05 on x-axis) is   The the sensitive human subpopulation.  0.25 mg.kgbw-1.d-1   Table 6.12 summarizes the results of the derivation of the HLV of the example substance   EXA according to the three approaches considered. Again it should be emphasized that no  general conclusions can be derived from the quantitative outcome. The example merely  shows the methodological differences and the difference in results. 







Table 6.12: Summary of results of the derivation of the HLV of EXA Method 

 NOAEL + default AFs of 10 NOAEL + Probabilistic AFs  Probabilistic CED and AFs

      

     dose Critical in animal study  mg.kgbw-1.d-1  NOAEL = 400 NOAEL = 400  CED = 967  (90%-C.I.: 865 - 1095)

AF1 

 10  GM=4 GSD =6  GM=4  GSD = 6

AF2

10 10

  

10

AF3









HLV -1 -1 .d  mg.kgbw 

 10  GM = 2 GSD =4 GM =2 GSD = 4

 0.4 0.12   0.25

151

6


Chapter 6


6.6 Recommendations and conclusions 6.6.1 MOS versus NOAEL The human risk characterisation according to the EU Technical Guidance Documents for new and existing substances (EC, 1996a) is based on the NOAEL/exposure ratio, i.e. the Margin of Safety (MOS). It can be concluded that more explicit guidance for human risk characterisation is needed to reduce variability and to achieve higher transparency and consistency between risk assessments of individual Member States. A formal, harmonised set of default assessment factors for the derivation of a Human Limit Value (HLV) or for judgement of the MOS is considered essential here. 6.6.2 Assessment factors In this discussion document the quantification of assessment factors is addressed. Historical data were analyzed to determine the distribution of the factors for interspecies differences and for the duration of exposure. The probabilistic approach toward the combination of factors is described. Various approaches toward human risk characterisation with respect to the application of assessment factors have been summarised and evaluated. The two methods distinguished comprise: 1. Multiplication of assessment factors (traditional approach using default factors of 10); 2. Probabilistic multiplication of distributions of database-derived assessment factors. In these methods correlations between factors are not taken into account. From the analysis of the various assessment factor approaches the following can be concluded: 1. For interspecies extrapolation, allometric scaling on the basis of caloric demands (the 0.75 power of body weight) is considered preferable above scaling on body weight. 2. The traditional extrapolation approach, based on more or less arbitrarily chosen factors of 10, is simple to apply but obscures the relative contributions of scientific arguments and policy judgements. Inflexible and default assessment factors may result in limited utilisation of existing knowledge (Beck et al., 1993). The other default approach, as well as the application of a toxicity profile derived factor (Hakkert et al., 1996), make better use of the data available and therefore are more directed towards a scientific estimation of safe exposure levels for humans. 3. The worst case character of the traditional default assessment factors is doubtful in view of the data analysed. The P95-values of the proposed distributions for the interspecies factor and the semi-chronic to chronic timescale factor are considerably higher than 10 and the limited data on intraspecies variation also indicate that a default factor of 10 may not be sufficient. However, it is also noted that the real distributions may be less wide than observed on the basis of historical data. 152


4. The derivation of approximations of the distribution of assessment factors from historical data, i.e. on the basis of NOAEL-ratios, has limitations. NOAEL-ratios are assessed without always knowing the quality of the underlying data. Furthermore, it should be recognised that the use of the NOAEL instead of the NAELtrue brings along the variation (error) in the NOAELs. The NOAELs are only rough estimates of the true NAEL. It is noted for example that true risks at the NOAEL may vary from 1 to over 10% (US EPA, 1995; Leisenring and Ryan, 1992). Therefore, the geometric standard deviations of the NOAEL-ratios assessed in the studies will overestimate the variation among the ratios of the CEDs. Unfortunately, it is problematic to quantify the measurement error of a NOAEL and to correct for this (Slob and Pieters, 1998). 5. The application of assessment factors derived from currently estimated distributions of assessment factors (example 3) may lead to very wide distributions of the overall assessment factor unless chemical-specific data can be introduced (Annex I). This large variation can be expected, but will also arise from the conservatism in the method for the derivation of the assessment factors used (see point 2 above). 6. Probabilistic multiplication of distributions of assessment factors (example 3) is preferred above simple multiplication of percentiles to avoid extreme conservatism without indication how conservative it may be. The result of the probabilistic combination of distributions is in the form of an assessment distribution, so that the degree of conservatism is quantifiable in any particular assessment. As a matter of fact, this approach allows for deriving a HLV as a function of an a priori chosen degree of conservatism. In addition, the approach allows for estimating the lower and upper bounds for possible health effects in the sensitive population at a given exposure level. 7. A prerequisite for the application of method3 is that consensus is reached on default distributions for the assessment factors. 6.6.3 The Benchmark dose concept Furthermore, the framework proposed by Slob and Pieters (1998) is described and may be considered as the ‘ideal’ approach for deriving exposure limits and for quantification of the risk of exceeding these limits. This method is considered a complete probabilistic approach, offering the possibilities of comparing the various uncertainties involved in typical risk assessment, including the uncertainty in the exposure estimate, the uncertainty in the toxicological starting point (Benchmark dose concept), and the uncertainty in assessment factors.

153

6


Chapter 6


The Benchmark-dose concept takes into account the information on the dose-response and the uncertainties in the estimation of the ‘true’ experimental threshold in the animal, depending on the quality of the particular study from which the data are used. The method presented also allows for estimating the lower and upper bounds for possible health effects in the sensitive population at a given exposure level. The use of assessment factors for LOAEL-NOAEL extrapolation and for the ‘steepness’ of a dose-effect curve is completely redundant applying the Benchmark-dose concept. Additionally to the uncertainties in deriving distributions of assessment factors as described above, this method has the following drawbacks: 1. Consensus needs to be reached on the definition of Critical Effect Sizes for all toxicological endpoints that may be relevant. Current toxicological and biological knowledge does not provide sufficient basis to unequivocally establish the breaking point between non-adverse and adverse effect sizes for most endpoints. 2. The method is less straightforward than the NOAEL-approach and requires some statistical experience in fitting mathematical dose-response models to data. 3. The data available in many cases will not allow modelling, since they were not generated with the intention of dose-response modelling. A study design with more, but smaller dose groups may be helpful here. 6.6.4 The way forward Replacing the traditional application of default assessment factors of 10 (example 1) by more database-derived defaults and distributions of assessment factors should be considered. Probabilistic multiplication of distributions of these factors (example 2) is then preferred to avoid extreme conservatism. Replacing the current methods by this framework in future should be considered: • Determination of the value of default assessment factors: The value of various assessment factors as well as its combination by multiplication is a matter of current debate. Consensus needs to be reached on the distributions of assessment factors for, for example, inter- and intraspecies extrapolation and duration of exposure. • Probabilistic approach towards the combination of assessment factors: If consensus can be reached on these distributions, they may be combined by a probabilistic approach. This can be considered as a first step for the implementation of a complete probabilistic method of risk assessment, but more work is needed to fully override the traditional method: • Determination of the CED: Parallel to the determination of the NOAEL, experience can be gained in defining CESs and in modelling the dose-response data to derive CEDs and to establish distributions of EFs on the basis of CED-ratios. 154


•

•

Probabilistic approach towards the determination of the HLV: Subsequently, experience can be gained in arriving at the distribution of a particular CED and in the probabilistic derivation of the HLV. Probabilistic approach towards risk characterisation and risk management: Risk assessors and risk managers continuously need to examine the pros and cons of each of the above stages and their implications for decision making.

In this way further research in this area will benefit from the experience gained. At the same time both risk managers and the public will have to adapt themselves to the new type of judgements they face: rather than relying on one point estimate for the HLV, a decision has to be made on the acceptable degree of confidence in the estimation of the HLV as well as in the estimation of exposure. An important aim should be to avoid overconservative estimates which may lead to high costs for risk reduction measures. 6.6.5 Recommendations for further research The approach presented still contains a lot of uncertainties. Directing future research toward the elucidation of the major aspects of these uncertainties along the following lines is recommended: 1. Interspecies extrapolation: additional data analysis is needed to investigate the variability in the available allometric scaling methods to account for pharmacokinetic differences. This needs a comprehensive comparison of available NOAELs or preferably CEDs of different species for different endpoints. Research is also needed to gain insight into pharmacodynamic extrapolation. 2. Intraspecies extrapolation: research into human variability in sensitivity to chemical substances so far is very limited both with regard to pharmacokinetic and to pharmacodynamic variation. Data analysis should be continued towards the derivation of the distribution of these factors. 3. The assessment factor adjusting for differences in timescale between the experimental result and the exposure scenario considered does not differentiate among substances, for example with respect to mechanism of toxicity or effect. The evaluation of databases with respect to the influence of study design, the interpretation of tests with respect to the derivation of the NOAELs or preferably CEDs and the correction for interspecies variation is recommended. The inhalatory and dermal routes also have not been sufficiently addressed and this research should be extended. 4. The probabilistic approach towards the derivation of the Critical Effect Dose has the drawback that consensus needs to be reached on the definition of Critical Effect Sizes for all toxicological endpoints that may be relevant. Research in this area should address the adversity of changes in effect parameters observed in experimental animals in order to be able to define the Critical Effect Size. 155

6


Chapter 6


6.7 Summary The general goal of this discussion paper is to contribute towards further harmonisation of the human health risk assessment. Although much of the contents of this article is applicable to the human health risk assessment of chemical substances in general, it concentrates on the assessment of new and existing substances within the scope of European Union legislation. More specifically, it intends to be a contribution towards the development of a formal, harmonised set of assessment factors to be applied within the scope of the EU risk assessment for new and existing substances. MOS versus NOAEL In the European Union, Directive 92/32/EC and EC Council Regulation (EC) 793/93 required the risk assessment of new and existing substances in accordance with a detailed package of Technical Guidance Documents (TGD). In human risk assessment an attempt is made to identify the hazards of the substances and to relate them to exposure. For those substances for which a threshold for toxicity is assumed to exist, a No-Observed-AdverseEffect Level (NOAEL) has to be derived or, if this is not possible, a Lowest-ObservedAdverse-Effect Level (LOAEL). The risk is characterised by comparing estimated (or measured) concentrations in air or on skin or total daily intakes to the results of the effects assessment. This Margin of Safety (MOS) is determined separately for each population potentially exposed and for each effect. Another widely used approach is the explicit derivation of a Human Limit Value (HLV11) from experimental or epidemiological toxicity data by dividing the NOAEL or LOAEL by an overall assessment factor. This overall assessment factor is a multiple of several factors, accounting among others for inter- and intraspecies variations, differences in exposure time scales, the nature of the adverse effects and the adequacy of the database. In this report the Margin of Safety approach of the 1996 TGD is compared to the assessment factor approach. It is noted that the 1996 TGD does not provide any quantitative guidance on the size of the Margin of Safety and concluded that decision criteria for human risk characterisation need to be made explicit. This could be achieved by establishing a formal, harmonised set of default assessment factors accompanied by elaborate guidance. The default set should only be applied in the absence of data which permit a more substance-specific, scientific choice. The default set should allow for differentiation with regard to exposure scenarios, including groups at risk, and the toxicological database. Harmonisation should not hamper further developments, i.e. should not be seen as standardisation for everything and for ever.

11

HLV: general term covering various limit values such as the ADI, RfD, PNAEL and HBORV (see section 6.2.2.)

156


Assessment factors This report gives an overview of published extrapolation methods based on the assessment factor approach. It discusses the status quo with regard to the type of factors to be identified, the range of values assigned as well as the presence or absence of a scientific basis for these values. It is shown that all methods use experts for judgement of the underlying toxicological database and the severity of the effects. Few approaches are based on scientific data, but most methods basically rely on the default 100-fold factor used to derive the Acceptable Daily Intake (ADI). It is recommended to investigate the probabilistic nature of assessment factors and to try to derive their distribution. An attempt is made to estimate the distribution of several assessment factors from historical data, i.e. NOAEL-ratios. This analysis shows that the alleged worst case character of the traditional default assessment factors is doubtful: the 95-percentiles of the proposed distributions for the interspecies factor and the semi-chronic to chronic factor are considerably higher than 10 and the limited data on intraspecies variation also indicate that a default factor of 10 may not be sufficient. More work is needed to better characterise the distributions of assessment factors. Probabilistic multiplication of these distributions is preferred above simple multiplication to avoid extreme conservatism without indication how conservative it may be. The Benchmark dose concept The NOAEL selected from the toxicological database may be a poor substitute for the unknown, true NAEL. New developments are presented with regard to the estimation of a NAEL. The already widely discussed Benchmark Dose concept can be extended to obtain an uncertainty distribution of the Critical Effect Dose (CED). This CED-distribution can be combined with estimated uncertainty distributions for assessment factors. In this way the full distribution of the HLV will be derived and not only a point estimate, whereas information on dose-response relations is taken account of. The method potentially can reveal the full range of variability and uncertainty in both the HLV and the exposure estimate as well as elicit expert judgment in a transparent way. It thus allows the risk manager to make a quantitative cost-benefit analysis. However, consensus on the definition of the Critical Effect Size is needed and this requires further research. The method also requires a certain statistical experience to fit mathematical dose-response models and in many cases the data available will not allow modelling, since current protocols are not intended to generate dose-response curves. The various methods discussed are applied to an example substance to show the technical procedures involved.

157

6


Chapter 6


The way forward Replacing the traditional application of default assessment factors of 10 by more databasederived defaults and distributions of assessment factors should be considered. Probabilistic multiplication of distributions of these factors is then preferred above simple multiplication to avoid extreme conservatism. This can be considered as a first step for the implementation of a completely probabilistic method of risk assessment but more work is needed to fully override the traditional method. Through this stepwise implementation further research in this area will benefit from the experience gained. At the same time both risk managers and the public will have to adapt themselves to the new type of judgements they face: rather than relying on one point estimate for the HLV, a decision has to be made on the acceptable degree of confidence in the estimation of the HLV as well as in the estimation of exposure.

158


Annex The assessment factor approach for establishment of a toxicity profile derived overall assessment factor versus the traditional approach used in risk assessment. Table: Toxicity profile of active substance Study Inhalation toxicity subacute, rat (28-days) (exposure: 6 hours/day 5 days/week)

100 (mg/m3)

1000 (mg/m3) LOAEL: liver effects: increased weight, clinical chemistry Higher doses: effects on liver and red blood cell parameters Local effects: none

50

500

LOAEL: liver effects: clinical chemistry Higher doses: effects on liver, kidneys and red blood cell parameters

semi-chronic, rat (90days)

10

90

LOAEL: liver effects: increased weight, clinical chemistry and effects on red blood cell parameters Higher doses: effects on liver, kidneys, (among which microscopical changes), urinalysis and red blood cell parameters

chronic, rat (104 wk)

5

30

semi-chronic, dog (90days)

4

40

LOAEL: effects on the liver and kidneys (increased weight, clinical chemistry and urinalysis) Higher doses: effects on liver, kidneys (among which microscopical changes) and red blood cell parameters Tumours: liver tumours (benign) were observed at dose levels of 150 mg.kgbw-1.d-1 and higher dose levels LOAEL: effects on body weight and red blood cell parameters Higher doses: effects on body weight, kidneys and red blood cell parameters

carcinogenicity study

carcinogenicity study

the test substance produced liver tumours (benign and malign) at dose levels of 300 mg.kgbw-1.d-1 and above.

80 200

200 500

40 100 250

100 250

maternal toxicity developmental toxicity (200 mg.kgbw-1.d-1 highest dose) no teratogenic effects were observed maternal toxicity developmental toxicity teratogenic effects (hydrocephaly/encephalocele; 250 mg.kgbw-1.d-1 highest dose)

7 50 ≥250

50 250

Oral toxicity subacute, rat (28-days)

Carcinogenicity carcinogenicity, mouse (104 wk) Teratogenicity gavage, rat gavage, rabbit

Reproduction toxicity oral, rat (2-generation study)

1

NOAEL LOAEL Effects1 mg.kgbw-1.d-1 mg.kgbw-1.d-1

Genotoxicity in vitro, bacteria negative in vitro, mammalian cells negative in vivo, mammalian negative

parental toxicity developmental toxicity reproduction toxicity (250 mg.kgbw-1.d-1 highest dose) conclusion: the test substance has no genotoxic potential

Only the adverse effects that are regarded as most important and those that determine the NOAEL are

mentioned 159

6


Chapter 6


The following factors are used for the establishment of a level for chronic inhalation exposure of the general population on basis of the subacute inhalation study in the rat: 1 Interspecies differences: This assessment factor is normally composed of two factors, one accounting for difference in caloric demands (experimental species factors of 4, 1.4 and 7 are used for rats, dogs and mice, respectively), and a default value of 3 accounting for remaining uncertainty. In the case an inhalation study is used as starting point, no factor for caloric demand is used, because animals breath according to their caloric demand. 2 Intraspecies differences: A default value of 10 is used, which compensates for differences in sensitivity within the general population. 3 Difference in duration of exposure between the experimental conditions and anticipated exposure pattern: The present toxicity profile demonstrates that after oral exposure in the rat, a factor of 5 could be used for extrapolation of results from subacute to semi-chronic exposure (NOAELsubacute of 50 mg.kgbw-1.d-1 versus NOAELsemi-chronic of 10 mg.kgbw-1.d-1) and a factor 2 could be used for the extrapolation of results from semi-chronic to chronic exposure (NOAELsemi-chronic of 10 mg.kgbw-1.d-1 versus NOAELchronic of 5 mg.kgbw-1.d-1). Combination of these factors results in a factor of 10 for the extrapolation of results from subacute to chronic exposure. It is noted that this factor is considerably lower than the default value of 100, which is traditionally used. 4 Critical effect: The critical effect of the present substance does not need compensation for the type of critical effect; therefore an assessment factor 1 is used. 5 Dose-response: In case of the present substance, the available dose-response relationships do not justify compensation for the steepness of shallowness of the curve; therefore a factor of 1 is used. 6 Confidence in the data base: A factor may be used for limitation of the entire toxicological data base. In case of the present substance, there are no indications for such a factor. In accordance with the above mentioned considerations the (overall) assessment factors in the table that follows are applicable to the 28-day inhalation study in the rat. It is assumed here that route-to-route extrapolation on the basis of the chronic oral test is not possible. Aspect 1. interspecies 2. intraspecies 3. duration of exposure 4. critical effect 5. dose response 6. confidence of the data base overall assessment factor

n.i.: not indicated.

160

Assessment factor approach 3 10 10 1 1 1 300

Traditional approach 10 10 100 n.i. n.i n.i 10000


Acknowledgements This investigation has been performed by order of the Ministry of Housing, Spatial Planning and the Environment (VROM), Directorate-General for Environmental Protection, Direction of Chemicals, Safety and Radiation Protection (DGM-SVS) and the Ministry of Social Affairs and Employment (SZW) of The Netherlands.

6

161


Chapter 6


7 Evaluating uncertainties in an integrated approach for chemical risk assessment under REACH: more certain decisions?

Theo Vermeire

Chapter 7


7. Evaluating uncertainties in an integrated approach for chemical risk assessment under REACH: more certain decisions? This final chapter will discuss how methodological approaches for risk and uncertainty assessment can be embedded in an improved integrated risk assessment process. Section 7.1 will first address the decision-support tools for the determination of risk and associated uncertainty described in Chapters 3 to 6. Section 7.2 concentrates on improvements in the risk assessment process based on the frameworks for integrated risk assessment (IRA) and uncertainty management (Walker et al., 2003) introduced in Chapters 1 and 2. Section 7.3 highlights the consequences for decision-making under REACH leading up to the final conclusions and recommendations (Section 7.4).

164

Discussion and conclusions

7.1 Methodology for risk and uncertainty assessment 7.1.1 Risk assessment methodology This thesis concentrates on the risk assessment methodology for industrial chemicals as applied within the current regulatory framework in Europe, REACH. Chapter 3 explained the decision-support tool EUSES used for the risk assessment methodology for industrial chemicals in the EU. Although this tool is based on the Technical Guidance Documents (TGD) operating until mid-2008 under the pre-REACH EU legislation, much of it is still in line with the updated Technical Guidance Documents for REACH (ECHA, 2008a). Further to the REACH-approach described in Chapter 1, the most important changes in EUSES (version 2) needed for its application in the REACH Chemical Safety Assessment (CSA) will be described first. The methodology will be placed in the context of IRA in Section 7.2. A fundamental difference between REACH and the previous legislation on new and existing chemicals is the ‘shift of paradigm’, i.e. the shift of the burden of proof of safety from authorities to industry and consequently the more central role of risk reduction measures in the risk assessment (Bodar et al., 2002). Under the former legislation authorities were obliged to show whether risks were attached to the production and use of industrial chemicals. The estimation of releases and exposures usually did not clearly include risk reduction options except for obvious ones such as the presence of a sewage treatment plant in environmental risk assessment or local exhaust ventilation in the risk assessment for workers. If the risk assessment showed a risk, and could not be improved further after extensive consultation with stakeholders, authorities had to propose a strategy for risk reduction. Under REACH, industry has to provide sufficient evidence that substances can be used safely for humans and the environment in all stages of its life cycle. Industry therefore has to prepare the risk assessment, taking into account all risk management measures envisaged, and document this in the CSA. In principle the CSA should always show risk characterization ratios below 1, i.e. safe use. The conditions of use of the chemical, including the risk reduction and waste management options and release rates, are described in the Exposure Scenario (ES). EUSES 2 does not have an ES module in which all ‘conditions of use’ parameters are shown together, though the environmental release estimation module has some features of it. The conclusion is that a next version of EUSES should show a clear separation between an ES-module and an exposure assessment module. The ES-module should describe the conditions of use of a substance including risk reduction options and efficiencies and waste management options and efficiencies. The ES-module should provide the input for the exposure assessment module. The release estimation tool in EUSES can be a supporting element of the environmental ES-module,

165

7


Chapter 7


providing a much needed expert system for the estimation of release rates as part of a branch-specific ES-library. The current release estimation tool needs updating: many EU and OECD Emission Scenario Documents have been implemented in the current version as well as an expert system with default emission rates. However, there is no clear separation between emission factors and emission reduction measures. The EUSES 2 environmental exposure assessment module is in accordance with the new TGD and can be used with input from the ES. The consumer and worker exposure assessments in EUSES2 provide some of the first tier models of the TGD. The exposure assessment module of the next version of EUSES should be improved by incorporating all consumer and worker exposure assessment models described in the REACH TGD. The environmental effects assessment in EUSES 2 is according to the REACH TGD and does not need major changes. In 2005, the human effects assessment in the TGD was adapted to a pre-version of the 2008 update. This is described in Chapter 3. The essence of this update is the explicit recommendation for the use of assessment factors to evaluate the Margin of Safety between the NOAEL or other dose descriptors and the estimated exposure for threshold substances and the determination of a lifetime cancer risk for non-threshold substances. The assessment factors should take into account uncertainty and variability in the extrapolation from experimental data to the human situation. The value of the assessment factors is based on an historical analysis and on database analysis (see Chapter 6 and Section 7.1.2). The assessment factors are applied to the appropriate dose-descriptor for each endpoint after correction for differences in bioavailability, route of exposure, respiratory volumes and experimental exposure conditions. Under REACH this has further evolved into the use of the term Derived No-Effect Level (DNEL) for the quotient of the dose descriptor and the overall assessment factor for threshold substances and the term Derived Minimal-Effect Level (DMEL) for the linearised extrapolation for non-threshold substances. EUSES 2 should be updated for the differences in terminology and slight differences in the approach, mainly with regard to the corrections to the appropriate starting point of the dose descriptor for the extrapolations. The next section discusses the past and present approach towards uncertainty in the risk assessment for industrial chemicals in the EU using the methodology described in this thesis. The discussion concentrates on the environmental risk assessment for humans.

166


7.1.2 Uncertainty assessment The probabilistic approach Chapters 4, 5 and 6 together show that a PRA in the EU framework for risk assessment of industrial chemicals is feasible with currently available techniques. From a scientific point of view, there is sufficient argumentation to strive for a probabilistic framework in chemical risk assessment. In a probabilistic assessment one can include all the available information. Incorporation of the possibility to perform sensitivity and uncertainty analysis in risk assessments could provide more quantitative insight into the range of possible outcomes and therefore better inform both risk assessors and risk managers. The example outlined in Chapter 5 shows the possibility of including uncertainty and variability in a typical risk assessment according to the risk assessment methodology explained in Chapters 2 and 3 and the principles for uncertainty analysis discussed in Chapter 4 (option B in Figure 4.2). It includes the uncertainty in the exposure estimate, the uncertainty in the toxicological starting point (Benchmark approach), and the uncertainty and variability in assessment factors. The method establishes the distribution of the Risk Characterisation Ratio (RCR) for exposure of humans to DBP via the environment, given the EU Standard Environment defined in the TGD (ECHA, 2008a) and given assumptions on distributions of chemical-specific parameters (Jager et al., 2001b), distributions of assessment factors (Chapter 6) and the distribution of the Critical Effect Dose (Chapter 6). It is also shown that the degree of conservatism in the deterministic risk assessment can be evaluated by a comparison with the overall probabilistic distribution (see also Box 7.1). Presently, a deterministic RCR below 1 would indicate “no concern” to the risk manager. A risk manager would probably think differently if he knows that the probability of exceeding this value is very likely (see Figure 4.1). It is stressed, however, that the RCR, even a probabilistic one, is not a true risk level, since the relationship between the RCR and the toxicological impact is unknown. In this particular example we estimated the probability that the Total Human Dose will lead to adverse reproductive effects in a single sensitive member of the population selected at random. In the DBP example, sensitivity analysis showed that the overall uncertainty was dominated by the combined assessment factors (80% of the overall uncertainty). The methodology is therefore able to show where the risk assessment can be improved in the most time- and cost-efficient manner and whether it is necessary and achievable to reduce the uncertainty further.

167

7


Chapter 7


Box 7.1 Risk Reduction Recommendations for DBP The final EU DBP Risk Assessment Report (RAR) for humans exposed to the environment (EC, 2003d) came to conclusion ii (i.e. ‘There is at present no need for further information or testing or risk reduction measures beyond those which are being applied already’). In the final Commission Recommendation on risk reduction measures (EC, 2006) it was recommended that authorities should lay down in permits conditions, emission limit values or equivalent parameters or technical measures regarding dibutylphthalate, in order for the installations concerned to operate according to the best available techniques (BAT). This was, however, inspired by a risk reduction recommendation for the protection of plants in the environment in the RAR. In Chapter 5 it is shown that, based on known uncertainties and variability, the point estimate for the total daily dose for humans exposed via the environment generally is at the 20th percentile of the distribution and that the deterministic RCR for humans exposed via the environment due to processing of DBP as a softener alone is between the 60th and 70th percentile of the uncertainty distribution. Conclusion ii therefore seems doubtful.

Uncertainty due to lack of knowledge and variability A number of methodological issues have to be discussed. First of all, as explained in Chapter 1, a clear separation needs to be made between uncertainty due to lack of knowledge and variability to be able to answer different risk questions. In Chapter 5, variability in sensitivity between individuals in the general population and uncertainty due to lack of knowledge are propagated together and the result (Figure 5.6) can be interpreted as an uncertainty distribution for the risk level (here the Risk Characterisation Ratio) of a single sensitive member of the population selected at random. In the example of Chapter 6 (Table 11), the distribution of intraspecies variability is ignored – a presumably worst case extrapolation factor of 10 is used - and the cumulative probability curve (not shown) then would present the uncertainty distribution for the risk levels in the most sensitive individuals of the population. These two types of uncertainty could also be treated separately in a two-dimensional probabilistic analysis which would result in probability curves showing risk levels for different percentiles of the population, together with confidence bounds showing the combined effect of those uncertainties that have been quantified (e.g. Hoffman and Hammonds, 1994; Van der Voet and Slob, 2007; IPCS, 2008b). These examples show that different risk questions determine the methodological approach.

168


Secondly, it should be made very clear which uncertainties due to lack of knowledge and variabilities are included in the assessment and which not. It is noted that categorisation of uncertainties is a problematic issue: for instance, the uncertainty of the interspecies factor (remaining uncertainty after metabolic scaling) and acute to chronic assessment factors is, erroneously, attributed to variability by Verdonck et al. (2007), whereas it is attributed to uncertainty due to lack of knowledge by Van der Voet and Slob (2007). This needs to be looked at further. In the methodological approach presented in Chapter 5, the emphasis is on uncertainty due to lack of knowledge. Variability is only included in the effects assessment where the variability in sensitivity between individuals in the general human population is represented by a general distribution of the intraspecies assessment factor. Variability in environmental and human exposure is excluded by taking a representative environment (EU Standard Environment) and a typical human being (adult, 70 kg with a fixed food and drinking water consumption pattern and inhaling a fixed volume of air per day). This ‘natural variability in time or space’ (Chapter 4) is complicated further by uncertainty in its quantification (Van der Voet and Slob, 2007). As suggested in Chapter 4, one way to take environmental or individual variability into account is by repeating the approach used for different, equally plausible, exposure situations (e.g., a best case, an average case and a worst case scenario), leading to alternative distributions. This approach necessitates a discussion on relevant exposure situations and on a clear-cut separation in chemical-specific and scenario parameters. Human variability in exposure can also be taken into account by incorporating distributions of consumption data and ventilation rates for the populations concerned (Pieters et al., 2004; Bosgra et al., 2005; Van der Voet and Slob, 2007). Another type of uncertainty not included in the probabilistic DBP assessment is model uncertainty, being an example of uncertainty due to lack of knowledge, mainly due to ignorance. In Chapter 4 it is pointed out that the simplification in models, e.g. with regard to the system and the situations to be modelled, constitute a source of uncertainty. Chapter 3 contains a reflection on uncertainty in exposure models in the evaluating of the validation status of EUSES 2. There have been scientific validations largely concentrating on the reliability, accuracy and usefulness of the individual exposure models in EUSES. However, validation activities for individual models are seldom directly applicable to EUSES, since it is a generic instrument, using fixed standard scenarios and averaged-out variations in time and space. Moreover, measured data used for validation are often non-representative for the standard scenarios used. As a consequence, the quantification of uncertainty in models due to lack of knowledge is problematic, if not impossible. Section 7.2.2 will discuss how quantifiable and unquantifiable uncertainties due to lack of knowledge and variability can be approached in risk assessments. 169

7


Chapter 7


Uncertainty and variability in effects assessment Uncertainty and variability in the human effects assessment is addressed by the application of assessment factors to a suitable dose descriptor such as the NOAEL, LOAEL or BMD. For most substances appropriate data are lacking to apply chemical-specific assessment or adjustment factors (CSAFs) such as advocated by IPCS (Renwick, 1993a; Edler et al., 2002; IPCS, 2005). A second choice, the use of data on analogues which act with the same mode of action as the chemical under consideration (ECHA, 2008a), can also rarely be applied. Therefore, one has to rely on default factors. It is concluded in Chapter 6 that the application of traditional default assessment factors has several drawbacks, such as poor scientific justification and unknown conservatism, mixing of science and policy judgments and limited use of existing knowledge. The database derived quantification of such factors for threshold substances and of the uncertainty in the dose descriptor is also discussed in Chapter 6. It is shown how the overall uncertainty in the HLV can be estimated in a fully probabilistic approach. This approach and similar ones have been evaluated (Edler et al., 2002; ECHA, 2008a; Chapter 6) and applied both in the EU (Pieters et al., 2004, Bosgra et al., 2005; Van der Voet and Slob, 2007; Chapters 4 and 5) and in the USA (Baird et al., 1996; Evans et al., 2001). They have only partly been accepted in the regulatory arena. In the REACH Guidance (ECHA, 2008a) the application of a harmonised set of fixed default assessment factors has been proposed which is, to a certain extent, based on the distributions proposed in Chapter 6 and further work summarised in Vermeire et al. (2001b) and Schneider et al (2005). Meanwhile, the literature on database derived distributions of default human assessment factors is growing (in addition to citations in Chapter 6: on intraspecies factor by Hattis et al., 1999; interspecies factor by Walton et al., 2001a and b; Rennen et al., 2001; Kalberlah et al., 2002; Groeneveld et al., 2004; Bokkers and Slob, 2007; time scale factor by Kalberlah et al., 2002; Bokkers and Slob, 2005; oral-to-inhalation factor by Rennen et al., 2004). Dorne and coworkers investigated pathway specific default assessment factors for kinetics (summarised by Falk-Filipsson et al., 2007). However, the traditional approach, using poorly founded assessment factors of 10 for inter- and intraspecies differences and adjustments for time scale and dose-response, is strongly embedded in regulatory toxicology. In the updated TGD, for instance, metabolic scaling (i.e. a factor of 4 for rats) is recommended, but also an ‘additional factor’ of 2.5 for ‘other interspecies differences, i.e. toxicokinetic differences not related to metabolic rate (small part) and toxicodynamic differences (larger part)’. No motivation is given for the size of this additional factor, but it is obvious this factor is inspired by the widely used traditional factor of 10. As shown in Chapter 6 and by others (Baird et al., 1996; Rennen et al., 2001; Bokkers et al., 2007)

170


the worst case character of the traditional factors is doubtful: the 95-percentiles of the proposed distributions for the interspecies factor and the semi-chronic to chronic factor are considerably higher than 10 and the limited data on intraspecies variation also indicate that a default factor of 10 may not always be sufficient. The distributions for assessment factors discussed in this thesis have several shortcomings. The interspecies factor distribution and the time scale factor are based on ratios of NOAELs and therefore also include the error in the NOAEL. The distributions that are estimated from ratios of NOAELs will be unnecessarily wide (Slob and Pieter, 1998; Renwick et al., 2003a). The NOAEL is not the same as the true No-Adverse-Effect Level and, although the NOAEL is still widely used as a conservative estimate of the true threshold dose, the quality (precision) of the estimate cannot be assessed as argued in Chapter 6. Indeed, the NOAEL may still be associated with statistically non-detectable, but perhaps biologically significant, adverse effects (to over 10%, Section 6.6.2). It has been shown that more accurate distributions of assessment factors can be derived based on CED-ratios of the Benchmark approach (Bokkers and Slob, 2005 and 2007). The intraspecies factor distribution used in Chapter 5 and 6 is a theoretical one derived from the traditional factor of 10. Obviously, a distribution based on empirical human data should be the aim. There is a wide genetic diversity in many of the enzymes involved in the inactivation and bio-activation of chemicals, with many of the pathways having bimodal or trimodal distributions indicative of polymorphism. There is concern that the default assessment factors may not provide adequate protection in the case of certain of these polymorphisms (Renwick et al., 2003; Falke-Filipsson et al., 2007). Unfortunately, distributions on intrahuman variability are only available for some selected subpopulations and polymorphisms, mostly for drugs, but not for the general population exposed to industrial chemicals. An interesting approach towards the derivation of distributions of assessment factors is the Bayesian combination of prior knowledge from a default distribution and posterior chemical-specific information as shown for the ecological interspecies assessment factor by Roelofs et al. (2003). In Chapters 5 and 6 it is shown that the Benchmark Dose concept can be extended to obtain an uncertainty distribution of the Critical Effect Dose (CED). This CED-distribution can be combined with estimated uncertainty distributions for assessment factors. The method potentially can reveal the full range of variability and uncertainty in both the HLV and the exposure estimate as well as elicit expert judgment on the size of the Critical Effect in a transparent way. It gives more weight to the expert evaluation of toxicity and dose-response than the traditional NOAEL method, which has major drawbacks as noted above. However, consensus on the definition of the Critical Effect Size (CEZ) is needed and this requires further research. For most endpoints, no general toxicological consensus exists on what

171

7


Chapter 7


effect size may demarcate adverse from non-adverse. The minimal adverse effect size may differ between different endpoints as shown in a study analysing within-animal variation in routinely studied toxicological parameters (Dekkers et al., 2006). Experts have to agree on the CEZ. The Benchmark Dose method also requires a certain statistical experience to fit mathematical dose-response models. In quite some cases the data available will not allow modelling, since current protocols are not intended to generate dose-response curves (e.g. Bosgra et al., 2005). It has been postulated that, in order to overcome the latter, new protocols using more dose levels with less animals per dose group should be developed (Slob et al, 2005). When the BMD methodology is also considered in the design of the study, it may be expected to be more efficient, in the sense of getting more information out of the same number of animals used. Acceptance of probabilistic methods in risk assessment Based on the examples in Chapters 4 to 6 and the above analysis it can be concluded that PRA potentially gives more information to both risk assessors and risk managers, because it gives more quantitative insight into the range of possible outcomes in a risk assessment and the degree of cumulated conservatism in the deterministic risk assessment (see also Moore and Elliott, 1996). In a deterministic assessment the main concern usually is to avoid acceptance of harmful substances: i.e. the errors made decrease chances of detecting false negatives or Type II errors12. This is fine if the point estimate clearly shows insignificant risks, erring on the side of safety is an acceptable policy, one does not worry about costly risk reduction measures or the final decision is based on socio-economic factors (e.g. very high, low benefits versus costs). However, if the policy is to reach an optimum decision on prioritisation, acceptance or rejection of harmful substances and cost-effective risk reduction measures, one would like to know the full distribution of the risk to know the balance between false positives (Type I errors) and false negatives as was pointed out in Chapter 4 (Figure 4.1). PRA can use all information about quantifiable variability and uncertainty in both the exposure and the effects assessment and forces experts to reveal the nature and extent of their judgment on types of uncertainty, distributions, the shape of the dose-response curve and the nature of the critical effect and the CES. Sensitivity analysis is able to reveal the relative impact of uncertainties in parameters on the final result and can reveal where the risk assessment can be improved in the most time- and cost-efficient manner and whether it is necessary and achievable to reduce the uncertainty further. 12 It is noted that the main direction of most errors in experimental animal studies in fact increases the chances of detecting a false negative, e.g., few dose levels, short exposure periods, low genetic variability, low statistical power, use of 5% probability level. This is countered by using high doses (Gee, 2008).

172

 


Apart from several methodological problems in PRA which already have been discussed above, several other barriers with respect to implementation of PRA have to be mentioned. The method clearly is more complicated, time-consuming and costly as compared to calculating point estimates and requires expertise which is not always available. PRA requires more data and needs development of probability distributions. Both risk assessors and risk managers often are unfamiliar with PRA and need to learn new skills. This should first of all be solved by training of both risk assessors and risk managers. In addition, this training should be supported by applying tiered approaches (see Section 7.2.2) and development of guidance (e.g., Vermeire et al., 2001b) and graphical support. Graphs like the cumulative probability curves shown in Chapter 5 (uncertainty and variability compounded) and Figure 7.1 (uncertainty and variability propagated separately) are very helpful in risk communication.   











  

 

   





Figure 7.1: Cumulative probability plot showing the variability in the RCR (middle black curve) as well as the 10, 25, 75 and 90 percentiles of the uncertainty distribution. The light grey area represents the total simulated confidence band and depends on the number of iterations in the uncertainty analysis. The area under the curves left of the vertical line at x=0 shows the probability that the RCR is below 1.

173

7


Chapter 7


If used and communicated properly, PRA can provide more transparency and confidence in risk distributions and insight into real differences between substances with regard to uncertainty and risk. Small scale interviews with regulators and scientists (Chapter 4) revealed that, whereas the scientific community broadly accepts uncertainty analysis as a necessity when presenting model results, risk managers are more hesitant in the expectation of more and time-consuming work and decreased transparency. Frewer et al. (2008) interviewed a substantial number of different end-users of risk assessments and concluded that the regulatory acceptance of PRA assessment strongly depends on the extent to which assessors are able to effectively interpret the reliability and utility of the outputs. Individual understanding of the methodology per se appeared less important. Therefore, effective communication strategies should be developed. This is strongly supported, given also the conclusion above that not all uncertainties can be clearly quantified. In Section 7.2 it will be argued how this strategy could be shaped.

7.2 Improvements in the process 7.2.1 Integrated risk assessment The framework of Integrated Risk Assessment (IRA) was developed to improve the quality and efficiency of the assessment of risks of adverse effects on human health and the environment from chemicals, physical factors, and other environmental stressors and to provide more complete and coherent inputs to the decision making process (Suter et al., 2003; Suter et al., 2005; Vermeire et al, 2007). The underlying thought is that both the scientific discussion and the regulatory responses can benefit from a more integrated, interdisciplinary approach leading to sharing of information, decreased uncertainties and fully informed decisions. Potential benefits The principles of the IRA framework as well as the potential benefits were explained in Chapter 1. In Chapter 2 the framework was applied to the assessment of the risks of organophosphorous (OP) pesticides to a farming community. This case study was developed to communicate the IRA approach, to illustrate how IRA assessments might be conducted, and to highlight the benefits of integration. This case study, as well as three others on POPs, organotins and UV (Suter, 2003), is illustrative only: it describes how integration might be accomplished over the entire risk assessment process. The OP case study particularly shows the following characteristics:

174


1. IRA prevents conclusions based on incomplete assessments; 2. The database is expanded, including both human and environmental data; 3. Results can be expressed in a coherent way across species, exposure, adverse effects, dose-response and eventually risks; 4. Communalities (and differences) in sources and emissions, distribution routes, fate, exposure scenarios, and adverse effects are efficiently used; 5. In the risk characterization, a common set of evidence, criteria, and interpretation of those criteria are used to determine the cause of human and ecological effects; 6. The results of health and ecological risk assessments are presented in a common format that facilitates comparison of results and allows more effective risk management and communication. This study and others (Munns et al., 2003; Di Guilio and Benson, 2002; Bridges, 2003; SCC, 2003) together suggest substantial benefits to be gained by the integration of ecological and human health risk assessments. At the same time, it also appears to be difficult to present really hard evidence for some of these benefits (see below). The most important of the perceived benefits are increased assessment efficiency with regard to data collection and methodology, increased cost effectiveness of assessment activities in view of shared resources, increased predictive and diagnostic capability on toxicokinetics and toxicodynamics and increased coherence of assessment results in view of shared methodology and risk characterization. With regard to data collection and methodology it can, for instance, be argued that assessment uncertainty will decrease, or at least be better characterised, by confirmation of mechanisms of action and by increased knowledge in weighing of multiple lines of evidence and read-across from conventional mammalian toxicology, ecotoxicology, human epidemiology, and eco-epidemiology, supported by alternative methods such as (Quantitative) Structure-Activity Relationships and in vitro data. Thus, integration could also help to meet the societal and political pressure to avoid vertebrate testing where possible. Further work is needed to investigate when and how the total Weight-of Evidence (WoE) from such integration will show a positive balance. It may also answer the question how alternative methods add to or detract from the total uncertainty in the effects assessment.

175

7


Chapter 7


As pointed out in Chapter 1, the problem formulation stage is particularly relevant for both information sharing and defining the needs of a fully informed decision that addresses the concerns of all risk managers and stakeholders13 (Suter et al., 2005). Problem formulation should define the goals, objectives, and scope of the risk assessment, including the structure and extent of the system to be assessed. Problem formulation also identifies the hypotheses to be tested and the assessment endpoints likely to be of interest, postulate a conceptual model that describes the relationship between the assessment endpoint(s) and stressor(s), and lay out a rational analysis plan for the risk assessment. The IRA framework does not prescribe when and how the risk assessors, risk managers and other stakeholders interact, but encourages interactions appropriate to the legal and social context. It can be expected that interactions between these groups will be most pertinent at the problem formulation stage. However, risk assessors with different interests may also want to interact at the analysis stage, for instance to get more data. At all times, the processes of risk assessment and risk management remain separate, but in close interaction with each other. Weaknesses Some weaknesses have to be pointed out as well, not so much in the approach per se, but rather in the demonstration of its benefits and in organisational backing (Vermeire et al., 2007):  Although several cases have been studied to demonstrate the benefits of IRA, it proves to be difficult to demonstrate convincingly that this approach will be efficient and cost effective. The benefits found are based on inferences from historical evidence and a number of descriptive case studies, but not on cost-benefit studies.  These case studies also revealed that an increased quality of the assessment seems likely, but hard to prove.  Although many regulations call for protection of both human health and the environment, scientifically and institutionally these areas often have developed independently.  The knowledge on shared modes and mechanisms of action needs to be expanded. It is clear that further demonstrations of the scientific and regulatory benefits of the IRA approach are needed, including its cost effectiveness. It can also be noted that the systematic analysis of uncertainties and its contribution to risk characterisation and decision making is not very prominent in the IRA-framework. This, therefore, needs more attention. 13 It is useful to repeat a few definitions of Chapter 1 here: risk assessors are experts providing decision-support through performing risk assessments. They can be affiliated to the public sector, industry or other stakeholders. Risk managers are considered to be regulatory decision- and policymakers on chemicals. If ‘risk manager’ is used more broadly, also including decision-makers in industry, this will be made clear. Stakeholders are members of the society outside the government such as representatives of industry, downstream users, non-governmental organisations, public interest groups, and private citizens.

176


IRA in the EU The case study of Chapter 2 was built on exposure assessment and risk characterization methodology developed and applied in the European Union (EU). In the European Union, some regulatory frameworks, notably that on industrial chemicals and biocides, already require a partly integrated risk assessment (Vermeire et al., 2007). As was shown in Chapter 1, this requirement also extends to the new EU chemicals policy as developed in REACH, which is in line with the European Environment and Health Strategy (EC, 2003b). The European Environment and Health Strategy calls for an integrated approach with regard to information, research, environmental and health concerns, and understanding of the cycle of pollutants, intervention and stakeholders. The strategy aims, among others, for “linking … environmental, health and research information to enable an integrated approach showing the cycle of a pollutant, assessing global exposure and associated health effects and identifying the most productive action routes.” This seems to give ample opportunity towards integrated approaches as defined in this thesis. Risk assessment under IRA and REACH If the IRA framework can be considered to be a gold standard for risk assessments, how does the REACH risk assessment framework compare to it? Both frameworks define the role of science and political and societal values as essential parts of the process. Both frameworks show risk management and stakeholder activities as parallel and concurrent activities that may interact in various ways depending on the legal context and the assessment problem. In IRA, problem formulation, including planning and scoping of the assessment, is an important stage for such interactions with considerations of regulatory requirements and constraints, societal values, stakeholder participation and risk communication. Under REACH, part of the problem formulation is addressed, harmonized and laid down in extensive guidance documents (ECHA, 2008a) thus promoting, among others, coherence in terminology, definitions, assessment questions and endpoints, specification of uses and conditions of use (in the Exposure Scenario), conceptual models, use of exposure and effect models, uncertainty analysis, socio-economic analysis and risk communication. Stakeholders participated in the development of such guidance. REACH further specifies formal committees for interactions at regulatory level (REACH Article 133 Committee), Member State level (Member States Committee), risk assessor level (Risk Assessment Committee), risk assessor and risk manager level (Socio-Economic Analysis Committee) and enforcement authorities (Forum for exchange of information). Stakeholders may be invited to the meetings of the Committees as observers. However, apart from these opportunities for communication, it is clear that the initiative for at least part of this process under REACH has shifted from the regulator to the industry supply chain. Chemical Safety Assessments will be prepared by registrants within the boundaries

177

7


Chapter 7


of REACH and the REACH guidance available. Intensive interactions at risk management and risk assessor level across the supply chain are envisaged to be able to integrate all information available, especially on sources, uses and stressor characteristics and to share analysis plans. It is strongly recommended that industry will not only interact in the supply chain, but also with regulatory bodies and other stakeholders at this stage, since the CSA will be the foundation for the evaluation, authorisation and restriction process carried out by regulators and regulatory scientists. The need for such interactions should be proportional to the degree of complexity of the CSA. An impetus towards increased integration of human health and environmental risk assessment under REACH is the requirement for the chemicals industry to demonstrate that the manufacture, use and disposal of chemicals are safe to humans and the environment. In the assessment, a central position is taken by the so-called exposure scenarios, defined as the set of conditions that describe how the substance is manufactured or used during its life cycle and how the manufacturer controls, or recommends downstream users to control, exposure to humans and the environment. Industry will be required to develop exposure scenarios for workers and consumers, exposed directly and via the environment, as well as exposure scenarios for ecosystems. These scenarios should consider manufacture, use, service life and waste disposal. In these exposure scenarios, risk management measures to be taken should be based on an integrated assessment of the risks, e.g. abatement techniques recommended for the protection of workers should not lead to unacceptable risks for the environment and vice versa. Another important element of integration between the human and ecological risk assessment under REACH is exposure via the environment of humans, predatory mammals and birds. This element was already operational under the former EU-legislation for industrial chemicals. The characterization of this risk is based on comparison of the measured or modelled intakes of humans and predators to Derived No-Effect Levels which are, in the absence of specific data, often extrapolated from experimental mammalian toxicity data (such as a No-Observed-Adverse Effect Level for rats). Both the risk assessment for predators and for humans exposed via the environment may be refined using toxicokinetic data, if available. As discussed in Chapter 2, IRA may facilitate the uncertainty analysis and the communication of its results, among others through a common concept, terminology and analytical approach. The IRA framework does not include guidance for the systematic analysis of uncertainties and its contribution to risk characterisation and decision making

178


The REACH TGD does advocate a common approach to the uncertainty analysis and gives guidance for uncertainty analysis as well as general recommendations for communicating uncertainty (ECHA, 2008b), which will be discussed in the next section. In addition, it was pointed out above that uncertainty in the effects assessment may decrease by confirmation of mechanisms of action and increased knowledge gained by the weighing of multiple lines of evidence and read-across across species and chemicals. Thus, IRA stimulates the development of integrated testing strategies and can stimulate achieving one of the goals of REACH, i.e. to reduce the number of animals used for testing (Briggs 2003; Bradbury et al. 2004; Höfer et al. 2004; Vermeire et al., 2007). A report of the EU Joint Research Centre has calculated that for REACH the cost and animal saving potential of Integrated Testing Strategies over a period of 11 years can be around 1 billion Euros and 1.5 million animals (van der Jagt et al. 2004). Several ongoing or planned research activities with specific integrated risk assessment topics can be identified. In addition to the European Strategy described above, several research projects have become part of the EU research programs (e.g., http://www.credocluster.info/; http://osiris.ufz.de). In conclusion: many elements of the IRA framework can be found in the REACH risk assessment approach:  Both frameworks define the roles of decision-makers, stakeholders and risk assessors as parallel and concurrent;  Problem formulation is an important element in both frameworks, though under REACH this is more pre-defined in guidance documents and formal procedures. It remains to be seen whether the interactions between risk managers and risk assessors at industry and regulatory level combined will take shape;  The REACH guidance advocates, more strongly than IRA, a common approach towards uncertainty analysis and the communication of its results;  The REACH concept of Exposure Scenario is a potential driver for further integration of the human and environmental risk assessment;  Exposure of humans and predators via the environment is a distinct IRA-element in the REACH risk assessment methodology;  An important aim of REACH, the reduction of the use of experimental animals, is an important driver towards integrated approaches using knowledge from in vivo, in vitro, in silico and exposure data and on mechanisms of action. As risk assessment is becoming more mechanistic and molecular there may be new opportunities to create an integrated approach based on common mechanisms and a common systems-biological approach.

179

7


Chapter 7


7.2.2 Uncertainty management So far, the theoretical background of the risk assessment of chemicals under the REACH Regulation has been explained as well as important decision-support tools used for the determination of risk and its uncertainty. The REACH assessment approach has been tested by comparing it with the IRA framework. Both advocate an important role for problem formulation, including in the analysis plan a common approach towards uncertainty analysis and communication of its results. In this section the role of uncertainty analysis in the risk assessment process will be further investigated. Uncertainty management in IRA In Chapter 1 the uncertainty management scheme concept of Walker et al. (2003), further elaborated into a guidance document by Van der Sluijs et al. (2003 and 2005), was explained and related to the IRA framework. The REACH TGD section on uncertainty analysis (ECHA,2008b) and guidance documents on uncertainty analysis of IPCS (IPCS, 2008b) and EFSA (2006) all refer to this guidance. The uncertainty management scheme can be regarded as an additional layer in the IRA framework aiming to improve the management and communication of uncertainty in decision-making processes as expressed in Figure 7.2. As noted in US NRC (1996): ‘Perhaps the most important aspect is not the probability number, but the evidence and reasoning it summarizes’. The problem formulation stage, for instance, is strengthened by an explicit analysis of the implications of the problem type and structure for uncertainty assessment, the development of an initial ranking of main socio-political and institutional uncertainties and uncertainty identification and prioritisation. In this stage risk communication strategies can already be prepared by experts, risk managers and stakeholders. Uncertainty analysis is explicitly added to the analysis stage and reviewing; evaluation and reporting of the uncertainty assessment to the risk characterisation stage. The other elements of the uncertainty management scheme are already included in IRA. REACH is based on the principle that it is for manufacturers, importers and downstream users to ensure that they manufacture, place on the market or use such substances that do not adversely affect human health or the environment. As explained in Chapter 1, the Precautionary Principle (PP) is guiding. Precautionary measures, proportional to the desired level of protection, can only be warranted in the event of reasonable grounds for concern of a potential risk that cannot, or not in time, be determined with sufficient certainty. The precautionary principle should be considered within a risk assessment framework, including a scientific evaluation, as complete as possible, with the identification of the degree of uncertainty at each stage (EC, 2000). The enhanced IRA-framework (IRA+) can serve this purpose.

180

                                              







  

  

  

  

  

 

 



 



                                             




 

               

7



Figure 7.2: The enhanced IRA scheme (IRA+) 181


Chapter 7


In the risk characterization part of the REACH Chemical Safety Assessment, it is therefore recommended to list all quantifiable and non-quantifiable uncertainties and to evaluate their effects on the outcome of the assessment. Chapter 4 and this section give further background on the importance of such an analysis, its implications for the risk assessment and its complications. This analysis of uncertainties can be carried out as recommended in the uncertainty management concept (Walker et al., 2003; Van der Sluijs et al., 2003 and 2005). Tiered approach A tiered approach is advocated. Tiers proposed range from a worst case or conservative assessment, through one or more refined deterministic assessments to fully probabilistic assessments. (EFSA, 2006; Verdonck et al., 2007; IPCS, 2008b; ECHA, 2008b). At all stages, additional information received, e.g. monitoring data, dose-response data, information on uses, can improve the assessment and reduce, or sometimes increase, the uncertainty. In the first tier, the conservative approach, the uncertainties are treated intrinsically by using worst case exposure situations, assumptions and default values. One should realise that it is not always easy to analyse whether these exposure situations are really worst case. If it is not obvious from this analysis that risks are adequately controlled14, the deterministic risk assessment can be refined in subsequent tier 2 steps with an increasingly thorough analysis of uncertainties. The uncertainty assessment at the first refinement stage should list and classify all known quantifiable and non-quantifiable uncertainties as well as the influence a specific entry has on the risk and the effect of all uncertainties combined. The uncertainty matrix (Walker et al., 2003; Van der Sluijs et al., 2003) and the source listing of the REACH Guidance (ECHA, 2008b) are very useful here as also noted by Verdonck et al., 2007. The third tier, the probabilistic approach outlined in Chapters 5 and 6 concentrates on the assessment of the quantifiable uncertainties. From the above, it can be concluded that such an approach needs to be embedded in an IRA approach (such as outlined in Chapter 2 for OP-esters), which should include a clear uncertainty management component (Figure 7.2). In this way also non-quantifiable uncertainties can be revealed and taken into account. The Problem Formulation stage of the assessment of DBP under REACH could be envisaged to include a qualitative analysis and prioritisation of the uncertainties as outlined in the 14 Under REACH the Chemical Safety Assessment includes risk reduction measures to control exposure. The risk characterization therefore is not an indication of the true risk, but an indication of safety (the risk after application of the risk reduction measures proposed).

182


Annex to this chapter for the DBP-example of Chapter 5. Each source of uncertainty should be analysed separately for its influence on the outcome of the risk assessment. This can be done qualitatively, deterministically or probabilistically. However, as noted, it will never be possible to quantify all uncertainties. In the risk characterisation, the combined effect of all identified uncertainties should therefore be evaluated carefully. Classification of uncertainties The retrospective, qualitative uncertainty analysis in Tables A.1 and A.2 shows that a large number of sources of uncertainty can be made explicit, classified and qualified. Many of the uncertainties in this example are classified as statistical or scenario uncertainty which are, in principle, quantifiable. In Chapter 5, the uncertainty analysis is performed on the basis of the ‘statistical uncertainties’. Variability can be described by frequency-based functions; uncertainties due to lack of knowledge can partly be described by expert-based probability functions. However, a good number of the sources of uncertainty are not, or partly not, quantifiable: they belong to the domain of recognized ignorance. The relative importance of each of the uncertainties, indicated by the codes A (of crucial importance), B (important) and C (of medium importance) is a subjective choice, based on the relative impact the source might have on the distribution of risk levels if this source would be quantifiable. Sensitivity analysis may help here for the statistical uncertainties and scenario analysis for the scenario uncertainties. From Chapter 5 we know that the quantifiable uncertainties are highest in the inter- and intraspecies extrapolation. The scoring in the last two columns for the ‘qualification of the knowledge base’ and the ‘value-ladenness of choices’ are again subjective choices. Methods exist to approach this table input more systematically (pedigree scoring and value mapping, Van der Sluijs, 2003 and 2005). If this qualitative analysis were carried out for the first tier deterministic assessment, the ABC scoring could be given an additional coding indicating the direction of the influence of the uncertainty on the risk level, .i.e. overestimation of underestimation due to the choices made (as recommended by EFSA, 2006, and ECHA, 2008b). Weight-of-Evidence and uncertainty In risk assessment, an important tool for characterizing and reducing uncertainty due to lack of knowledge is the Weight-of-Evidence (WoE) process, usually carried out by experts. Lack of knowledge can occur due to inexactness (measurement error), lack of observations, conflicting evidence, or sheer ignorance (Van Asselt, 2000). WoE is an important element of quality assurance procedures in risk assessment, since it deals with crucial uncertainties in the decision-process and the different interpretations of these uncertainties. Examples of expert decisions vary from the selection of a value or distribution for an uncertain parameter (e.g. the water purification factor, the intraspecies factor, the Critical Effect

183

7


Chapter 7


Size in Chapter 5) to the choice of a mode or mechanism of action and endpoint (e.g. the non-genotoxic action of DBP, the choice of the critical study for the selection of the NOAEL, the selection of the NOAEL in Chapter 5) or the characterization of the risk (the interpretation of the RCR-ratio in Chapter 5). Under REACH, a transparent WoE process will be critical in determining the classification of chemicals, particularly as CMR of PBT or vPvB, since these determine important risk management decisions such as the selection of chemicals for a full exposure assessment (all chemicals classified as dangerous15) or for authorisation (CMR/PBT/vPvB) So, in this process, scientists need to assess more specific and technical uncertainties in the underlying data as well as in their own interpretations of these data (Levin et al., 2004). Unfortunately, formal procedures and consistent terminology for WoE-processes are lacking (Levin, 2004; Weed, 2005; Hart et al., 2007; Gee, 2008).

7.3 Decision making The risk management process can be described as one in which decision-makers weigh not only the results of a risk assessment, but also legislative-political, socio-economic ethicalcultural factors and technical feasibility (Hope, 2007; Van Leeuwen, 2007). Recognizing this, this thesis has discussed the important stages in risk assessment according to the IRA+framework. Based on examples of uncertainty analysis in the risk assessment methodology of industrial chemicals in the EU, it was shown that 1) uncertainty analysis should be an essential element in this framework, 2) that integration of human health and ecological risk information may reduce this uncertainty and 3) that uncertainty analysis should already be addressed right at the start of each assessment, at the problem formulation stage. In problem formulation, the implications of the question for the risk assessment and the uncertainty analysis should be identified and discussed with all parties concerned. Given the problem and the data available, which types of uncertainties are relevant? Are uncertainties mostly scientific or are socio-political or institutional dimensions also important such as divergent values of stakeholders or different views on the acceptability of risks and uncertainties? Which uncertainties are known, which ones quantifiable? Which uncertainties will be ignored or are caused by ignorance? Which methods of uncertainty analysis will be used? Many policy-makers expect risk assessors to provide certainties rather than uncertain risk estimates, especially if they regard a risk estimate as the only basis for their decision (Are we doomed or safe? Hope, 2007). That uncertainties exist, however, usually is also well understood by both scientists, policy makers and stakeholders. Nobody will contest the view expressed in the REACH TGD (ECHA, 2008b) that, in order to produce a Chemical Safety Assessment that is robust, reliable and adequate, it is useful to consider the degree 15

Classified as dangerous in accordance with Directive 67/548/EEC on classification and labelling

184


of uncertainty in each part of the assessment. But how will an increased attention for uncertainty in an IRA+-framework, as proposed here, help decision-makers, including risk managers in industry, in choosing a course of action and what will be the practical implications, specifically under REACH? Advantages of IRA+ for decision making under REACH Three advantages of a conceptual framework for integrated risk assessment including the systematic treatment of uncertainty in decision-support can be highlighted: 1. Improved communication among scientists, stakeholders and policymakers; 2. The framework will help in making an inventory of where the most policy-relevant uncertainties are expected and how they can be characterised. It can also uncover the effects of further efforts to reduce or characterise uncertainties in the outcome of the risk assessment (Verdonck et al., 2007). 3. More effective and efficient decisions on testing strategies and on the priorities for the Chemical Safety Assessment based on all data available; Below, each of these 3 advantages will be discussed with regard to REACH. IRA+ and improved communication under REACH Risk management decisions on industrial chemicals are taken at Registration, Evaluation, Authorisation and Restriction. Intensive interactions between decision-makers, experts and stakeholders at all stages of the REACH-process will be needed to discuss, from a very early stage, the information available, potential risks and the uncertainties and differences in values, interests and decision-criteria. As argued earlier, part of the problem formulation process is laid down in extensive guidance, including exposure and risk assessment models, and in formal committees, but needs to be broader in scope. A continued dialogue is needed on improvement of the guidance and to test, review and refine the overall process (Money et al., 2007). In addition, a problem formulation per chemical or group of chemicals is needed, the scope of which needs to be proportional to the potential risks: the higher the potential risk, the stronger and elaborate the process should be. The question rises whether the present mechanisms of interaction envisaged to operate under REACH will be sufficient to ensure a thorough and transparent problem formulation and uncertainty management. REACH aims at intensive interactions among stakeholders in the supply-chain, but no clear procedures are foreseen for interactions between stakeholders, scientists and decisionmakers at the start of any assessment of chemicals to be registered. The interactions at evaluation, authorisation and restriction stages are defined at committee level, but needs to be developed further for more informal interactions. The intensity of such communication, especially at registration, should be proportional to the complexity of the Chemical Safety Assessment, the degree of risk foreseen and the degree of uncertainty due to lack of knowledge. 185

7


Chapter 7


Another difficulty in this process is the fact that detailed scientific information is essential for understanding risks and uncertainties for preparing proper decisions, whereas decisionmakers, stakeholders and members of the public affected by them often are no experts at all in this area. On the one hand this requires a strong IRA-process and high communication skills of experts and decision-makers, and, at the other hand, training. IRA+ and uncertainty analysis under REACH The precautionary approaches in REACH, discussed above and in Chapter 1, require first of all a transparent and thorough evaluation of the potentially adverse effects and, secondly, an understanding of all types of uncertainties and of the influence they have on the outcome of the assessment. A tiered approach can be followed in all REACH stages based on the complexity of the Chemical Safety assessment, the degree of risk foreseen and the degree of uncertainty. As argued above, the IRA+-framework is considered to be particularly suited to identify as early as possible, all types of uncertainty in the risk assessment - and not only quantifiable ones - and to discuss how to analyse, prioritise, quantify and communicate them in a dialogue between risk assessors, risk managers and stakeholders. This will prepare the ground for the more expert-like stage of analysis and for a risk characterisation which will lead to decision-support with less surprise, communication problems and distrust between the various parties involved. Decision-makers, also in industry, will particularly be interested to know what ‘lack of knowledge’ exists, to what extent, in which elements, which options are available to resolve or reduce it, and at what cost-benefit ratio. They also need an understanding of the probability that the assessment is leading to a false negative or false positive conclusion, depending on the assumptions and decisions made. Thirdly, insight into variability, allows risk managers to show the distribution of the quantifiable risk in a population. The IRA+ framework can facilitate this analysis supported by robust tools for a tiered integrated risk assessment, uncertainty analysis and risk communication. This thesis has focussed on the former two and shows that such tools are available and, though deficiencies still exist, are operational. The acceptance of the analysis of the quantifiable uncertainties with PRA as the latest addition to the toolbox of risk assessment is rather slow, but the tools discussed here, developed over the past 10 years in the USA and in the EU, finally found their way in scientific recommendations (SCC, 2003) and in the most recent REACH Guidance (ECHA, 2008b). Preliminary examples of a thorough discussion of uncertainties and their influence on the final result have also been demonstrated in the former EU existing chemicals program: in the risk assessment report of zinc and zinc compounds important quantifiable and non-quantifiable uncertainties are discussed qualitatively and, with regard to the environmental effects assessment, analysed probabilistically (Bodar et al., 2005). In order for all parties to get more familiar with this type of analysis, the guidance available should 186


be extended by showing real-life applications, prepared in a regulator-industry-stakeholder expert working group. Priorities for reduction of quantifiable uncertainty can be found in the examples discussed. These have shown that, in general, uncertainty in the Chemical Safety Assessments is expected to be particularly high in the emission estimation and risk reduction factors, both elements of the Exposure Scenario, and in the application of default assessment factors. IRA+ and integration of data under REACH IRA+ also emphasizes that decisions should be based on all data available and searching for communalities (Section 7.2.1). First, this will prevent risk reduction or prioritisation decisions which do not protect all human and environmental targets, or even increase the risk to specific targets. Second, this allows decisions on testing strategies to be based on knowledge from in vivo, in vitro, in silico and exposure data on human or ecological receptors. Each test method or data element for a particular endpoint has its own contribution to the overall WoA taking into account their sensitivity, specificity, variability and uncertainty. Based on the overall WoA it can be decided whether further testing is needed. Third, a similar process can be followed for the exposure assessment based on modelling and monitoring data. As pointed out above, this approach can have great potential in reducing the use of experimental animal.

7.4 Conclusions and recommendations The aim of this thesis was to investigate in what way the scientific process of risk assessment can improve decision-making in the light of existing uncertainty, with a focus on REACH. The thesis seeks this improvement in both process and methodology. An improved framework for integrated risk assessment, IRA+, was presented, which includes a strong uncertainty management component. Uncertainty management was further shown to be supported by qualitative and quantitative tools. The quantitative tools were tested using the REACH risk assessment methodology. Based on examples of uncertainty analysis in the risk assessment methodology of industrial chemicals in the EU, it was shown that 1) uncertainty analysis should be an essential element in this framework, 2) that integration of human health and ecological risk information may reduce this uncertainty and 3) that uncertainty analysis should already be addressed right at the start of each assessment, at the problem formulation stage. The following conclusions and recommendations relate to the IRA and IRA+-frameworks, the uncertainty assessment and management component of this framework and the REACH 187

7


Chapter 7


risk assessment methodology. Special emphasis will be given to the decision-support provided in these issues. On IRA: The most important of the perceived benefits of IRA are increased assessment efficiency with regard to data collection and methodology, increased cost effectiveness of assessment activities in view of shared resources, increased predictive and diagnostic capability on toxicokinetics and toxicodynamics and increased coherence of assessment results in view of shared methodology and risk characterization. IRA also emphasizes that decisions should be based on all data available and searching for communalities. This will avoid risk reduction or prioritisation decisions which do not protect all human and environmental targets, or even increase the risk to specific targets. It is found that the IRA framework and the REACH risk assessment approach have many elements in common and reinforce each other. The problem formulation stage is particularly relevant for both information sharing and defining the needs of a fully informed decision that addresses the concerns of all risk managers and stakeholders. In registering chemicals under REACH, it is strongly recommended that industry will also interact with regulatory bodies and other stakeholders in a problem formulation approach, since the Chemical Safety Assessment will be the foundation for the evaluation, authorisation and restriction process carried out by regulators and regulatory scientists. The intensity of such communication, especially at registration, should be proportional to the complexity of the Chemical Safety Assessment, the degree of risk foreseen and the degree of uncertainty due to lack of knowledge. At all times the processes of risk assessment and risk management remain separate, but in close interaction with each other. Further demonstrations of the scientific, economic and regulatory benefits of the IRA approach are needed. It is recommended to perform cost-benefit studies to this regard. The integration of information could also help to develop testing strategies with the aim of avoiding vertebrate testing where possible. An important aim of REACH, the reduction of the use of experimental animals, is an important driver towards integrated approaches using knowledge from in vivo, in vitro, in silico experiments and data on exposure and mechanisms of action across species. As risk assessment is involving more and more mechanistic and molecular information, there may be new opportunities to create an integrated approach based on common mechanisms and a common systems-biological approach. Further work is needed here to investigate when and how the total Weight-of Evidence from such integration will show a positive balance with regard to costs and animals saved. It may also answer the question how alternative methods add to or detract from the total uncertainty in the effects assessment. 188


On IRA+: The IRA+-framework combines the perceived benefits of the IRA-framework and those of the uncertainty management framework. Advantages discussed, with a focus on REACH, are improved communication, detection and characterisation of the most policy-relevant uncertainties, and improved decisions on testing strategies and priorities for the Chemical Safety Assessment. The uncertainty management scheme can be regarded as an additional layer in the IRA framework aiming to improve the management and communication of uncertainty in IRA decision-making processes (Figure 7.2). The problem formulation stage, for instance, is strengthened by an explicit analysis of the implications of the problem type and structure for the assessment of quantifiable and non-quantifiable uncertainties, the development of an initial ranking of main socio-political and institutional uncertainties and uncertainty identification and prioritisation. Uncertainty analysis is explicitly added to the analysis stage and reviewing, evaluation and reporting of the uncertainty assessment to the risk characterisation stage. The IRA+ framework facilitates the analysis and prioritisation of uncertainties, supported by robust tools for a tiered integrated risk assessment. In order for all parties to get more familiar with this type of analysis, the guidance available should be extended by showing real-life applications, prepared in a regulator-industry-stakeholder expert working group. REACH is based on the Precautionary Principle. Precautionary measures, proportional to the desired level of protection, can only be warranted in the event of reasonable grounds for concern of a potential risk that cannot, or not in time, be determined with sufficient certainty. The Precautionary Principle should be considered as a risk management tool supported by a risk assessment framework which includes a scientific evaluation, as complete as possible, and an identification of the degree of uncertainty at each stage. The IRA+-framework can serve this purpose: it supports decision making, stimulates interactions between risk assessors, decision-makers and stakeholders and provides an approach for a tiered uncertainty analysis. On methodology: The risk assessment tool EUSES has been shown to be fit for purpose under REACH with some modifications, specifically with regard to the implementation of the Exposure Scenario concept and more risk reduction options. Limited validation activities show EUSES to be a good compromise between complexity and practicability. More work needs to be done to increase the domain of applicability, in view of the limited use of EUSES outside the domain of the persistent, non-dissociating substances of intermediate lipophilicity.

189

7


Chapter 7


It has been shown that a PRA for humans exposed via the environment in the EU framework for risk assessment of industrial chemicals is feasible with currently available techniques and that the degree of conservatism in the deterministic risk assessment can be evaluated by a comparison with the overall probabilistic distribution. The different stages of uncertainty analysis and PRA would be greatly facilitated if firmly integrated with risk assessment tools like EUSES. The probabilistic tools can be integrated into a tiered uncertainty approach as part of the IRA+ framework, to avoid unnecessary work and costs for many chemicals. If a fully deterministic, conservative REACH Chemical Safety Assessment indicates concern, it is essential to list all quantifiable and non-quantifiable uncertainties in the next tier and to evaluate their effects on the outcome of the assessment. This analysis of uncertainties can be carried out as recommended in the uncertainty management concept discussed. In any uncertainty analysis a clear separation needs to be made between quantifiable and nonquantifiable uncertainties, and, if possible, between uncertainty due to lack of knowledge and variability to be able to answer different risk questions. Secondly, it should be made very clear which uncertainties are included in the assessment and which not. In risk assessment, an important tool for characterizing and reducing uncertainty due to lack of knowledge is the Weight-of-Evidence (WoE) process, usually carried out by experts. In this process, scientists need to assess more specific and technical uncertainties in the underlying data as well as in their own interpretations of these data. Unfortunately, formal procedures and consistent terminology for WoE-processes are lacking. As shown, international efforts are ongoing to improve this situation and it is strongly recommended to support these initiatives. It is concluded that PRA potentially gives better decision-support to risk managers, because it gives more quantitative insight into the range of possible outcomes in a risk assessment and the degree of cumulated conservatism in the deterministic risk assessment. In deterministic assessments, it is impossible to determine where point estimates lie in the range of possibilities and they give a false sense of accuracy. Exemplary is the application of default assessment factors, shown in this thesis to have poor scientific justification and unknown conservatism, mixing of science and policy judgments and limited use of existing knowledge. Even the alleged worst case character of the traditional factors is doubtful. PRA can use all information about quantifiable variability and uncertainty in both the exposure and the effects assessment and forces experts to reveal the nature and extent of their judgment on types of uncertainty, distributions, shape of the dose-response curve and the nature of the critical effect. Sensitivity analysis is able to reveal the relative impact of uncertainties in parameters on the final result, where the risk assessment can be improved 190


in the most time- and cost-efficient manner and whether it is necessary and achievable to reduce the uncertainty further. The examples discussed show a high contribution to the overall quantifiable uncertainty from uncertainty in emission estimation and in the application of default assessment factors. For a further implementation of PRA, it is necessary to build up experience. If the tiered REACH TGD guidance is followed, PRAs on more substances will become available. The PRA methodology discussed in this thesis and elsewhere can be compared to the traditional, lower tier deterministic approach. Research in this area will benefit from the experience gained. Important research areas are both the characterisation of both default and chemicalspecific distributions used for exposure and effects input parameters, methods to address environmental and human variability in the REACH CSA and to address variability separate from uncertainty due to lack of knowledge. Specific attention should be given to research into the intraspecies assessment factor distribution and the influence of route of exposure on default distributions. An important difficulty in the application of this process and methodology is the knowledge gap: both risk assessors and risk managers often are unfamiliar with PRA and need to get familiar with these new approaches and learn new skills. This should first of all be solved by training of both risk assessors and risk managers. It further requires a strong IRA+-process with a tiered approach, graphical support (e.g., Figures 5.6 and 7.1) and high communication skills of both experts and risk managers. The guidance available should be extended by showing real-life applications, preferably prepared in regulator-industrystakeholder expert working groups. Experts and risk managers need to investigate together how to use the results of uncertainty analysis in decision-making and how to communicate these in a transparent way. Final remarks And what will be the answer to the question of this thesis on the title page: do we get more certain decisions by incorporating uncertainty in risk assessment of chemicals? From this work it clearly follows: yes. If process and methodology follow the direction shown, decision-support will be more transparent, will lead to less communication problems and will improve the trust between various parties involved. Decisions which fully take into account the uncertainties in the assessments performed, including the influence of divergent opinions and assumptions of experts and stakeholders, will be better informed and will lead to transparent decisions which can be communicated in a clear way.

191

7



192

technical parameters B24

C3

B13 C16 C20 C21 C22 C23 B29 B30 B31 B32 B33 C34

Model

structure

C3

B19

Scenario uncertainty A1 A2 A4 C5

x

B18

C6

x x x x x x x x x x

A7 C8 C9 C10 B11 B12

x

Recognised Ignorance x

Level of uncertainty Statistical uncertainty

Expert judgement

Context

Location↓

Uncertainty Matrix

x

x x

x x x x x x x x x x x x

x x x x x x x

x

x x x x x

x

x x x x

x x x

x

x

x x x x x x

x

x x x

Nature of Qualification of uncertainty knowledge base Lack of Variability Weak Fair knowledge x x x x x x

x

x

x

x

x

Strong

x x x x x


x

x x x

x

Small

x

x

Medium

x

x x

Large

Value-ladenness/subjectivity of choices

Table A.1: Uncertainty matrix for the risk assessment of DBP (Chapter 5) (The ABC coding refers to the relevance of the specific uncertainty sources: Ax =of crucial importance, Bx = important, Cx = of medium importance; the subscript refers to the sources in Table 7.2)

Annex: Qualification and prioritisation of uncertainties

Chapter 7

Data Outputs C35

B13 A14 A15 B17 B25 B26 B27 B28

Model

inputs

Statistical uncertainty

x A36 A37 A38

Scenario Recognised uncertainty ignorance

Level of uncertainty

Location↓

Uncertainty Matrix

x

x

x x


x x x x

x

Variability Weak

x x

Fair

Qualification of knowledge base

Lack of kowledge

Nature of uncertainty

Table A.1: Uncertainty matrix for the risk assessment of DBP (continued)

x

Strong

x

x x x x x x

x

Small

Medium

Value-ladenness of choices

x x x

x

Large


7

193



194

Exposure assessment 18 Assumption: well-mixed environmental compartments 19 Assumption: continuous, constant emissions 20 Bioconcentration factor fish 21 Bioaccumulation factor meat 22 Bioaccumulation factor milk 23 Tissue-soil Concentration Factor 24 Leaf conductance 25 Purification factor drinking water 26 Respirable fraction

16 Allometric scaling factor 17 Time assessment factor semi-chronic to chronic

15 Interspecies assessment factor

14 Intraspecies assessment factor

8 Assumption: critical endpoint is reproductive toxicity 9 Assumption: data set gives full picture of human toxicity 10 Choice of critical study 11 Choice of critical effect 12 Critical Effect Size 13 Critical Effect Dose

Brief description of sources General 1 Assumption: EU standard environment 2 Assumption: homogeneous EU population 3 Assumptions for type of parameter distributions 4 Simplification to one use scenario for local exposure 5 Conceptual model 6 Model implementation ExcelTM-EUSES 1.00 Hazard assessment 7 Assumption: threshold for toxicity

Boundary condition for exposure models, simplification, common approach, influence ambiguous Boundary condition for exposure models, simplification, ignores unknown variability Model uncertainty described by Jager et al. (1997) and Schwarz (2000), simplification but well validated Model uncertainty described by Jager et al. (1997) and Schwarz (2000), oversimplification Model uncertainty described by Jager et al. (1997) and Schwarz (2000), oversimplification Model uncertainty described by Jager et al. (1997) and Schwarz (2000), oversimplification Input parameter uncertainty described by Jager et al. (1997) and Schwarz (2000) Input parameter uncertainty described by Jager et al. (1997) and Schwarz (2000) Input parameter uncertainty described by Jager et al. (1997)

No great controversy among expert groups, threshold model in general is challenged, choice of other assumption (no threshold) has great influence of risk assessment Assumption backed by EU-RAR (EC, 2003b), other expert groups may disagree, database not ideal Weight of Evidence based on animal experiments; human data are lacking (EC, 2003b) Six animal studies available, EU-RAR considers this study the most appropriate EU-RAR shows uncertainty on developmental toxicity, mechanism of action, neurological effects Controversy on judging degree of adversity (Section 6.4.3), predefined value used, expert choice Variability in dose-response analysis, uncertainty in dose-response data, uncertainty in regression model (maximum likelihood estimate) (Section 6.4 ) Default factor,for variability within total general population, theoretical distribution based on ‘historical’factor of 10 (Section 6.3.2.3) Default factor for uncertainty in extrapolation between humans and experimental animals, derived from database of uncertain NOAELs Modelled scaling factor describing metabolic differences between species (Section 6.3.2.2) Default factor for uncertainty in time extrapolation, derived from database of uncertain NOAELs (Section 6.3.2.4)

High variability in scenario’s, independent of knowledge base, policy choice to ignore this High variability in human population, independent of knowledge base, policy choice to ignore this Influence on mean small, on variance higher (Lessmann et al.,2005), general assumptions made Processing as softener, subjective choice for worst case scenario leading to the highest releases Figure 5.1, state of the art conceptual model, unknown or ignored routes considered less important Meets common quality criteria (Schwartz, 2000)

Explanation of position in Table A.1

Table A.2: Description of sources of uncertainty for DBP case (Chapter 5)

Chapter 7

Brief description of sources 27 Bioavailability via inhalation route 28 Bioavailability via oral route 29 PEC-air 30 PEC-surface water 31 PEC-grassland soil 32 PEC-pore water grassland 33 PEC-pore water agricultural land 34 PEC-groundwater 35 Monitoring data for PECs Risk characterisation 36 Assumption: RCR > 1 means concern 37 If RCR>1 and refinement not possible: conclusion for data requirements or risk reduction 38 Aggregate and cumulative risks of other phthalates Subject of high controversy, database insufficient

Impact (incidence, severity of effects) unknown, policy choice Impact (incidence, severity of effects) unknown, policy choice

Explanation of position in Table A.1 Input parameter uncertainty described by Jager et al. (1997) Input parameter uncertainty described by Jager et al. (1997) Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Model uncertainty described by Jager et al. (2001), compounded uncertainty and variability Described by Jager et al. (2001), data may not be representative

Table A.2: Description of sources of uncertainty for DBP case (continued)


7

195


Chapter 7


Abbreviations

Abbreviations


Abbreviations

ADI AER AF AOEL BCF BMD(L) BMF CED CES ChE CMR COMEST CSA CSAF DBP DDT DMEL DNEL EC ECX ECETOC ECHA EEA EEC EF EINECS EPA ES ESD EUSES FAO FDA GM GSD HLV HBORV HC HEDSET

Acceptable Daily Intake AOEL versus Exposure Ratio Assessment Factor Acceptable Operator Exposure Level Bioconcentration Factor Benchmark Dose (Level) Biomagnification Factor Critical Effect Dose Critical Effect Size cholinesterase Carcinogenic, Mutagenic, Reprotoxic World Commission on the Ethics of Scientific Knowledge and Technology Chemical Safety Assessment Chemical-Specific Adjustment Factor dibutylphthalate dichlorophenyl trichloroethane Derived Minimal-Effect Level Derived No-Effect Level European Commission Effect Concentration for X% of a testpopulation after a specified exposure time European Centre for Ecotoxicology and Toxicology of Chemicals European Chemicals Agency European Environmental Agency European Economic Community Extrapolation Factor European INventory of Existing Commercial chemical Substances Environmental Protection Agency (USA) Exposure Scenario Emission Scenario Document European Union System for the Evaluation of Substances Food and Agricultural Organisation Food and Drug Administration (USA) Geometric Mean Geometric Standard Deviation Human Limit Value Health Based Occupational Reference Value Health Council (of The Netherlands) Harmonised Electronic Data SET (OECD)

199


Abbreviations


ID IHCP IOMC IPCS IRA JECFA JMPR LC50 LLN LOAEL LR MAD MF MOE MOS MRR NAEL NOAEL NOEC NRC OECD OPIDN OPs PBT PCB PEC PNAEL PNEC PP PRA Px QSAR RCR RfD RIVM RMOE RMOS SAICM

200

(Hazard) Identification Institute of Health and Consumer Protection (JRC) Inter-Organization Programme for the Sound Management of Chemicals International Programme on Chemical Safety Integrated Risk Assessment Joint Expert Committee on Food Additives Joint Meeting of Experts on Pesticide Residues Median Lethal Concentration for 50% of test organisms Lewis/Lynch/Nikiforov model Lowest-Observed-Adverse-Effect Level Lifetime cancer Risk Mutual Acceptance of Data Modifying Factor Margin Of Exposure Margin Of Safety MOS Reference-MOS Ratio No-Adverse-Effect level No-Observed-Adverse-Effect Level No-Observed-Effect Concentration National Research Council (USA) Organisation for Economic Co-operation and Development Organophosphate-induced delayed neuropathy Organophosphorous Pesticides Persistent, Bioaccumulative, Toxic polychlorinated biphenyls Predicted Environmental Concentration Predicted No-Adverse-Effect level Predicted No-Effect Concentration Precautionary Principle Probabilistic Risk Assessment xth percentile Quantitative Structure Activity Relationship Risk Characterisation Ratio Reference Dose National Institute of Public Health and the Environment Reference Margin Of Exposure Reference-MOS Strategic Approach to International Chemical Management

Abbreviations

SF SLUD STP TDI TEF TGD TNG TNO TSCA UF UN USES vPvB WHO


Safety Factor salivation, lacrimation, urination and defecation (syndrome) Sewage Treatment Plant Tolerable Daily Intake Toxic Equivalency Factor Technical Guidance Documents Technical Notes for Guidance Netherlands Organisation for Applied Scientific Research Toxic Substances Control Act (USA) Uncertainty Factor United Nations Uniform System for the Evaluation of Substances (The Netherlands) very persistent, very bioaccumulative World Health Organisation

201

Abbreviations


References

References


References

Aldenberg T, Slob W. 1993. Confidence limits for hazardous concentrations based on logistically distributed NOEC toxicity data. Ecotox Environ Safety 25:48-63. Anonymous. 1995. Occupational risk assessment of pesticides. Part I. Guidance for the setting of Acceptable Operator Exposure Levels. Workshop held on April 27 1995, The Hague, The Netherlands. Azaroff LS. 1999. Biomarkers of exposure to organophosphorous insecticides among farmers’ families in rural El Salvador: factors associated with exposure. Environ Res Section A 80: 138147. Baird SJS, Cohen JT, Graham JD, Shlyakhter AI, Evans JS. 1996. Noncancer risk assessment: a probabilistic alternative to current practice. Hum Ecol Risk Assess 2:79-102. Bartell. SM. 1996. Some thoughts concerning quotients risks and decision-making. Hum Ecol Risk Assess. 2:25-29. BAUA. 1998. TRGS 901 Ergänzungen. Kriterien für die Ableitung von gesundheitsbasierten Luftgrenzwerten bei limitierter Datenlage. Bundesarbeitsblatt 10:74-76. Beck BD, Conolly RB, Dourson ML, Guth D, Hattis D, Kimmel C, Lewis, SC. 1993. Improvements in quantitative noncancer risk assessment. Fund Appl Toxicol 20:1-14. Bennett RS. 1989. Role of dietary choices in the ability of bobwhite to discriminate between insecticide-treated and untreated food. Environ Toxicol Chem 8:731-738. Berding V. 2000. Validation of a regional distribution model in environmental risk assessment of substances. Doctoral Thesis, University of Osnabrück, Germany. Bernstein PL. 1996. Against the Gods: the remarkable story of risk. Wiley, New York.,USA. ISBN13: 9780471295631. Bigbee JW, Sharma KV, Gupta JJ, Dupree, JL. 1999. Morphogenic role for acetylcholinesterase in axonal outgrowth during neural development. Environ Health Persp 107 Suppl. 1:81-87. Bigwood EJ. 1973. The acceptable daily intake of food additives. Crit Rev Toxicol, June pp. 41-93. Bodar C, De Bruijn J, Vermeire T, Van der Zandt P. 2002. Trends in risk assessment of chemicals in the European Union. Hum Ecol Risk Assess 8:1825-2002. 205


References


Bodar CWM, Pronk MEJ, Sijm D. 2005. Rge European Union risk assessment of zinc and zinc compounds: the process and the facts. Integr Environ Assess Manage 1:301-319. Bokkers BGH, Slob W. 2005. A comparison of ratio distributions based on the NOAEL and the Benchmark approach for subchronic-to-chronic extrapolation. Toxicol Sci 85:1033–1040. Bokkers BGH, Slob W. 2007. Deriving a data-based interspecies assessment factor using the NOAEL and the Benchmark Dose Approach. Crit Rev Toxicol 37:355–373. Borzelleca JF. 2001. History of toxicology. In: Hayes AW (ed). Principles and methods of toxicology, Fourth edition. Raven Press, New York.,USA. ISBN 1-56032-814-2. Bosgra S, Bos, PMJ, Vermeire TG, Luit RJ, Slob W. 2005. Probabilistic risk characterization: an example with di(2-ethylhexyl)phthalate. Regul Toxicol Pharm 43:104-113. Bradbury SP, Feijtel TCJ., van Leeuwen CJ. 2004. Meeting the scientific needs of ecological risk assessment in a regulatory context. Environ Sci Technol, December 1: 463A – 470A. Bridges J. 2003. Human health and environmental risk assessment: the need for a more harmonised approach. Chemosphere 52:1347-1351. Briggs J. 2003. Human health and environmental risk assessment: the need for a more harmonised approach. Chemosphere 52: 1347-1351. Bremmer HJ, Blom WM, van Hoeven-Arentzen PH, van Raaij MTM, Straetmans EHFM, van Veen M.P, van Engelen JGM. 2002. Report No. 613340003. Pest control products fact sheet. To assess the risks for consumer. RIVM, Bilthoven, The Netherlands. Calabrese EJ. 1985. Uncertainty factors and interindividual variation. Regul Toxicol Pharm 5:190196. Calabrese EJ, Gilbert CE. 1993. Lack of total independence of uncertainty factors (Ufs): Implications for the size of the total uncertainty factor. Regul Toxicol Pharm 17:44-51. CEFIC. 2004. LRI, the Long Range Research Initiative, CEFIC - European Chemical Industry Council, Brussels, Belgium. www.cefic-lri.org.

206

References

COMEST. 2005. The precautionary principle. World Commission on the Ethics of Scientfic Knowledge and Technology (COMEST), United Nations Educational, Scientific and Cultural Organisation (UNESCO), Division of Ethics and Science and Technology, Paris, France. Copeland TL, Paustenbach DJ, Harris MA, Otani J. 1993. Comparing the results of a Monte Carlo analysis with EPA’s reasonable maximum exposed individual (RMEI): a case study of a former wood treatment site. Regul Toxicol Pharm 18:275-312. Costa LG, Murphy SD. 1983. Unidirectional cross-tolerance between the carbamate insecticide propoxur and the organophosphate disulfoton in mice. Fund Appl Toxicol 3:483-488. Crommentuijn C, Sijm D, Bruijn J, van Leeuwen K, van de Plassche E. 2000. Maximum permissible and negligible concentrations for some organic substances and pesticides. J Environ Manage 58:297-312 Crump KS 1984. A new method for determining allowable daily intakes. Fund Appl Toxicol 4:854-871. Custer TW, Mitchell CA. 1987. Exposure to insecticides of brushland wildlife within the lower Rio Grande Valley, Texas, USA. Environ Pollut 45:207-220. Daly AK, Cholerton S, Gregory W, Idle JR. 1993. Metabolic polymorphisms. Pharmac Therapeut 57:129-160. Davis JM, Jarabek AM, Mage DT, Graham JA. 1998. The EPA health risk assessment of methylcyclopentadienyl manganese tricarbonyl (MMT). Risk Anal 18:57-70. Dekkers S, Telman J, Rennen MAJ, Appel MJ, De Heer C. 2006. Within-animal vaiation as an indication of the minimal magnitude of the Critical Effect Size for continuous toxicological parameters applicable in the Benchmark Dose approach. Risk Anal 26:876-880. Den Hollander HA, Van de Meent D. 2004. Appendix to SimpleBox 3.0: A multimedia mass balance model for evaluating the environmental fate of chemicals. RIVM report 601200003a/ 2004. Bilthoven, the Netherlands. http://ecb.jrc.it/DOCUMENTS/Existing-Chemicals/EUSES/ EUSES_2.0_Documentation/RIVM_report_601200003a_SimpleBox_3.0.doc De Nijs ACM, De Greef J. 1992. Ecotoxicological risk assessment of the cationic fabric softener DTDMAC II. Exposure modelling. Chemosphere 24:611-627.

207


References


Diderich R. 2007. The OECD Chemicals Programme. In: Van Leeuwen CJ, Vermeire TG (eds). 2007. Risk assessment of chemicals: an introduction, Second edition. Springer, Dordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book), pp 623638.. Di Giulio RT, Benson eds. 2002. Interconnections between Human Health and Ecological Integrity. SETAC Press, Pensacola, FL, USA. Di Toro DM, Zarba CS, Hansen DJ, Berry WJ, Schwarz RC, Cowan CE, Pavlou SP, Allen HE, Thomas NA, Paquim PR. 1991. Technical basis of establishing sediment quality criteria for non-ionic organic chemicals using equilibrium partitioning. Environ Toxicol Chem 10:1541-1583. DK-EPA. 2001. Environmental factors and health. Principles for derivation of health based guidance values and quality criteria for chemical substances. Appendix 1. Ministry of Environment and Energy, Copenhagen, Denmark. Dourson ML. 1996. Evolution of science-based uncertainty factors in non-cancer risk assessment. Regul Toxol Pharm 24:108-120. Dourson ML, Lu FC. 1995. Safety/risk assessment of chemicals compared for different expert groups. Biomed Environ Sci 8: 1-13. Dourson ML, Stara JF. 1983. Regulatory history and experimental support of uncertainty factors. Regul Toxicol Pharm 3:224-238 EC. 1967. Council Directive of 27 July 1967 on the approximation of the laws, regulations and administrative processes relating to the classfication, packaging and labelling of dangerous substances (67/548/EEC) Official Journal of the European Communities 196. EC. 1973. First Programme of Action on the Environment. Official Journal of the European Communities 16, No. C112, December 20, 1973. EC. 1992. Council Directive of 30 April 1992 amending for the seventh time Directive 67/548/EEC on the approximation of the laws, regulations and administrative provisions relating to the classification, packaging and labelling of dangerous substances (92/32/EEC). Official Journal of the European Communities L154.

208

References

EC. 1993a. Commission Directive 93/67/EEC of 20 July 1993, laying down the principles for the assessment of risks to man and the environment of substances notified in accordance with Council Directive 67/548/EEC. Official Journal of the European Communities L227. EC. 1993b. Technical guidance documents in support of the risk assessment Directive (93/67/EEC) for new substances notified in accordance with the requirements of Council Directive 67/548/EEC. Commission of the European Communities, Brussels, Belgium. EC. 1993c. Council Regulation (EC) 793/93 of 23 March 1993 on the evaluation and control of the risks of existing substances. Official Journal of the European Communities L84. EC. 1994. Commission Regulation (EC) 1488/94 of 28 June 1994, laying down the principles for the assessment of risks to man and the environment of existing substances in accordance with Council Regulation (EEC) No. 793/93. Official Journal of the European Communities L161. EC. 1996a. Technical Guidance Documents in support of Directive 93/67/EEC on risk assessment of new notified substances and Regulation (EC) No. 1488/94 on risk assessment of existing substances, Parts I,II,III, IV. Luxembourg, Office for Official Publications of the European Community L-2965, EC catalogue numbers CR-48-96-001, 002, 003, 004-EN-C. EC. 1996b. EUSES, the European Union System for the Evaluation of Substances. Bilthoven, The Netherlands, National Institute for Public Health and the Environment (RIVM). Available through the European Chemicals Bureau, Ispra, Italy . EC. 1998. Directive 98/8/EC of the European Parliament and of the Council of February 16, 1998 concerning the placing of biocidal products on the market. Official Journal of the European Communities L123. EC. 1999. European Union Risk Assessment Report, Dibutylphthalate. European Chemicals Bureau, Institute of Health and Consumer Protection, Ispra, Italy. Draft May 1999, prepared by The Netherlands’ rapporteur. EC. 2000. Communication from the Commisison on the precautionary principle. COM(2000)1, 02.02.2000. Commission of the European Communities, Brussels, Belgium. EC. 2002a. Technical Notes for Guidance. Human exposure to biocidal products – Guidance on exposure estimation. European Chemicals Bureau, Ispra, Italy.

209


References


EC. 2002b. Technical Notes for Guidance on Annex I, IA and IB inclusion - Technical Notes for Guidance in Support of Directive 98/8/EC of the European Parliament and the Council concerning the placing of biocidal products on the market. Principles and Practical procedures for the inclusion of active substances in Annexes I, IA and IB. European Chemicals Bureau, Ispra, Italy. EC. 2003a. Technical Guidance Document on Risk Assessment (TGD) in support of Commission Directive 93/67/EEC, Commission Regulation (EC) No 1488/94 and Directive 98/8/EC) European Chemicals Bureau, Ispra, Italy. EC. 2003b. Communication from the Commission to the Council, the European Parliament and the European Economic and Social Committee – A European Environment and Health Strategy. COM/2003/0338. Commission of the European Communities, Brussels, Belgium. EC. 2003c. Proposal for a regulation of the Europan Parliament and of the Council concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency and amending Directive 1999/45/EC and Regulation (EC) {on Persistent Organic Pollutants} Proposal for a Directive of the Europan Parliament and of the Council amending Council Directive 67/548/EEC in order to adapt it to Regulation (EC) of the European Parliament and of the Council concerning the registration, evaluation, authorisation and restriction of chemicals. European Commission, Brussels, Belgium, COM(2003)644. EC. 2003d. European Union Risk Assessment Report, dibutyl phthalate, CAS No: 84-74-2, EINECS No: 201-557-4, with addendum 2004, first Priority List, Vol.29. European Chemicals Bureau, Institute of Health and Consumer Protection, Joint Research Centre, Ispra, Italy. LB-NA-19840EN-C Office for Official Publications of the European Commission, ISBN 92-894-1276-3. EC. 2004a. Communication from the to the Council, the European Parliament, the European Economic and Social Committee – “The European Environment & Health Action Plan 2004”. COM(2004)416. Commission of the European Communities., Brussels, Belgium. http://europa. eu.int/comm/environment/health/index_en.htm. EC. 2004b. European Union System for the Evaluation of Substances 2.0 (EUSES 2.0). Prepared for the European Chemicals Bureau by the National Institute for Public Health and the Environment (RIVM), Bilthoven, the Netherlands. Available from the European Chemicals Bureau, http://ecb.jrc.it. EC. 2006. Commission recommendation of 11 April 2006 on risk reduction measures for the substances: Dibutylphthalate; 3,4-Dichloroaniline; Di-‘isodecyl’phthalate; 1,2-Benzenedicarboxylic acid, di-C9-11-branched alkyl esters, C10-rich; Di-

210

References

‘isononyl’phthalate; 1,2-Benzenedicarboxylic acid, di-C8-10-branched alkyl esters, C9-rich; Ethylenediaminetetraacetate; Methyl acetate; Monochloroacetic acid; n-Pentane; Tetrasodium Ethylenediaminetetraacetate, (2006/283/EC). Off. J. Eur. Union L104:45-47. EC. 2007. Corrigendum to Regulation (EC) No 1907/2006 of the European Parliament and the Council of 18 December 2006 concerning the Registration, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency and amending Directive 1999/45/EC and repealing Council Regulation (EC) No. 1488/94 as well as Council Directive 76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/105/EC, and 2000/21/EC (OJ L 396, 30.12.2006). Official Journal of the European Communities L136 Volume 50, 29 May 2007. ECHA.2008a. Guidance for information requirements and chemical safety assessment. Guidance for the implementation of REACH. European Chemicals Agency, Helsinki, Finland, May 2008. www.echa.europa.eu. ECHA.2008b. Guidance for information requirements and chemical safety assessment. Guidance for the implementation of REACH. Chapter R.19: Uncertainty analysis in the Chemical Safety Assessment. European Chemicals Agency, Helsinki, Finland, May 2008. www.echa.europa.eu. ECETOC. 1995. Assessment factors in human health risk assessment. ECETOC Technical Report no. 68. European Centre for Ecotoxicology and Toxicology of Chemicals, Brussels, Belgium. ECETOC. 2003. Derivation of assessment factors for human health risk assessment. European Centre for Ecotoxicology and Toxicology, Brussels, Belgium, ECETOC Technical Report No. 86. Ecobichon DJ, Joy RM. 1994. Pesticides and neurological diseases, 2nd edition. CRC Press. Boca Raton, USA. Edler L, Poirier K, Dourson M, Kleiner J, Mileson B, Nordmann H, Renwick A, Slob W, Walton K, Wurtzen G. 2002. Mathematical modelling and quantitative methods. Food Chem Toxicol 40:283–326. EEA. 2001. Late lessons from early warnings: the precautionary principle1896-2000. Environmental issue report 22, European Environmental Agency, Copenhagen, Denmark. ISBN 92-9167-323-4.. EEA. 2007. Europe’s environment, the fourth assessment. European Environmental Agency, Copenhagen, Denmark. ISBN 978-92-9176-932-4. Available from www.eea.europa.eu..

211


References


EFSA. 2006. Guidance of the Scientific Committee on a request from EFSA related to uncertainties in dietary exposure assessment. The EFSA Journal 438:1-54. Evans JS, Rhomberg L, Williams PL, Wilson AM, Baird SJS. 2001. Reproductive and developmental risks from ethylene oxide: a probabilistic characterization of possible regulatory thresholds. Risk Anal 21:697-718. Falk-Filipsson A, Hanberg A, Victorin K, Warholm M, Wallen M. 2007. Assessment factors – Applications in health risk assessment of chemicals. Environ Res 104:108-127. FAO/WHO. 1997. Risk management and food safety. Paper 65, Report of a joint FAO/WHO expert consultation, 27-31 January 1997. FAO Food and Nutrition, Rome, Italy. Feldman HA, McMahon TA. 1983. The 3/4 mass exponent for energy metabolism is not a statistical artifact. Resp Physiol 52:149-163. Finley BL, Paustenbach DJ. 1994. The benefits of probabilistic exposure assessment: Three case studies involving contaminated air, water, and soil. Risk Anal 14:53-73. Finley BL, Proctor D, Scott P, Harrington N, Paustenbach D, Price P. 1994. Recommended distributions for exposure factors frequently used in health risk assessment. Risk Anal 14: 533553. Frewer LJ, Fischer ARH, Van den Brink PJ, Byrne P, Brock T, Brown C, Crocker T, Goerlitz G, Hart A, Scholderer J, Solomon K. 2008. Potential for the adoption of probabilistic risk assessments by end-users and decision-makers. Hum Ecol Risk Assess 14:166-178. Funtowicz SO, Ravetz JR. 1990. Uncertainty and quality in science for policy. Kluwer Academic Publishers, Dordrecht, The Netherlands. Gee D. 2008. Establishing evidence for early action: the prevention of reproductive and developmental harm. Basic Clin Pharmacol Toxicol 102:71-72. Groeneveld CN, Paulussen JJG, Houben GF, Hakkert BC. 1998. Extrapolation for exposure duration in oral toxicity. TNO report V98.276. TNO Nutrition and Food Research Institute, Zeist, The Netherlands.

212

References

Groeneveld C, Hakkert B, Bos P, de Heer C. 2004. Extrapolation for exposure duration in oral toxicity: a quantitative analysis of historical toxicity data.. Hum Ecol Risk Assess 10:709-716. Hakkert BC, Stevenson H, Bos PMJ, van Hemmen JJ. 1996. Methods for the establishment of HealthBased Recommended Occupational Exposure Limits for existing substances. TNO Nutrition and Food Research, TNO report V96.463, Zeist, The Netherlands. Hart A, Roelofs W, Hardy AR, Macleod A. 2007. Comparative review of risk terminology. A comparative review of terminology and expressions used by the three scientific committees established by Commission Decision 2004/210/EC and by their predecessors established by Commission Decision 97/579/EC (repealed by Commission Decision 2004/210/EC). Report No. S12.454739. The Central Science Laboratory, DEFRA, Sand Hutton, York, UK. Hattis D, Erdreich L, Ballew M. 1987. Human variability ans susceptability to toxic chemicals - a preliminary analysis of pharmacokinetics data from normal volunteers. Risk Anal 7:414-426. Hattis D, Banati P, Goble R, Burmaster D. 1999. Human interindividual variability in parameters related to health risks. Risk Anal 19:711-726. Health Council. 1985. Advies inzake uitgangspunten voor normstelling; De inzichtelijke opbouw van advieswaarden voor niet-mutagene, niet-carcinogene en niet-immunotoxische stoffen.. Report no. 31 (in dutch). Health Council of The Netherlands. Health Council, The Hague, The Netherlands. Health Council. 1996a. Risk is more than a number. Reflexions on the development of the environmental risk management approach. 1996/03E. Health Council of the Netherlands, Committee on risk measures and risk assessment, The Hague, The Netherlands. ISBN: 90-5549112-8. Health Council 1996b. Toxicology-based recommended exposure limits. Report No. 1996/12. Health Council of The Netherlands Rijswijk. Health Council, Rijswijk, The Netherlands. Höfer T, Gerner I, Gundert-Remy U, Liebsch M, Schulte A, Spielmann H, Vogel R and Wettig K. 2004. Animal testing and alternative approaches for the human health risk assessment under the proposed new European chemicals regulation. Arch Toxicol 78: 549-564. Hoffman FO, Hammonds JS. 1994. Propagation of uncertainty in risk assessments: The need to distinguish between uncertainty due to lack of knowledge and uncertainty due to variability. Risk Anal 14:707-712. 213


References


Hoover SM. 1999. Exposure to persistent organochlorines in Canadian breast milk: a probabilistic assessment. Risk Anal 19:527-545. Hope. 2007. What’s wrong with risk assessment? Hum Ecol Risk Assess 13:1159-1163. Hrubec J, Toet C. 1992. Predictability of the removal of organic compounds by drinking water treatment. RIVM Report No. 714301007. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Huff RA, Abou-Donia MB. 1995. In vitro effect of chlorpyrifos oxon on muscarinic receptors and adenylate cyclase. Neurotoxicology 16:281-290. IHCP. 2004. European Information System on risks of chemicals in products (EIS-CHEMRISKS). Institute of Health and Consumer Protection, Joint Research Centre, Ispra, Italy. http://www.jrc.cec. eu.int/pce/eis_activities.htm ILSI. 1998. Aggregate exposure assessment. An ILSI Risk Science Instute Workshop Report. Washington D.C., USA, International Life Sciences Institute (ILSI). ISBN: 1-57881-040-X. IOMC. 2006. Participation in the implementation of the strategic approach to international chemical management. Interorganization Programme for the sound management of chemicals, ICCM Conference Room Paper. International Conference on Chemicals Management, Dubai, United Arab Emirates. IPCS. 1994. Assessing human health risks of chemicals: derivation of guidance values for healthbased exposure limits. Geneva, Switzerland, International Programe on Chemical Safety, WHO/ UNEP/ILO, Environmental Health Criteria 170. IPCS. 1999. Principles for the assessment of risks to human health from exposure to chemicals. International Programme on Chemicals Safety, World Health Organisation, Geneva, Switzerland, WHO/UNEP/ILO. Environmental Health Criteria 210. IPCS. 2000. Guidance document for the use of data in development of Chemical-Specific Adjustment Factors (CSAF) for inter-species differences and human variability in dose/concentration repsonse assessment. International Programme on Chemical Safety, World Health Organisation, Geneva, WHO/PCS/01.4 . www.ipcsharmonize.org/uncertain.html). IPCS. 2001. Report on Integrated Risk Assessment. WHO/IPCS/IRA/01/12. World Health Organization, Geneva, Switzerland. 214

References

IPCS. 2005. Chemical-specific adjustment factors for interspecies differences and human variability: guidance document for use of data in dose/concentration–response assessment.., IPCS Harmonization Project Document No. 2. http://whqlibdoc.who.int/ publications/2005/9241546786_eng.pdf. IPCS. 2008a.Description of selected key generic terms used in chemical hazard/risk assessment. Joint Project with OECD. International Programme on Chemical Safety, Geneva, Switzerland. http:// www.who.int/ipcs. IPCS. 2008b. Part 1: Guidance Document on characterizing and communicating uncertainty in exposure assessment. Harmonization Project document No.6. International Programme on Chemical Safety, World Health Organisation, Geneva, Switzerland. Jager DT. 1995a. Feasibility of validating the Uniform System for the Evaluation of Substances (USES). Report No. 679102026. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Jager DT. 1995b. Uncertainty Analysis of the Uniform System for the Evaluation of Substances (USES): Example Calculations. Report No. 679102032., National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Jager DT, Slob W. 1995. Uncertainty Analysis of the Uniform System for the Evaluation of Substances (USES). National Institute for Public Health and the the Environment (RIVM), Report No. 679102027, Bilthoven, The Netherlands. Jager T, Rikken MGJ, Van der Poel P. 1997. Uncertainty analysis of EUSES: Improving risk management by probabilistic risk assessment. RIVM Report No. 679102 039.National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Jager T, ed. 1998a. Evaluation of EUSES: inventory of experiences and validation activities. Report No. 679102048. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Jager T. 1998b. Uncertainty analysis of EUSES: interviews with representatives from Member States and industry. Report no. 679102 047. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands.

215


References


Jager T, Den Hollander HA, Janssen GB, Van der Poel P, Rikken MGJ, Vermeire TG. 2000. Probabilistic risk assessment for new and existing chemicals: Sample calculations. RIVM-report No. 679102049. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Jager T, VermeireTG, Rikken, MG, van der Poel P. 2001a. Opportunities for a probabilistic risk assessment of chemicals in the European Union. Chemosphere 43: 257-64. Jager T, Den Hollander H, Van der Poel, P, Rikken MGJ, Vermeire T. 2001b. A probabilistc environmental risk assessment for dibutylphthalate (DBP). Hum Ecol Risk Assess 6:1681-1697. Jardine C, Hrudey SE, Shortreed JH, Craig L, Krewski D, Furgal C, McColl S. 2003. Risk management frameworks for human health and environmental risks. J Toxicol Env Health B 6:569-720. Johnson MK. 1975. The delayed neuropathy caused by some organophosphorous esters: Mechanism and challenge. Crit Rev Toxicol 3:289-310. Kalberlah F, Schneider K. 1998. Quantification of extrapolation factors. Final report of the research project No. 116 06 113 of the Federal Environmental Agency. Dortmund/Berlin, Schriftenreihe der Bundesanstalt für Arbeitsschutz un Arbeitsmedizin – Forschung – FB 797. Kalberlah F, Schneider K, Schuhmacher US, Voss J-U, Ioannidis I, Oltmanns J. 1999. Zeitextrapolation und Interspezies-extrapolation bei lokal wirksamen Stoffen met begrenzter Datenlage. Dortmund/ Berlin, Schriftenreihe der Bundesanstalt für Arbeitsschutz und Arbeitsmedizin. Fb 862. Ann Occup Hyg 46:175-185, 2002. Kalberlah F, Föst U, Schneider K. 2002 Time extrapolation and interspecies extrapolation for locally acting substances in case of limited toxicological data. Ann Occup Hyg 46:175-185. Kavlock RJ, Schmid JE, Setzer RW. 1996. A simulation study of the influence of study design on the estimation of benchmark doses for developmental toxicity. Risk Anal 16:399-410. Klepper O, Bakker J, Traas TP, Van de Meent D. 1998. Mapping the potentially affected fraction (PAF) of species as a basis for comparison of ecotoxicological risks between substances and regions. J Hazard Mater 61:337-344. Klinke A, Renn O. 2002. A new approach to risk evaluation and management: risk-based, precautionbased, and discourse-based strategies. Risk Anal 22:1071-1094. 216

References

Kramer HJ, van den Ham WA, Slob W, Pieters MN. 1996. Conversion factors estimating indicative chronic No-Observed-Adverse-Effect Levels from short-term toxicity data. Regul Toxicol Pharm 23:249-255. Lehman AJ, Fitzhugh OG. 1954. 100-Fold margin of safety. Assoc Food Drug Off 18:33-35. Leisenring W, Ryan L. 1992. Statistical properties of the NOAEL. Regul Toxicol Pharm 15:161-171. Lessmann K, Beyer A, Klasmeier J, Matthies M. 2005. Influnece of distributional shape of substance parameters on exposure model output. Risk Anal 25:1137-1145 Levin R, Jansson SO, Rudén C. 2004. Indicators of uncertainty in chemical risk assessments. Regul Toxicol Pharm 39:33-43. Lewis SC, Lynch JR, Nikiforov AI. 1990. A new approach to deriving community exposure guidelines from “No-observed-adverse-effect levels”, Regul Toxicol Pharm 11:314-330. Lönngren R. 1992. International approaches to chemicals control. A historical overview. The National Chemicals Inspectorate (KEMI), Stockholm, Sweden. ISBN 91-7932-032-5. Lu FC. 1979. Assessment at an international level of health hazards to man of chemicals shown to be carcinogenic in laboratory animals. In: Coulston, E.F. (ed.), Regulatory aspects of carcinogenesis. New York, Academic Press, pp. 315-328. Mackay D, Di Guardo A, Paterson S, Kicsi G. Cowan CE. 1996. Assessing the fate of new and existing chemicals: a five-stage process. Environ Toxicol Chem 15:1618-1626. Mackay D, Paterson S, Shiu WY. 1992. Generic models for evaluating the regional fate of chemicals. Chemosphere 24:695-717. MacPhail RC, Glowa JR. 1999. Quantitative risk assessment in neurotoxicology: Past, present and future. In Tilson, HA. Harry GJ, eds, Neurotoxicology, Taylor and Francis, New York, USA, pp. 367-382. McColl RS. 1989. Biological safety factors in toxicological risk assessment. Ottawa, Canada, Report produced for the Environmental Health Directorate, Health Protection Branch, Health and Welfare Canada. Ministry of Supply and Services Canada, Cat. H49-49/1990E, ISBN 0-66217638-3.

217


References


McKone TE, Ryan PB. 1989. Human exposure to chemicals through food chains: An uncertainty analysis. Environ Sci Technol 23:1154-1163. McKone TE. 1994 Uncertainty and variability in human exposures to soil contaminants through home-grown food: a Monte Carlo assessment. Risk Anal 14:449-463. McNamara BP, Mehlman MA, Shapiro RE, Blumenthal H eds. 1976. Concepts in health evaluation of commercial and industrial chemicals. In: Advances in modern technology, 1, part 1, New concepts in safety. Washington DC, USA. Hemisphere pp. 61-140. Mineau P, ed. 1991. Cholinesterase-inhibiting insecticides: their impact on wildlife and the environment. Elsevier, New York, USA. Mineau P, Boersma DC, Collins B. 1994. An analysis of avian reproduction studies submitted for pesticide registration. Ecotoxicol Environ Safety 29:304-329. Mineau P, Collins BT, Baril A. 1996. On the use of scaling factors to improve interspecies extrapolation of acute toxicity in birds. Regul Toxicol. Pharm 24:24-29. Money, CD, Van Hemmen JJ, Vermeire TG. 2007. Scientific governance and the process for exposure scenario development in REACH. J Expo Sci Env Epid 17 (supplement 1):S34-S37. Moore DRJ. 1996. Using Monte Carlo analysis to quantify uncertainty in ecological risk assessment: Are we gilding the lily or bronzing the dandelion? Hum Ecol Risk Assess 2:628-633. Moore DRJ, Elliott BJ. 1996. Should uncertainty be quantified in human and ecological risk assessments used for decision making? Hum Ecol Risk Assess 2:11-24. Munns Jr W, Suter II GW, Damstra T, Kroes R, Reiter L, Marafante E. 2003. Integrated risk assessment – Results from an international workshop. Hum Ecol Risk Assess 9:379-386. Naumann BD, Weidemann PA, Dixit R, Grossman SJ, Shen CF, Sargent EV. 1997. Use of toxicokineticand toxicodynamic data to reduce uncertainties when setting occupational exposure limits for pharmaceuticals. Hum Ecol Risk Assess 3:555-565. Nessel CS, Lewis C, Stauber KL, Adgate JL. 1995. Subchronic to chronic exposure extrapolation: toxicologic evidence for a reduced uncertainty factor. Hum Ecol Risk Assess 1:516-526.

218

References

Nichols JK, Crawford PJ. 1983. Managing chemicals in the 1980s. Organisation for Economic Cooperation and Development (OECD), Paris, France. ISBN 92-64-12431-1. OECD. 1989. Report of the OECD Workshop on ecological effects assessment. Organisation of Economic Co-operation and Development, Paris, France. OECD Environment Monographs No. 26. OECD. 1992. Report of the OECD workshop on the extrapolation of laboratory aquatic toxicity data on the real environment. OECD Environment Monographs No. 59. OECD, Paris, France. OECD. 1993. Occupational and consumer exposure assessments. Organisation of Economic Cooperation and Development, Paris, France. OECD Environment Monographs No. 70. Pandian MD, Furtaw EJ. et al. 1997. THERdbASE: Total human exposure relational database and advanced simulation environment. Harry Reid Center for Environmental Studies, University of Nevada, Las Vegas, USA. Developed under contract to the EPA, ORD. Park CN, Hawkins NC. 1993. An overview of cancer risk assessment. Toxicol Method 3:63-86. Paustenbach PD. 2000. The practice of exposure assessment: a state-of-the-art review. J Toxicol Environ Health B, 3:179-291. Pieters MN, Kramer HJ, Slob W. 1998. Evaluation of the uncertainty factor for subchronic-to-chronic extrapolation: statistical analysis of toxicity data. Regul Toxicol Pharm 27:108-111. Pieters MN, Bakker M, Slob W. 2004. Reduced intake of deoxynivalenol in The Netherlands: a risk assessment update. Toxicol Lett 153:145–153. Polder MD, Hulzebos EM, Jager DT. 1998. Bioconcentration of gaseous organic chemicals in plant leaves: comparison of experimental data with model predictions. Environ Toxicol Chem 17:962968. Power M, McCarthy LS. 1997. Fallacies in ecological risk assessment practices. Environ Sci Technol 31:370-375. Power M, McCarty LS. 1998. A comparative analysis of environmental risk assessment/risk management frameworks. Environ Sci Technol 32:224A-231A.

219


References


Price PS, Keenan RE, Swartout JC, Gillis CA, Carlson-Lynch H, Dourson ML. 1997. An approach for modeling noncancer dose responses with an emphasis on uncertainty. Risk Anal 17:427 - 437. Renn O. 2008. Concepts of risk: an interdisciplinary review; Part 1: Disciplinary risk concepts. GAIA 17:50-66. Rennen MAJ, Hakkert BC, Stevenson H., Bos PMJ. 1999. Interspecies extrapolation: A quantitative analysis of historical toxicity data. TNO-report No. V99.210. TNO Nutrition and Food Research Institute, Zeist, The Netherlands. (Accepted for publication in: Comments on Toxicology. Special issue on Non-cancer Health Risk Assessment in the 21st Century). Rennen MAJ, Hakkert BC, Stevenson H, Bos PMJ. 2001. Data-based derived values for the interspecies extrapolation, a quantitative analysis of historical toxicity data. Comment Toxicol 7:423–436. Rennen MAJ, Bouwman T, Wilschut A, Bessems JGM, De Heer C. 2004. Oral-to-inhalation route extrapolation in occupational health risk assessment: a critical assessment. Regul Toxicol Pharm 39:5-11. Renwick AG. 1991. Safety factors establishment of acceptable daily intakes. Food Addit Contam 8:135-140. Renwick AG. 1993a. Data-derived safety factors for the evaluation of food additives and environmental contaminants. Food Addit Contam 10:275-305. Renwick AG. 1993b. An analysis of the risk of exceeding the acceptable or tolerable daily intake. Regul Toxicol Pharm 18:463-480. Rikken MGJ, Lijzen JPA. 2004. Update of risk assessment models for the indirect human exposure. Report No. 601516011, RIVM, Bilthoven, The Netherlands. RIVM, VROM, VWS. 1998. Uniform System for the Evaluation of Substances 2.0 (USES 2.0). National Institute for Public Health and the Environment (RIVM), Ministry of Housing, Spatial Planning and the Environment (VROM), Ministry of Health, Welfare and Sport (VWS), The Netherlands. RIVM report no. 679102044. Available from RIVM/BDA, Bilthoven, The Netherlands. Available at http://www.rivm.nl/bda.

220

References

RIVM, VROM, WVC. 1994. Uniform System for the Evaluation of Substances (USES), version 1.0. National Institute for Public Health and Environmental Protection (RIVM), Ministry of Housing, Physical Planning and Environment (VROM), Ministry of Welfare, Health and Cultural Affairs (WVC). Ministry of Housing, Physical Planning and Environment. Distribution No. 11144/150. The Hague, The Netherlands. Rodricks J, Taylor, MR. 1990. Application of risk assessment to food safety decision making. In: Glickman TS, Gough M (eds). 1990. Readings in risk. Resources for the future, Washington DC, USA. ISBN 0-915707-55-1, pp 143-153. Roelofs W, Huijbregts MAJ, Jager, T, Ragas AMJ. 2003. Prediction of ecological no-effect concentrations for initial risk assessment: combining substance-specific data and database information. Environ Toxicol Chem 22:1387-1393. Rogers MD. 2003. Risk analysis under uncertainty, the Precautionary Principle, and the new EU chemicals strategy. Regul Toxicol Pharm 37:370-381. Rulis AM, Hattan DG. 1985. FDA’s priority based assessment of food additives. Regul Toxicol Pharm 5:152-174. Sabljic A, Güsten H, Verhaar H, Hermens J. 1995. QSAR modelling of soil sorption. Improvements and systematics of log Koc vs. log Kow correlations. Chemosphere 31: 4489-4514. Sandin P. 1999. Dimensions of the precautionary principle. Hum Ecol Risk Assess 5:889-907. SCC. 2003. The future of risk assessment in the European Union. The second report on the harmonization of risk assessment procedures. Scientific Steering Committee, Hhealt and Consumer Protection Directorate, European Commission, Brussels, Belgium. Schaddelee M. 1997. Interspecies en intraspecies variabiliteit. Trainee report. RIVM, Bilthoven, The Netherlands. Schneider K, Hassauer M, Oltmanns J, Schuhmacher-Wolz U, Elmshäuser E, and Mosbach-Schulz O. 2004. Warscheinlichkeitsrechnung als Hilfsmittel zur Wirkungsabschätzung bei Arbeitnehmern. Schriftenreihe der Bundesanstalt für Arbeitsschutz und Arbeitsmedizin Fb 1012.

221


References


Schneider K, Hassauer M, Oltmanns W, Schuhmacher-Wolz U, Elmshäuser E, Mosbach-Schulz O. 2005. Uncertainty analysis in workplace effect assessment. Project F1824/F1825/F1826, Federal Institute for Occupational Safety and Health, Dortmund, Germany. http://www.baua.de/english/ fors/f1824-26e.htm. Schwarz S. 2000. Quality assurance of exposure models for environmental risk assessment of substances. Doctoral Thesis. Institute of Environmental Systems Research, Department of Mathematics and Compputer Science, University of Osnabrück, Osnabrück, Germany. Schwartz S, Berding S, Trapp S, Matthies M. 1998. Quality criteria for environmental risk assessment software – Using the example of EUSES. Environ Sci Pollut R 5:217-222. Seiler FA, Alvarez JL. 1996. On the selection of distributions for stochastic variables. Risk Anal 16:529. Simcox NJ, Fenske RA, Wolz SA, Lee IC. Kalman DA. 1995. Pesticides in household dust and soil: exposure pathways for children of agricultural families. Env Health Persp 103:1126-1134. Slob W. 1987. Strategies in applying statistics in ecological research. PhD thesis, Free University, Amsterdam, The Netherlands. Slob W. 1994. Uncertainty analysis in multiplicative models. Risk Anal 14:571-576. Slob W, Pieters MN. 1997. Few large, or many small groups? An evaluation of toxicological study designs using computer simulations. Report no. 620110006. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Slob W, Pieters MN. 1998. A probabilistic approach for deriving acceptable human intake limits and human risks from toxicological studies: general framework. Risk Anal 18:787-798. Slob W, Moerbeek M, Rauniomaa E, Piersma AH, 2005. A statistical evaluation of toxicity study designs for the estimation of the Benchmark dose in continuous endpoints. Toxicol Sci 84:167185. Slovic. 2000. The perception of risk: Risk, society and policy. Earthscan Publications, London, UK. ISBN 1-85383-527-7 Sonich-Mullin C. 1995. Harmonisation of approaches to the assessment of risk from exposure to chemicals. Risk Newsletter, 4th quarter 1995, pp. 6-7. 222

References

Stevenson H, Bos PMJ, de Raat WK. 1995a. Review of applied factors to derive health-based resommended exposure levels. TNO-report V95.092. TNO Nutrition and Food Research Institute, Zeist, The Netherlands. Stevenson H, Bos PMJ, van Hemmen JJ. 1995b. Possibilities for a harmonised approach on assessment factors. TNO-report V95.697. TNO Nutrition and Food Research Institute, Zeist, The Netherlands. Struijs J. 1996. Simpletreat 3.0: a model to predict the distribution and elimination of chemicals by sewage treatment plants. Report No. 719101025. (RIVM), Bilthoven, The Netherlands Suter II GW. 2007. Ecological risk assessment in the United States Environmental Protection Agency: a historical overview. Integr Environ Assess Manage 4:285–289. Suter II GW , Vermeire T, Munns Jr W, Sekizawa, J. 2003. Framework for the integration of health and ecological risk assessment. Hum Ecol Risk Assess 9:281-301. Suter II GW, Vermeire T, Munns Jr WR, Sekizawa. J. 2005. An integrated framework for health and ecological risk assessment. Toxicol Appl Pharm 207:S611 – S616. Swartout JC, Price PS, Dourson ML, Carlson-Lynch HL, Keenan RE. 1998. A probabilistic framework for the Reference Dose (Probabilisitc RfD). Risk Anal 18:271-282. Teuschler LK, Dourson ML, Stiteler WM, McClure, P. and Tully, H. 1999. Health risk above the reference dose for multiple chemicals. Regul Toxicol Pharm 30: S19-S26. Thompson KM, Graham JD. 1996. Going beyond the single number: Using probabilistic risk assessment to improve risk management. Hum Ecol Risk Assess 2:1008-1034. Traas TP, Luttik R, Jongbloed RH. 1996. A probabilistic model for deriving soil quality criteria based on secondary poisoning of top predators. I. Model description and uncertainty analysis. Ecotox Environ Safe 34:264-278. Trapp S, Matthies M. 1995. Generic one compartment model for uptake of organic chemicals by foliar vegetation. Environ Sci Technol 29(9):2333-2338. Trapp S, Schwartz, S. 2000. Proposals to overcome limitations in the EU chemical risk assessment scheme. Chemosphere 41:965-971.

223


References


Travis CC, Arms AD. 1988. Bioconcentration of organics in beef, milk and vegetation. Environ Sci Technol 22:271-274. Truhaut R. 1977. Eco-Toxicology - Objectives, Principles and Perspectives. Ecotox Environ Safe 1:151-173. UN. 1992. Environmentally sound management of toxic chemicals including prevention of illegal international traffic in toxic and dangerous products. Agenda 21, Chapter 19. Commission for Sustainable Development, United Nations Conference on Environment and Development, Rio de Janeiro, Brazil. UN-WCED. 1987. Our common future. United Nations World Commission on Environment and Development. Oxford University Press, Oxford, UK. ISBN 0-19-282080-X. US EPA. 1992. Framework for Ecological Risk Assessment. EPA/630/R-92/001. Risk Assessment Forum, Environmental Protection Agency, Washington DC, USA. US EPA. 1993. IRIS – Reference Dose (RfD): description and use in health risk assessments. Background document 1A. US Environmental Protection Agency. http://www.epa.gov/iris/rfd.htm. US EPA, 2001. Preliminary OP cumulative risk assessment. Environmental Protection Agency, Washington DC, USA. Available at http://www.epa.gov/pesticides US EPA. 1995. The use of the Benchmark Dose approach in health risk assessment. EPA/630/R-94/007. US Environmental Protection Agency, Risk Assessment Forum. Washington, USA. US EPA. 1997. Guiding principles for Monte Carlo analysis (EPA/630/R-97/001). U.S. Environmental Protection Agency, Washington DC, USA. US NRC. 1983. Risk assessment in the Federal Government: managing the process. National Research Council, National Academy Press, Washington, USA. US NRC 1993. Pesticides in the diets of infants and children. National Research Council, Committee on Pesticides in the Diets of Infants and Children, Board on Agriculture and Board on Environmental Studies and Toxicology, Commission on Life Sciences, National Academy Press, Washington DC, USA, ISBN 0-309-04875-3. US NRC. 1996. Understanding risk: information decisions in a democratic society. National Research Council, National Academy Press, Washington DC USA. 224

References

Van Asselt MBA. 2000. Perspectives on uncertainty and risk. The PRIMA approach to decision support. Kluwer Academic Publishers, Dordrecht, The Netherlands. ISBN 0-7923-6656-5. Van Asselt MBA, Vos E. 2006. The precautionary principle and the uncertainty paradox. J Risk Res 9:313-336. Van de Gevel IA, Hakkert BC. 1997. Methods for interspecies extrapolation: an evaluation. TNOreport V97.529. TNO Nutrition and Food Research, Zeist, The Netherlands. Van de Meent D. 1993. SimpleBox: a generic multimedia fate evaluation model. Report no. 672720001. National Institue of Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van der Jagt K, Munn S, Torslov J, de Bruijn J. 2004. Alternative approaches can reduce the use of test animals under REACH. Joint Research Centre, European Commission, Ispra, Italy. Van der Sluijs JP, Risbey JS, Kloprogge P, Ravetz JR, Funtowicz SO, Quintana SC, Pereira AG, De Marchi B, Petersen A, Janssen PHM, Hoppe R, Huijs SWF. 2003. RIVM/MNP guidance for uncertainty assessment and communication. Detailed Guidance. Utrecht University, Utrecht, The Netherlands, ISBN 90-393-3536-2. Van der Sluijs JP, Craye M, Funtowicz S, Kloprogge P, Ravetz J, Risbey J. 2005. Combining quantitative and qualitative measures of uncertainty in model-based environmental assessment: the NUSAP system. Risk Anal 25:481-492. Van der Voet H, Slob W. 2007. Integration of probabilistic exposure assessment and probabilistic hazard chatacterization. Risk Anal 27:351-371. Van Jaarsveld JA. 1990. An operational atmospheric transport model for priority substances, specifications and instructions for use. Report No. 222501002. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van Leeuwen CJ, Hermens JLM. 1995. Risk assessment of chemicals: an introduction., Kluwer Academic Publishers, Dordrecht, The Netherlands, ISBN 0-7923-3740-9. Van Leeuwen CJ. 2007. General introduction. In: Van Leeuwen, C.J., Vermeire, T.G. (eds) Risk assessment of chemicals: an introduction, Second edition. Springer, Dordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book).

225


References


Van Leeuwen CJ, Hansen BG, De Bruijn JHM. 2007. The management of industrial chemicals in the EU. In: Van leeuwen CJ, Vermeire TG (eds) Risk assessment of chemicals: an introduction, Second edition. Springer, Dordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book), pp 511-551. Van Veen MP. 1996. A general model for exposure and uptake from consumer products. Risk Anal 16:331-338. Van Veen, MP. 1997. CONSEXPO 2.0, consumer exposure and uptake models. Report no. 612810 005. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van Veen MP. 2001. CONSEXPO 3.0. Consumer exposure and uptake models. Report No. 612810011 (with CD-ROM). RIVM, Bilthoven, The Netherlands. Veith GD, Defoe DL, Bergstedt BV. 1979. Measuring and estimating the bioconcentration factor for chemicals in fish. J Fish Res Board Can 36:1040-1048. Verdonck FAM, Souren A, Van Asselt MBA, Van Sprang PA, Vanrolleghem PA. 2007. Improving uncertainty analysis in European Union risk assessment of chemicals. Integr Environ Assess Manage 3:333-343. Vermeire TG, van der Zandt PTJ, Roelfzema H, van Leeuwen CJ. 1994. Uniform System for the Evaluation of Substances I: Principles and Structure. Chemosphere 29:23-38. Vermeire TG, Jager DT, Bussian B, den Haan K, Hansen B, Lundberg I, Niessen H, Robertson S, Tyle H, van der Zandt PTJ. 1997. European Union System for the Evaluation of Substances, EUSES, Principles and Structure. Chemosphere 34: 1823-1836. Vermeire TG, Stevenson H, Pieters MN, Rennen M, Slob W, Hakkert BC. 1999a. Assessment factors for human health risk assessment: a discussion paper. Crit Rev Toxicol 29: 439-490. Vermeire TG, Bos P, van Haelst A, Pieters M, Pronk M, Van Raaij M, Rennen M.. 1999b. Report of the Joint EU/RIVM/TNO Workshop on Interpretations of Margins Of Safety in Human Health Risk Assessment, 21-22 April 1999b. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. August 30, 1999.

226

References

Vermeire T, Jager T, Janssen G, Bos P, Pieters M. 2001a. A probabilistic human health risk assessment for environmental exposure to dibutylphthalate. Hum Ecol Risk Assess 7: 1663-1679. Vermeire T, Pieters M, Rennen M, Bos P. 2001b. Probabilistic assessment factors for human health risk assessment: a practical guide. RIVM Report No. 601516005, TNO Report No. V3489. National Institute for Public Health and the Environment (RIVM), Bilthoven; TNO, Zeist, The Netherlands. www.rivm.nl/bibliotheek/rapporten. Vermeire T, MacPhail R, Waters M. 2003. Integrated human and ecological risk assessment: a case study of organophosphorous pesticides in the environment. Hum Ecol Risk Assess 9: 343-357. Vermeire T, Rikken M, Attias L, Boccardi P, Boeije G, de Bruijn J, Brooke D, Comber M, Dolan B, Fischer S, Heinemeyer G, Koch V. Lijzen J, Müller B, Murray-Smith R, Tadeo J. 2004. European Union System for the Evaluation of Substances, The second version. Chemosphere 59: 473-485. Vermeire T, Munns Jr WR, Sekizawa J, Suter II GW, Van der Kraak G. 2007. An Assessment of Integrated Risk Assessment. Hum Ecol Risk Assess 13:339-354. Vettorazzi G. 1980. Safety factors and their application in the toxicological evaluation. In: The evaluation of toxicological data for the protection of public health. Oxford, UK, Pergamon, pp. 207-223. Vollmer G, Giannoni L, Sokull-Klüttgen B, Karcher W. 1996. Risk assessment, theory and practice. Proceedings of the Workshop Risk Assessment: a Workshop on Practical Experience, European Chemicals Bureau. 27-28 March, 1996. Luxembourg, Office for Official Publications of the European Communities. Ispra, Italy. Walker WE, Harremoës P, Rotmans J, Van der Sluijs JP, Van Asselt MBA, Janssen P, Krayer von Krauss MP. 2003. Defining uncertainty, A conceptual basis for uncertainty management in model-based decision support. Integr Assess 4:5-17. Walton K, Dorne JL, Renwick AG. 2001a. Uncertainty factors for chemical risk assessment: interspecies differences in the in vivo pharmacokinetics and metabolism of human CYP1A2 substrates. Food Chem Toxicol 39:667-680. Walton K, Dorne JL and Renwick AG. 2001b. Uncertainty factors for chemical risk assessment: interspecies differences in glucuronidation. Food Chem Toxicol 39:1175-1190.

227


References


Ward, TR. Mundy WR. 1996. Organophosphorus compounds preferentially affect second messenger systems coupled to M2/M4 receptors in rat frontal cortex. Brain Res Bull 39:49-55. Weed DL.. 2005. Weight of Evidence: a review of concept and methods. Risk Anal 25:1545-1557. Weil CS, McCollister DD. 1963. Relationship between short- and long-term feeding studies in designing an effective toxicity test. J Agr Food Chem 11:486-491. WHO. 1979. Agreed terms on health effects evaluation and risk and hazard assessment of environmental agents. EHE/EHC/79.19. Internal report of a working group. World Health Organisation, Geneva,, Switzerland. Wilschut A, Houben GF, Hakkert BC. 1998. Evaluation of route-to-route extrapolation in health risk assessment for dermal and respiratory exposure to chemicals. TNO Nutrition and Food Research Institute. TNO report V97.520, Zeist, The Netherlands Wine RN, Li L-H, Barnes LH, Gulati, D.K. and Chapin, R.El. 1997. Reproductive toxicity of di-nbutylphthalate in a continuous breeding protocol in Sprague-Dawley rats. Environ Health Persp 105:102-107. Woutersen RA, Opdam JJG, Stevenson H. 1997 Health-based recommended (occupational) exposure limits: value of the benchmark-dose approach in comparison with the No Observed Adverse Effect Level method. TNO Nutrition and Food Research. TNO-report V97.164, Zeist, The Netherlands.

228

Summary

Summary


Summary

Chemicals are important in our everyday life. The chemical industry is the third largest industrial sector in the world. Chemicals add value to our lives, but also add costs due to adverse effects to health and the environment from exposure at production and use and at the waste stage. Chemicals control covers all regulatory and voluntary measures taken to minimize the risks of chemicals to human health and the environment. In this process, risk managers, both in the public sector and in industry, need to be advised by experts of various disciplines. A major discipline is risk assessment. With regard to chemicals, risk can be defined as the probability of adverse effects in an organism, system or (sub)population caused under specified circumstances by exposure to an agent Both risk managers and risk assessors nowadays recognise that uncertainties are part and parcel of risk assessment advice. This thesis concentrates on uncertainty and variability in the risk assessment methodology for industrial chemicals as applied within the current regulatory framework for industrial chemicals in Europe, REACH (Registration, Evaluation, Authorisation and Restriction of Chemicals). The methodological approaches discussed address the risk assessment for both humans and the environment. The aim of this thesis is to investigate in what way the scientific process of risk assessment can improve decision-making knowing that uncertainties are inherently linked to risk assessment. An important element of this decision-making is consideration of precautionary measures in the event of reasonable grounds for concern of a potential risk that cannot, or not in time, be determined with sufficient certainty. The investigation is built on two frameworks: the IPCS/WHO framework for Integrated Risk Assessment (IRA) and the framework for Uncertainty Management of Walker et al. (2003). The framework of Integrated Risk Assessment was developed to improve the quality and efficiency of the assessment of risks of adverse effects on human health and the environment from chemicals, physical factors, and other environmental stressors and to provide more complete and coherent inputs to the decision making process. The underlying idea is that both the scientific discussion and the regulatory responses can benefit from a more integrated, interdisciplinary approach leading to sharing of information, decreased uncertainties and fully informed decisions. The Uncertainty Management framework closely parallels the stages in the Integrated Risk Assessment scheme eand is considered very useful for highlighting different types of uncertainty in risk assessment. Chapter 1 introduces the subject and explains the two frameworks used. It provides the historical and legal context of chemical control and risk assessment at UN, OECD and EU level. This chapter also provides summary information on REACH and on one of its basic

231


Summary


principles, the Precautionary Principle. It is explained that the Precautionary principle can be applied when there exist considerable scientific uncertainties about causality, magnitude, and nature of harm and that some form of scientific analysis, including a thorough uncertainty analysis, is mandatory. Essential is an evaluation of the reliability of the assessment as well as the remaining uncertainties and a correct typology of the uncertainties. Chapter 2 illustrates the concept of integrated human and ecological risk assessment. The concept is applied to a case study of the use of organophosphorous pesticides in a typical farming community. This case study was developed to communicate the IRA approach, to illustrate how IRA assessments might be conducted, and to analyse the benefits and drawbacks of integration. It is argued that this integrated approach helps both the risk manager and the risk assessor in looking more broadly, towards the risk of the use of OPesters. The first stage, problem formulation, is particularly relevant for both information sharing and defining what information and advice is required for an optimal regulatory decision. Problem formulation is the process by which the risk assessment is defined and the plan for analysing and characterising the risk is developed. The methodology used in Chapter 2 is based on the risk assessment methodology developed and harmonised in Europe as described in the EU Technical Guidance Documents and implemented in the PC-application European Union System for the Evaluation of Substances EUSES. The 2004 version of EUSES is described in Chapter 3. Adaptations considered necessary for a full application under REACH are discussed in Chapter 7. In a discussion on the validation status the models used are considered to be state-of-the-art, providing deterministic conservative estimates for standard scenarios based on limited data requirements. In the EU risk assessment of industrial chemicals, it is current practice to characterise risks using a deterministic quotient of the exposure concentration, or the dose, and a no-effect level. A sense of uncertainty is tackled by introducing worst-case assumptions in the methodology. Chapter 4 explains how chemical risk management can benefit from considering those uncertainties, which can be characterised by distributions, in a probabilistic risk assessment (PRA). It discusses the advantages and possibilities of PRA and illustrates this by providing an example calculation. Chapter 5 continues this demonstration of the advantages and drawbacks of PRA, taking the EU risk assessment of the plasticizer dibutylphthalate (DBP) as starting point. This example shows the possibility of including and comparing the various uncertainties

232

Summary

involved in a typical risk assessment, covering the uncertainty in the exposure estimate, the uncertainty in the effect parameter modelled by a Benchmark approach, and the uncertainty in assessment factors used in the extrapolation from experimental animals to sensitive human beings. One important aspect of uncertainty analysis is the widely used application of assessment factors to derive human and ecotoxicological no-effect levels. This subject is reviewed extensively in Chapter 6 with regard to human assessment factors. In addition, this chapter discusses more extensively the probabilistic approach to establish a human health limit value based on Benchmark modelling as applied in Chapter 5. Chapter 7 finally discusses the relevant methodological aspects of risk assessment and uncertainty analysis. The EU risk assessment methodology as implemented in EUSES is compared to the new REACH guidance Although EUSES is based on the Technical Guidance Documents operating until mid-2008 under the pre-REACH EU legislation, much of it is still in line with the updated Guidance for REACH. Important changes and implications for EUSES are discussed this chapter. It further shows how the REACH risk assessment framework can benefit from the IRA-framework and how the IRA-framework in its turn can be improved by taking into account the current understanding of the role of uncertainty analysis in risk assessment and risk management. Uncertainty Management, including a method for classification of uncertainties and a tiered approach, is presented as an additional layer in the IRA framework aiming to improve the management and communication of uncertainty in decision-making processes. The main, general conclusion of this thesis is that both the process and the methodology of risk assessment as a decision-support tool under REACH can be improved. The process can be improved by the introduction of an IRA framework with a strong uncertainty management component (IRA+). The methodology can be improved by a tiered approach for uncertainty analysis, starting with simple deterministic approaches and, if necessary, classification and prioritisation of uncertainties and probabilistic approaches. Qualitative and quantitative tools for uncertainty assessment were shown to be available. Conclusions and recommendations on the process: The most important of the perceived benefits of IRA are increased assessment efficiency with regard to data collection and methodology, increased cost-effectiveness of assessment activities in view of shared resources, and increased coherence of assessment results in view of shared methodology and risk characterization. Decisions should be based on

233


Summary


all data available and searching for communalities. This will prevent risk reduction or prioritisation decisions which do not protect all human and environmental targets, or even increase the risk to specific targets. In registering chemicals under REACH, it is strongly recommended that industry will interact with regulatory bodies and other stakeholders in a problem formulation approach. The intensity of such communication, especially at registration, should be proportional to the complexity of the Chemical Safety Assessment, the degree of risk foreseen and the degree of uncertainty due to lack of knowledge. Further demonstrations of the scientific, economic and regulatory benefits of the IRA approach are needed. It is recommended to perform cost-benefit studies to this regard. The integration of information could also help to develop testing strategies with the aim of avoiding vertebrate testing where possible. An important aim of REACH, the reduction of the use of experimental animals, is an important driver towards integrated approaches combining knowledge from in vivo, in vitro, in silico experiments, exposure data and information on mechanisms of action across species. Further work is needed to investigate when and how the total Weight-of Evidence from such integration will show a positive balance with regard to costs and animals saved. The uncertainty management scheme can be regarded as an additional layer in the IRA framework aiming to improve the management and communication of uncertainty in IRA decision-making processes. This IRA+ framework stimulates interactions between risk assessors, decision-makers and stakeholders and facilitates the analysis and prioritisation of uncertainties, supported by robust tools for a tiered integrated risk and uncertainty assessment. This approach is also needed to support decision-making on the basis of the Precautionary Principle. The Precautionary Principle should be considered as a risk management tool supported by a risk assessment framework which includes a scientific evaluation, as complete as possible, and an identification of the degree of uncertainty at each stage as proposed in the IRA+-framework. Conclusions and recommendations on methodology: The risk assessment tool EUSES has been shown to be fit for purpose under REACH with some modifications, specifically with regard to the implementation of the Exposure Scenario concept and more risk reduction options. More work needs to be done to increase the domain of applicability, in view of the limited use of EUSES outside the domain of the persistent, non-dissociating substances of intermediate lipophilicity.

234

Summary

It has also been shown that a probabilistic risk assessment (PRA) in the EU framework for risk assessment of industrial chemicals is feasible with currently available techniques. PRA potentially gives better decision-support to risk managers, because it gives more quantitative insight into the range of possible outcomes in a risk assessment and the degree of cumulated conservatism in the deterministic risk assessment. Sensitivity analysis is able to reveal the relative impact of uncertainties in parameters on the final result, where the risk assessment can be improved in the most time- and cost-efficient manner and whether it is necessary and achievable to reduce the uncertainty further. The examples discussed show a high contribution to the overall quantifiable uncertainty from uncertainty and variability in emission estimation and in the application of default assessment factors. For a further implementation of PRA, it is necessary to build up experience. If the tiered REACH Guidance is followed, PRAs on more substances will become available. Important research areas are the characterisation of both default and chemical-specific distributions used for exposure and effects input parameters and methods to address environmental and human variability in the REACH CSA and to separately address variability, based on experimental data, and uncertainty due to lack of knowledge. The different stages of uncertainty analysis and PRA would be greatly facilitated if firmly integrated with risk assessment tools like EUSES. In any uncertainty analysis a clear separation needs to be made between quantifiable and non-quantifiable uncertainties and between true uncertainty (lack of knowledge) and variability to be able to answer different risk questions. Secondly, it should be made very clear which uncertainties are included in the assessment and which not. In risk assessment, an important tool for characterizing and reducing uncertainty due to lack of knowledge is the Weight-of-Evidence (WoE) process, usually carried out by experts. Unfortunately, formal procedures and consistent terminology for WoE-processes are lacking. As shown, international efforts are ongoing to improve this situation and it is strongly recommended to support these initiatives. An important difficulty in the application of this process and methodology is the knowledge gap: both risk assessors and risk managers often are unfamiliar with these new approaches and have to learn new skills. This should first of all be solved by training of both risk assessors and risk managers. It further requires a strong IRA+-process with a tiered approach, graphical support and high communication skills of both experts and risk managers. The guidance available should be extended by showing real-life applications, preferably prepared in regulator-industry-stakeholder expert working groups. Experts and

235


Summary


risk managers need to investigate together how to use the results of uncertainty analysis in decision-making and how to communicate these in a transparent way. If process and methodology follow the direction shown, decision-support will be more transparent, will lead to less communication problems and will improve the trust between various parties involved. Decisions which fully take into account the uncertainties in the assessments performed, including the influence of divergent opinions and assumptions of experts and stakeholders, will be better informed and will lead to transparent decisions which can be communicated in a clear way.

236

Samenvatting

Samenvatting


Samenvatting

Chemicaliën zijn belangrijk in ons dagelijkse leven. De chemische industrie is op wereldschaal de op twee na belangrijkste sector. Chemicaliën dragen bij aan ons welzijn, maar kunnen ook schadelijk zijn voor mens en milieu door blootstelling bij productie, gebruik en in het afvalstadium. Het risicobeheer van chemicaliën omvat alle wettelijke en vrijwillige maatregelen om de risico’s te verminderen. In dit proces worden risicobeheerders zowel bij de overheid als in de industrie geadviseerd door deskundigen op verschillende expertisegebieden. Risicobeoordeling is er daar een van. Wat betreft chemicaliën kan ‘risico’ gedefinieerd worden als de waarschijnlijkheid van schadelijke effecten in een organisme, systeem of (deel)populatie door blootstelling aan een stof onder specifieke omstandigheden. Zowel risicobeheerders als hun deskundige adviseurs erkennen alom dat er onzekerheden kleven aan risicobeoordelingen. Dit proefschrift concentreert zich op onzekerheid en variabiliteit in de risicobeoordelings methodologie voor industriële chemicaliën binnen het huidige wettelijke kader daarvoor in Europa: REACH (Registratie, Evaluatie, Autorisatie en Restrictie van Chemicaliën). De besproken methodieken bestrijken de risicobeoordeling voor zowel de mens als voor het milieu. Het doel van dit proefschrift is te onderzoeken hoe het wetenschappelijke proces van risicobeoordeling de besluitvorming kan verbeteren door rekening te houden met de onzekerheden die hieraan verbonden zijn. Een belangrijk onderdeel van deze besluitvorming is de overweging of uit voorzorg maatregelen genomen moeten worden. Het onderzoek bouwt voort op twee theoretische kaders: het IPCS/WHO denkkader voor geïntegreerde risicobeoordeling (Integrated Risk Assessment, IRA) en het denkkader voor het omgaan met onzekerheden van Walker et al. (2003). Het IRA denkraam is ontwikkeld om de kwaliteit en efficiëntie van de bepaling van de risico’s van chemicaliën, fysische en andere milieudrukfactoren te verbeteren en om de ondersteuning van de besluitvorming vollediger en meer samenhangend te maken. De achterliggende gedachte is dat zowel de wetenschappelijke discussie als de daaropvolgende beleidsmaatregelen kunnen profiteren van een meer geïntegreerde, interdisciplinaire aanpak waarin informatie vollediger is en wordt gedeeld en waarin onzekerheden worden verminderd. Het denkraam voor omgaan met onzekerheden sluit goed hierbij aan en is vooral bruikbaar omdat het aandacht vraagt voor de verschillende typen onzekerheden in de risicobeoordeling. Hoofdstuk 1 leidt het onderwerp in en legt de twee theoretische kaders uit. Het geeft de historische en wettelijke context van risicobeheer en risicobeoordeling van chemicaliën op UN-, OECD- en EU-niveau. REACH wordt kort beschreven evenals een van de basisprincipes: het Voorzorgbeginsel. Uitgelegd wordt dat dit Voorzorgbeginsel kan worden

239


Samenvatting


toegepast wanneer er aanzienlijke wetenschappelijke onzekerheid is over de causaliteit, grootte en aard van een gevaar en dat in een dergelijk situatie een wetenschappelijke analyse met een grondige beschouwing van de onzekerheden altijd noodzakelijk is. Dit speelt vooral wanneer er een gegronde reden tot zorg is om een potentieel risico dat niet, of niet op tijd, met voldoende zekerheid kan worden vastgesteld. Het gaat dan om de betrouwbaarheid van de beoordeling, de overblijvende onzekerheden en een correcte classificatie van die onzekerheden. Hoofdstuk 2 laat zien hoe een IRA ingevuld kan worden. IRA is hier toegepast op een typische boerengemeenschap waarin organofosfaten als bestrijdingsmiddel wordt gebruikt. Dit voorbeeld is ontwikkeld om de IRA-aanpak te communiceren en demonstreren en om de voor- en nadelen te onderzoeken. Duidelijk wordt gemaakt dat de geïntegreerde aanpak zowel de risicobeheerder als de risicobeoordelaar helpt om het probleem breder te bekijken. Vooral de eerste stap, de probleemdefinitie, is belangrijk om informatie te delen en aan te geven welke informatie en welk advies nodig is voor een optimaal besluit. Probleemdefinitie is een proces waarin de risicobeoordeling wordt afgebakend en gepreciseerd en het plan van aanpak voor het onderzoek naar en de karakterisering van de risico’s worden gemaakt. De methodologie die in Hoofdstuk 2 is toegepast is gebaseerd op methodologie die in Europa is ontwikkeld en geharmoniseerd. Ze is beschreven in EU Technische Richtlijnen en verwerkt in een PC-applicatie, het ‘European Union System for the Evaluation of Substances’ (EUSES). Hoofdstuk 3 beschrijft de 2004 versie van EUSES, terwijl de aanpassingen die noodzakelijk geacht worden voor haar volledige toepassing onder REACH in Hoofdstuk 7 worden besproken. Uit een discussie over de validatiestatus van EUSES blijkt dat de modellen ‘state-of-the-art’ zijn en deterministische, conservatieve schattingen geven voor standaard scenario’s op basis van beperkte data-eisen. In de EU risicobeoordeling van industriële chemicaliën worden de risico’s gewoonlijk gekarakteriseerd door middel van een deterministische vergelijking van de blootstellings concentratie of dosis met een geen-effect niveau. Onzekerheden worden impliciet meegenomen door conservatieve aannames in deze methodologie. Hoofdstuk 4 geeft aan hoe een beschouwing van die onzekerheden, die door verdelingen kunnen worden gekenschetst, in een probabilistische risicoschatting (PRA) het risicobeheer kan verbeteren. Voor- en nadelen van PRA worden besproken en inzichtelijk gemaakt met een voorbeeld.

240

Samenvatting

Hoofdstuk 5 vervolgt de discussie over voor- en nadelen van PRA aan de hand van de EU risicobeoordeling van dibutylftalaat, een weekmaker voor plastic. Dit voorbeeld laat zien hoe verschillende onzekerheden in een risicobeoordeling kunnen worden meegenomen en met elkaar kunnen worden vergeleken. Het gaat om onzekerheden in zowel de blootstellingschatting als in de effectbeoordeling. Belangrijk zijn daarbij de onzekerheden in de assessmentfactoren zoals die worden gebruikt bij de extrapolatie van proefdieren naar de gevoelige individuen in de algemene bevolking. Het wijdverbreide gebruik van assessmentfactoren in de afleiding van geen-effect niveaus voor de mens en ecotoxicologische geen-effect niveaus wordt diepgaand besproken in Hoofdstuk 6 voor wat betreft de assessmentfactoren voor de mens. Tevens wordt in dit hoofdstuk aandacht gegeven aan de probabilistische afleiding van een veilig niveau voor de mens op basis van Benchmark modellering zoals toegepast in Hoofdstuk 5. Hoofdstuk 7 tot slot bevat een discussie over de relevante methodologische aspecten van risicobeoordeling en onzekerheidsanalyse uit dit proefschrift. De EU risicobeoordelings methodologie van EUSES wordt vergeleken met de nieuwe REACH Leidraad. Hoewel EUSES gebaseerd is op de Technische Richtlijnen zoals die operationeel waren tot midden 2008 onder de pre-REACH EU-wetgeving, komt veel ervan nog overeen met de nieuwe REACH Leidraad. Hoofdstuk 7 geeft de belangrijkste, noodzakelijk geachte veranderingen en de implicaties daarvan voor EUSES. Het hoofdstuk laat verder zien hoe de REACH risicobeoordeling kan profiteren van het IRA-denkkader en hoe dit IRA-denkkader op haar beurt verbeterd kan worden door de nieuwste inzichten over de rol van onzekerheidsanalyse in risicobeoordeling en risicobeheer. Het omgaan met onzekerheden, inclusief een methode om deze te classificeren en inclusief een getrapte benadering voor onzekerheidsanalyse, kan als extra laag ingebouwd worden in het IRA denkkader. Dit versterkt het omgaan met, en de communicatie over onzekerheid in het besluitvormingsproces. De belangrijkste algemene conclusie van dit proefschrift is dat zowel het proces als de methodologie voor de risicobeoordeling als beslissingsondersteunend instrument onder REACH kan worden verbeterd. Het proces kan verbeterd worden door de toepassing van het denkkader voor geïntegreerde risicobeoordeling met een sterke onzekerheidsanalyse component (IRA+). De methodologie kan verbeteren door een getrapte aanpak van de onzekerheidsanalyse, beginnend met een eenvoudige deterministische aanpak en, indien nodig, classificatie en prioritering van onzekerheden en probabilistische risicobeoordeling (PRA). Dit proefschrift laat zien dat zowel kwalitatieve als kwantitatieve instrumenten hiervoor beschikbaar zijn.

241


Samenvatting


Conclusies en aanbevelingen over het proces: Als belangrijkste voordelen van IRA worden gezien: een efficiëntere risicobeoordeling voor wat betreft het verzamelen van gegevens en de inzet van methodologie, en een toename in de kosteneffectiviteit van risicobeoordelingen. Dat laatste komt vooral doordat de beschikbare middelen breed worden ingezet en door een grotere samenhang in de beoordelingsresultaten als gevolg van op elkaar afgestemde methodologie en risicokarakterisering. Beslissingen moeten op basis van alle beschikbare gegevens genomen worden en de samenhang daarin moet worden gezocht. Dit voorkomt dat besluiten over risicoreductie of prioritering worden genomen die niet alle beschermingsdoelen dekken, of zelfs het risico voor bepaalde beschermingsdoelen vergroten. Het verdient sterk de aanbeveling dat bij de registratie van chemicaliën onder REACH, de industrie de IRA aanpak van de probleemdefinitie volgt in interactie met overheden en andere belanghebbenden. De wetenschappelijke, economische en beleidsvoordelen van de IRAbenadering moeten nog beter worden aangetoond. Aanbevolen wordt hierbij om kostenbaten studies uit te voeren. De integratie van de informatie kan ook bijdragen aan de ontwikkeling van teststrategieën die het proefdiergebruik kunnen verminderen. Deze doelstelling van REACH is dan ook een belangrijke drijfveer voor de toepassing van geïntegreerde benaderingen met inzet van in vivo, in vitro, in silico en blootstellinggegevens, evenals van kennis over werkingsmechanismen. Nader onderzoek moet uitmaken wanneer en hoe de totale bewijslast van een dergelijke integratie positief uitpakt voor wat betreft vermindering van kosten en proefdieren. Het omgaan met onzekerheden kan als extra laag worden ingebouwd in het IRA denkkader. Dit zal het omgaan met en de communicatie over onzekerheid in het besluitvormings proces versterken. Dit IRA+ denkkader bevordert de interacties tussen risicobeoordelaars, beslissers en belanghebbenden en faciliteert de analyse en prioritering van onzekerheden, gesteund door een getrapte en geïntegreerde beoordeling van de risico’s en onzekerheden. Deze aanpak is ook nodig voor besluitvorming op basis van het Voorzorgbeginsel. Het Voorzorgbeginsel is een risicobeheerinstrument dat gesteund wordt door een risicobeoordeling met een zo volledig mogelijke wetenschappelijke evaluatie en een identificatie van de mate van onzekerheid in elke stap, zoals wordt voorgesteld in het IRA+-. raamwerk.

242

Samenvatting

Conclusies en aanbevelingen over de methodologie: Aangetoond is dat EUSES geschikt is als risicobeoordelingsmethode onder REACH. Enige aanpassingen zijn wel nodig, vooral in de inpassing van het REACH Blootstellingscenario concept en van opties voor risicoreductie. Nader onderzoek is ook nodig om het toepasbaarheidsgebied van EUSES voor chemicaliën te vergroten buiten het domein van de persistente, niet dissociërende verbindingen met intermediaire lipofiliciteit. Het onderzoek heeft ook laten zien dat een probabilistische risicobeoordeling (PRA) voor industriële chemicaliën mogelijk is met bestaande technieken. Een PRA kan betere beslissingsondersteunende informatie aan risicobeheerders geven, omdat het een meer kwantitatief inzicht geeft in de mogelijke uitkomsten van de risicobeoordeling. Ook geeft het inzicht in de mate van conservatisme in deterministische schattingen. Een gevoeligheidsanalyse kan zichtbaar maken wat de relatieve invloed van elke onzekerheid op het eindresultaat is, waar de risicobeoordeling het meest effectief kan worden verbeterd qua tijd en kosten en of het wel nodig is om de onzekerheid verder te reduceren. De voorbeelden in dit proefschrift tonen aan dat de totale kwantificeerbare onzekerheid vooral wordt bepaald door onzekerheid en variabiliteit in de emissieschatting en in de toepassing van assessmentfactoren. Meer ervaring met PRA is nodig voor bredere toepassing. Als de REACH richtlijnen worden opgevolgd zullen en meer PRA’s beschikbaar komen. Belangrijke terreinen van onderzoek hierbij zijn de karakterisering van default en stofspecifieke verdelingen voor blootstelling- en effectparameters, methoden om variabiliteit in mens en milieu onder REACH mee te nemen in de beoordeling en methoden om variabiliteit te scheiden van onzekerheid door gebrek aan kennis. De uitvoering van de verschillende stappen in de onzekerheidsanalyse kan veel makkelijker worden als ze ingebouwd worden in beoordelingsinstrumenten zoals EUSES. Om verschillende risicovragen te kunnen beantwoorden is het essentieel om kwantificeerbare en niet-kwantificeerbare onzekerheden enerzijds en onzekerheden door variabiliteit en gebrek aan kennis anderzijds duidelijk uit elkaar te houden. Verder moet altijd helder worden gemaakt welke onzekerheden in de beoordeling worden meegenomen en welke niet. In het veld van de risicobeoordeling is het afwegen van de bewijslast (‘Weight-ofEvidence’) een belangrijk proces om onzekerheden te karakteriseren en verminderen. Doorgaans is dit het terrein van deskundigen. Helaas ontbreken formele procedures

243


Samenvatting


en een consistente terminologie. Aangegeven is dat er internationaal pogingen worden ondernomen om hierin verbetering te brengen en dergelijke initiatieven worden dan ook sterk aanbevolen. Zowel risicobeoordelaars en risicobeheerders zijn vaak niet vertrouwd met deze nieuwe benaderingen. Deze kenniskloof is een belangrijke hinderpaal bij de toepassing van aanpak en methodologie voor onzekerheidsanalyse. Dit vereist op de eerste plaats training, Bovendien vraagt dit om een sterk IRA+ proces met een getrapte benadering, grafische ondersteuning, goede communicatie door deskundigen en training van de betrokken niet-deskundigen. De beschikbare richtlijnen moeten worden uitgebreid met realistische toepassingen die bij voorkeur samengesteld worden door een werkgroep van deskundigen van overheid, industrie en andere belanghebbenden. Deskundigen en risicobeheerders moeten onderzoeken hoe ze de resultaten van onzekerheidsanalyse in de besluitvorming op een transparante manier kunnen gebruiken en communiceren. Als bovenstaand proces en de methodologie worden gevolgd kan de besluitvorming transparanter en met minder communicatieproblemen plaatsvinden. Het vertrouwen tussen partijen zal hierdoor ook verbeteren. Besluiten zijn beter onderbouwd, transparanter en beter communiceerbaar naarmate ze in de risicobeoordeling meer rekening houden met onzekerheden met inbegrip van uiteenlopende meningen en aannames van deskundigen en belanghebbenden.

244

Curriculum vitae

Curriculum vitae


Curriculum vitae

Theodorus Gabriël (Theo) Vermeire (1953) studied chemistry and toxicology at the University of Utrecht. He received his MSc and teaching qualifications in 1978. Together with his wife Anita, he lived in Zambia for three years, working as teacher chemistry. Following this valuable experience, he started his career in risk assessment as toxicologist at the Ministry of Housing, Physical Planning and the Environment contributing to projects of the WHO International Programme on Chemical Safety (IPCS) and UNEP International Register of Potentially Toxic Chemicals (currently: UNEP Chemicals). In 1987, he was employed by the National Institute for Public Health and the Environment (RIVM) and has served in a good number of scientific and managerial functions up to this day. Major projects were the development of the Netherlands’ Uniform System for the Evaluation of Substances (industrial chemicals, plant protection products and biocides) and the European Union System for the Evaluation of Substances (industrial chemicals and biocides). His present position at RIVM is deputy head of the RIVM Expertise Centre for Substances SEC. He manages a group within SEC preparing risk assessments for mainly industrial chemicals and working on risk assessment methodology. As an expert with a wide knowledge on toxicology and risk assessment, he has been involved in many expert groups developing guidance and tools for risk assessment (e.g. for IPCS/WHO, EU, OECD) and in a substantial number of training courses in this area in and outside Europe. Recently, he has co-edited, and contributed to, the standard risk assessment volume “Risk assessment of Chemicals, an Introduction”. He is a member of the Scientific Committee of the European Environment Agency, the Scientific Committee on Emerging and Newly Identified Health Risks (SCENIHR) of the European Commission and the WHO/IPCS Steering Group for the Harmonisation of Approaches to the Assessment of Risk from Exposure to Chemicals. He is editor of the journal Human and Ecological Risk Assessment. Last but not least, he is the proud father of three sons.

Utrecht, 30th of March 2009

247


Curriculum vitae






Toxicology Advisory Centre (1997- 2008) Several hundreds of confidential advisory reports within the scope of the Dutch Chemical Substances Act ((author, co-author or Member of Advisory Board). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. IPCS (1984-1991) Draft Environmental Health Criteria tetrachloroethylene (31), methylene chloride (32), epichlorohydrin (33), ethylene oxide (55), propylene oxide (56), 1,2-dichloroethane (62), hydrazine (68), 1-propanol (102), 2-propanol (103), PCB’s (chapter on mammalian toxicology), acrolein (127), hexachlorobutadiene (156). WHO/ UNEP/ILO, International Programme on Chemical Safety, Geneva, Switzerland. Van der Heijden CA, Vermeire TG, Van de Wiel HJ, Van Jaarsveld. 1987. Industriële uitworp te Oss. Onderzoek naar de mogelijke gevolgen voor de volksgezondheid. Report No. 748701002. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van de Wiel HJ, Knaap AGAC, Van Apeldoorn M, Vermeire TG, Bos HP. 1987. Onderzoek naar de uitworp van een PVC-verwerkend bedrijf te Huizen en de mogelijke gevolgen hiervan voor de volksgezondheid. Report No. 748704004. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Kliest JJG, Vermeire TG, Van de Wiel HJ, Van Staden JJ, Bos HP, De Boer JLM, Knaap AGAC. 1988. Onderzoek naar de emissies van een thermische grondreinigingsinstallatie en de mogelijke gezondheidskundige gevolgen voor de omwonenden. Report No.718703001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. De Nijs ACM, Knoop JM, Vermeire TG. 1988. Risk assessment of new chemical substances. System realisation & validation. Report No. 718703001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. De Nijs ACM, Vermeire TG. 1990. Soil-plant and plant-mammal transfer factors. Report No. 670203001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Toet C, De Nijs ACM, Vermeire TG. 1991. Risk assessment system of new chemicals; system realization and validation II. Report No. 679102005. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands.

251




Vermeire TG, De Vries T, Speijers GJA, Wester PW, De Liefde A, Beekhof PK, Van der Heijden CA. 1989. OECD International collaborative study on acute toxicity testing. Report no. 618807001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Van der Heijden CA. 1990. Health assessment of hazardous air pollutants, The Netherlands. Toxicol Ind Health 6:235-243. Vermeire TG, Van Apeldoorn ME, De Fouw JC, Jansen PJCM. 1991. [Proposal for the human-toxicological derivation of C-values for soil pollution cases]. Report No. 725201005 (in Dutch). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire T. 1991. Hazard and risk assessment methodologies for human health. In: Proceedings of the 3rd US-Dutch expert workshop on comparative risk analysis - approaches for air pollution prevention, June 11-14, 1991, Bellevue (Seattle). Washington, USA. Vermeire TG, Van Iersel A, De Leeuw FAAM, Peijnenburg WJGM, Van der Poel P, Taalman R, Toet, C. 1992. Initial assessment of the hazards and risks of new chemicals to man and the environment. Report No. 679102006. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Kroese ED, Meijer GW, Mohn GR, Notenboom J, Peijnenburg WJGM, Piersma AH, Roghair CJ. 1993. Initial assessment of the hazards and risks of new chemicals to man and the environment, Part II. Report No. 679102018. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Van der Poel P, Van de Laar RTH, Roelfzema H. 1993. Estimation of consumer exposure to chemicals: application of simple models. Sci Total Environ 136:155-176. Vermeire TG, Van Iersel AAJ, De Leeuw FAAM, Peijnenburg WJGM, Van der Poel P, Taalman RDFM, Toet C. 1993. Initial assessment of the hazards and risks of new chemicals to man and the environment. Sci Total Environ (Supplement 2): 1597-1615. De Nijs TCM, Toet C, Vermeire TG, Van der Poel P, Tuinstra J. 1993. Dutch Risk Assessment system for New Chemicals, “DRANC”. Sci Total Environ (Supplement 2):1729-1748. 252


Vermeire TG. 1993. Voorstel voor de humaan-toxicologische onderbouwing van C-(toetsings)waarden. Report No. 715801001, addendum to Report No. 725201005 (in Dutch). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Van der Zandt PTJ, Roelfzema H, Van Leeuwen CJ. 1994. Uniform System for the Evaluation of Substances I; principles and structure. Chemosphere 29:23-38. Jager DT, Vermeire TG, Slooff W, Roelfzema H. 1994. Uniform System for the Evaluation of Substances II; effects assessment. Chemosphere 29:319-335. Vermeire T, Jager T. 1993. The Uniform System for the Evaluation of Substances, USES 1.0. Annual Scientific Report. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire T, Van Veen M, Van Leeuwen R. 1993. Risk assessment for man. Annual Scientific Report. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van Veen MP, Vermeire TG, Olling M. 1994. Consumentenblootstelling: een overzicht van blootstellings- en opnamemodellen. Report No. 612810001 (in dutch). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Van der Zandt P, Vermeire T. 1994. Uniform oordelen over risico’s van stoffen. Chemisch Magazine, juli/augustus 1994. Van der Zandt P, Vermeire T. 1994. Uniform System for the Evaluation of Substance: a decision-supporting tool in risk management. EUROTOX Newsletter 18(1):5-6. Van de Meent D, De Bruijn JHM, De Leeuw FAAM., De Nijs ACM, Vermeire TG. 1995. Exposure modelling. In: Van Leeuwen CJ, Hermens J (eds). Risk assessment of chemicals, an introduction. Kluwer Scientific Publishers, Dordrecht, The Netherlands. Vermeire TG, Van der Zandt PTJ. 1995. Chemical risk assessment and evaluation. In: Van Leeuwen CJ, Hermens J (eds) Risk assessment of chemicals, an introduction. Kluwer Scientific Publishers, Dordrecht, The Netherlands.

253




Vermeire TG, Van Veen MP. 1995. De schatting van de blootstelling van de mens aan stoffen en straling: definitierapport. Report No. 601132001 (in dutch). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Gingnagel P, Vermeire TG. 1995. Screening of chemical substances; application of the Uniform System for the Evaluation of Substances, USES 1.0. Report No. 601014009. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Van Veen MP, Jager DT. 1996. Human exposure assessment: consumer exposure and exposure through the environment. In: Vollmer G, Giannoni L, SokullKlüttgen B, Karcher W (eds). Risk assessment, theory and practice, Proceedings of the Workshop Risk assessment: a workshop on practical experience, 27-28 March, 1996. European Chemicals Bureau, Ispra, Italy. Hulzebos EM, Vermeire TG. 1996. Risk characterisation for the environment. In: Vollmer G, Giannoni L, Sokull-Klüttgen B, Karcher W (eds). Risk assessment, theory and practice, Proceedings of the Workshop Risk assessment: a workshop on practical experience, 27-28 March, 1996. European Chemicals Bureau, Ispra, Italy. Guinée J, Heijungs R, van Oers L, Van de Meent D, Vermeire T, Rikken M. 1996. LCA impact assessment of toxic substances. CML/RIVM-rapport (in dutch), Publikatiereeks Produktenbeleid nr. 1966/21.VROM, Den Haag. Vermeire TG, Jager DT, Bussian B, Den Haan K, Hansen B, Lundberg I, Niessen H, Robertson S, Tyle H, Van der Zandt PTJ. 1997. European Union System for the Evaluation of Substances, EUSES, Principles and Structure. Chemosphere 34:1823-1836. Guinée JB, Heijungs R, Van Oers LFCM, Sleeswijk AW, Van de Meent D, Vermeire TG, Rikken M. 1997. USES, Uniform System for the Evaluation of Substances, Inclusion of fate in LCA characterisation of toxic releases applying USES 1.0. Int J LCA 1:133-138. Van Leeuwen CJ, Bro-Rasmussen F, Feijtel TCJ, Arndt R, Bussuan BM, Calamari D, Glynn P, Grandy NJ, Hansen B, Van Hemmen J, Hurst P, King N, Koch R, Müller M, Solbé J, Speijers GAB, Vermeire T. 1996. Risk assessment and management of new and existing chemicals. Environ Toxicol Pharm 2:243-299.

254


Vermeire TG, Van Veen MP, Janssen MPM, Smetsers, RCGM. 1997. De schatting van de blootstelling van de mens aan stoffen en straling, De status van het RIVM-onderzoek. Report No. 601132002 (in dutch). National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire T, Linders J. 1997. Uniform System for the Evaluation of Substances (USES) Hazard comparison between substances. In: Lee SD, Schneider T (eds). Proceedings of the 4th US-Dutch International Symposium: Comparative risk analysis and priority setting for air pollution issues, June 7-11, Keystone, Colorado. Pittsburgh, Pennsylvenia, USA. Vermeire TG, Stevenson H, Pieters MN, Rennen M, Slob W, Hakkert BC. 1998. Assessment factors for human health risk assessment: a discussion paper. Report No. 620110007/TNO report No. V97.880. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Vermeire TG, Stevenson H, Pieters MN, Rennen M, Slob W, Hakkert BC. 1999. Assessment factors for human health risk assessment: a discussion paper. Crit Rev Toxicol 29(5):439-490. Vermeire TG, Bos P, van Haelst A, Pieters M, Pronk M, Van Raaij M, Rennen M.. 1999b. Report of the Joint EU/RIVM/TNO Workshop on Interpretations of Margins Of Safety in Human Health Risk Assessment, 21-22 April 1999b. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. August 30, 1999. Jager T, Den Hollander HA, Janssen GB, Van der Poel P, Rikken MGJ, Vermeire TG. 2000. Probabilistic risk assessment for new and existing chemicals: Sample calculations. RIVM-report No. 679102049. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Sekizawa J, Suter II G, Vermeire T, Munns W. 2000. An example of an integrated approach for health and environmental risk assessment: the case of organotin compounds. Water Sci Technol 42:305-313. Jansson B, Vighi M, Utne Skaare J, Vermeire T, Costello M. 2000. Exposure data in risk assessments. C2/JCD/csteeop/ExpAssess7032001/D(01). WG of the Scientific Committee on Toxicity, Ecotoxicity and the Environment (CSTEE). Brussels, Belgium.

255




Vermeire T, Pieters M, Rennen M, Bos P. 2001. Probabilistic assessment factors for human health risk assessment: a practical guide. RIVM Report No. 601516005, TNO Report No. V3489. National Institute for Public Health and the Environment (RIVM), Bilthoven; TNO, Zeist, The Netherlands. www.rivm.nl/bibliotheek/rapporten. Jager T, VermeireTG, Rikken, MG, van der Poel P. 2001. Opportunities for a probabilistic risk assessment of chemicals in the European Union. Chemosphere 43: 257-64. Vermeire T, Jager T, Janssen G, Bos P, Pieters M. 2001. A probabilistic human health risk assessment for environmental exposure to dibutylphthalate. Hum Ecol Risk Assess 7: 1663-1679. Jager T, Den Hollander H, Van der Poel, P, Rikken MGJ, Vermeire T. 2001b. A probabilistc environmental risk assessment for dibutylphthalate (DBP). Hum Ecol Risk Assess 6:16811697. Vermeire T. 2001. Case study information package, Organophosphorous pesticides in the environment. Presentation at SETAC-USA, Platform session on Integrated Ecological and Human Health Risk Assessment, Baltimore, 13 November, 2001. Bodar C, De Bruijn J, Vermeire T, Van der Zandt P. 2002. Trends in risk assessment of chemicals in the European Union. Hum Ecol Risk Assess 8:1825-2002. Bodar CWM, Berthault F, De Bruijn JHM, Van Leeuwen CJ, Pronk, MEJ, Vermeire TG. 2002. Evaluation of EU Risk Assessments Existing Chemicals (EC Regulation 793/93). Report No. 601504002. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Also published in 2003 in Chemosphere 53:10391047. Vermeire, TG, Bodar CWM. 2002. Computer-aided modelling in risk assessment. Toxicol Lett 135:S1. Suter II GW , Vermeire T, Munns Jr W, Sekizawa, J. 2003. Framework for the integration of health and ecological risk assessment. Hum Ecol Risk Assess 9:281-301. Vermeire T, MacPhail R, Waters M. 2003. Integrated human and ecological risk assessment: a case study of organophosphorous pesticides in the environment. Hum Ecol Risk Assess 9: 343-357.

256


Vermeire T, Rikken M, Attias L, Boccardi P, Boeije G, de Bruijn J, Brooke D, Comber M, Dolan B, Fischer S, Heinemeyer G, Koch V. Lijzen J, Müller B, Murray-Smith R, Tadeo J. 2004. European Union System for the Evaluation of Substances, The second version. Chemosphere 59: 473-485. Bosgra S, Bos, PMJ, Vermeire TG, Luit RJ, Slob W. 2005. Probabilistic risk characterization: an example with di(2-ethylhexyl)phthalate. Regul Toxicol Pharm 43:104-113. Vermeire T, Munns Jr WR, Sekizawa J, Suter II GW, Van der Kraak G. 2007. An Assessment of Integrated Risk Assessment. Hum Ecol Risk Assess 13:339-354. Vermeire TG, Aldenberg T, Dang Z, Janer G, de Knecht JA, van Loveren H, Peijnenburg WJGM, Piersma AH, Traas TP, Verschoor AJ, van Zijverden M, Hakkert B. 2007. Selected Integrated Testing Strategies (ITS) for the risk assessment of chemicals. Report No. 601050001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands. Janer G, Hakkert BC, Slob W, Vermeire T, Piersma AH. 2007. A retrospective analysis of the two-generation study: What is the added value of the second generation? Reproductive Toxicol 24:97–102. Janer G, Hakkert BC, Piersma AH, Vermeire T, Slob W. 2007. A retrospective analysis of the added value of the rat two-generation reproductive toxicity study versus the rat subchronic toxicity study. Reproductive Toxicol 24:103–113. Van Leeuwen CJ, Vermeire TG (eds). 2007. Risk assessment of chemicals: an introduction, Second edition. Springer Dordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book). Van Engelen JGM, Hakkinen PJ, Money C, Rikken MGJ, Vermeire TG. 2007 Human exposure assessment. In: Van Leeuwen CJ, Vermeire TG (eds). 2007. Risk assessment of chemicals: an introduction, Second edition. Springer Dordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book).

257




Vermeire TG, Baars AJ, Bessems JGM., Blaauboer BJ, Slob W, Muller JJA. 2007. Toxicity testing for human health risk assessment. In: Van Leeuwen CJ, Vermeire TG (eds). 2007. Risk assessment of chemicals: an introduction, Second edition. SpringerDordrecht, The Netherlands. ISBN 978-1-4020-6101-1 (handbook), ISBN 9781-4020-6102-8 (e-book). Zweers PGPC., Vermeire TG. 2007. Data: needs, availability, sources and evaluation. In: Van Leeuwen CJ, Vermeire TG (eds). 2007. Risk assessment of chemicals: an introduction, Second edition. SpringerDordrecht, The Netherlands. ISBN 978-1-40206101-1 (handbook), ISBN 978-1-4020-6102-8 (e-book). Money CD, Van Hemmen JJ, Vermeire TG. 2007. Scientific governance and the process for exposure scenario development in REACH. J Expo Sci Env Epid 17 (Supplement 1):S34-S37. Vermeire TG, Bakker J, Bessems JGM, Van de Bovenkamp M, Dang Z, Van Engelen JGM, Gunnarsdottir S, Hagens WI, Links I, Marquart H, Mikkers J, Van Zijverden M. 2008. Exposure informed testing under REACH. Report No. 601017001. National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands.

258

Acknowledgements

Acknowledgements


Acknowledgements

For many years in my professional career a good number of people have encouraged me to write this thesis. Whereas I am grateful to all for this confidence in me, I am also convinced this is the right moment in time, since I am now able to integrate risk assessment theory with valuable experience and practice. Having said this, it is also clear that the theory and practice presented in this book is the result of collaborative scientific and technical input from numerous colleagues in and outside RIVM with whom I had the privilege to work in very challenging projects. I would like to thank all of them. Specifically, I would like to thank those with whom I worked on the first approaches towards risk assessment methodologies which finally lead up to the EU Technical Guidance and its implementation EUSES, especially Tjalling Jager, Dik van de Meent, Ton de Nijs, Paul van der Poel, Mathieu Rikken, Jaap Struijs and Peter van der Zandt. Many thanks also to Guido Hummelen and Peter Tanis of TSA Delft for their skilful implementation work. I also acknowledge the EU EUSES Working Group (see Chapter 3) for their critical evaluations and advice. It was also great working on the review and quantification of human assessment factors with my RIVM and TNO colleagues Peter Bos, Betty Hakkert, Geert Janssen, Moniek Pieters, Monique Rennen, Wout Slob and Hanzen Stevenson. Thank you all. I also acknowledge the work of the Planning Group on Approaches to Integrated Risk Assessment of the WHO International Programme on Chemical Safety. I very much enjoyed the critical and cooperative spirit in this group and would like to give my special thanks here to Bob MacPhail, Wayne Munns Jr., Jun Sekizawa, Glenn Suter II and Glen van der Kraak. It is sad I cannot say thank you to Bob Kroes, who also was a great stimulating force, but unfortunately passed away too soon. I would like to thank my promoters Willem Seinen en Wout Slob for their enthusiasm, encouragement and the valuable discussions which helped me a lot in creating this thesis. Thanks also to the Reading Commission for spending their time to the evaluation of my manuscript: Jack de Bruijn, Rolaf van Leeuwen, Kees van Leeuwen, Dik van de Meent and Glenn Suter II. I thank Marc Sprenger, the Director General of RIVM for giving me the opportunity to work on this thesis. Priceless was the help I received from my dear colleagues Charles Bodar, Tjalling Jager, Hans Könemann, Jan Roels, Dick Sijm and Theo Traas. Tjalling, special thanks to you for the great cooperation we had, your very valuable scientific input, and the critical comments and good advice you managed to give me on the manuscript in spite of your severe illness. All my colleagues at SEC and SIR: thank you for your moral support and your interest in this project. Furthermore, I would like to thank Tom Aldenberg for his support in finalising Chapter 7 and my son Yuri for sorting out and polishing the references.

261


Acknowledgements


I thank my mother, who passed away too early, and my father for stimulating me in my education. As said above, many year of work has flown in this thesis. All those years, I could function and respond to sometimes very demanding questions due to a stable and stimulating home environment with my beloved wife Anita and my three great sons Ise, Tim and Yuri. Therefore, they deserve most of the credits. Theo Utrecht, 30th of March 2009

262