ORIGINAL PAPERS International Journal of Occupational Medicine and Environmental Health, Vol. 14, No. 4, 397—402, 2001
SIMULATIONS IN HEALTH RISK ASSESSMENT MAREK BIESIADA Department of Health Risk Assessment Institute of Occupational Medicine and Environmental Health, Sosnowiec, Poland
Abstract. Health risk assessment procedure provides a clear and systematic form of quantitative (or semi-quantitative) description of environmental health impact. It is well known that this approach is burdened with various types of uncertainties of different origin and nature. Therefore, the results of risk assessment should always contain both the “number” and the “measure of uncertainty”. The problem is that even if one does attempt to take account of the uncertainty, one does not know a priori what is the probability of getting a given risk value within the specified range of uncertainty. A promising tool for the assessment of risk which provides a means of describing the sensitivity with respect to different exposure factors and evaluating different intervention scenarios is the technique of Monte Carlo simulation. In this probabilistic approach all variables and parameters used in risk assessment may be regarded as distributions throughout the analysis. A process of repeated simulations is then used, during which the estimated quantity (risk in this case) is calculated many times (usually 10,000 or more) with randomly chosen values of variables and parameters, covering their range of variability and reproducing the assumed distribution density. The final result is given in the form of a probability distribution of risk. The idea of Monte Carlo simulations in health risk assessment concerning the exposure to heavy metals in drinking water is illustrated in the population living in the vicinity of the “Łubna” waste site, taken as an example. Key words: Risk assessment, Monte Carlo simulations
INTRODUCTION The main purpose of this paper is to give an overview of the risk assessment methodology with particular emphasis on the areas where numerical Monte Carlo simulations can be helpful. This is the reason for certain imbalance as far as the structure of the paper is concerned (e.g. a lengthy introduction as compared with other sections). The real data are invoked here only for the purpose of illustration. Since ancient times intuitive (and tacit) risk assessment has been fundamental for human survival [1]. But, it is only about 30 years [2] since risk assessment has became an efficient tool for risk management. The subject itself has many faces and dimensions, including financial, economic and social issues. We shall confine our attention to the human health aspect of the risk assessment methodology.
In its modern understanding, health risk assessment is a multidimensional task that requires a specific approach and involves all stakeholders into the whole process (including community consultation). Health risk assessment is based on the integration of fundamental knowledge of a number of disciplines, like environmental science, medicine, public health, toxicology or physiology. Goals of health risk assessment are threefold: ■ to provide objective (quantitative or semi quantitative) measures of health risks; ■ to provide an input to the decision making process; and ■ to become a part of cost-benefit analysis. Despite the longstanding debate over the meaning of risk and its perception, there is a consensus that risk should be understood as a probability (or likelihood) of developing certain adverse effects under specific conditions of expo-
The paper presented at the Conference “Metal in Eastern and Central Europe: Health effects, sources of contamination and methods of remediation”, Prague, Czech Republic, 8–10 November 2000. Address reprint requests to Dr M. Biesiada, Department of Epidemiology, Institute of Occupational Medicine and Environmental Health, Kościelna 13, 41-200 Sosnowiec, Poland (e-mail:
[email protected]).
IJOMEH, Vol. 14, No. 4, 2001
397
ORIGINAL PAPERS
M. BIESIADA
sure. However, transparent strategy has been elaborated only for chemical hazards. It is customary [3,4] to distinguish the following stages of risk assessment process: ■ hazard identification, ■ exposure assessment, ■ evaluation of dose-response relationship, ■ risk characterization and uncertainty analysis. To some extent these stages are independent elements (the first and the third ones are essentially toxicological and the major responsibility of an investigator is to gather appropriate information). Hazard identification means to find out whether a given chemical substance or physical factor is responsible for any adverse health effect, and if so - to identify what kind of effects does it produce. Exposure is usually quantified as the concentration of an agent in the medium contacted by human population integrated over the duration of contact. There are three elements necessary for an exposure to occur: ■ a source of contamination, ■ a susceptible person or population, and ■ a pathway – route along which the transfer from the source to the exposed person takes place. The fate of contaminants between source and exposed person involves the redistribution of contaminants between three major elements of the environment: soil, air and water, including also the sediment and biological agents. Then the individual's uptake of contaminants may occur through: gastrointestinal tract, respiratory system or skin (usually neglected). Already at this point it becomes clear that integrated risk assessment (i.e. exhaustive consideration of all exposure routes) demands the use of computer simulations due to inherent complexity of the problem. The idea behind the calculation of a dose is very simple and may be expressed by the following formula:
Dose agent; pathway = TF ⋅ CF ⋅
398
IJOMEH, Vol. 14, No. 4, 2001
C ⋅ UR BW
where: TF (time factor) denotes the scenario specific time averaging factor (time and frequency of contact normalized to lifetime exposure time); C denotes the concentration of contaminant in a specific medium (air – in (mg/m3), water – in (mg/l), soil - in (mg/kg), food – in (mg/kg)) and a specific agent (substance); UR is the uptake rate – e.g. an inhalation rate (m3/d) or consumption rate (l/d, mg/d, g/d) – age and gender specific factor; BW is the body weight - an age and gender specific factor; CF denotes a conversion factor – i.e. a coefficient normalizing the dose to (mg/(kg d)) units. For practical reasons, it is essential to make a fundamental distinction between carcinogenic and non-carcinogenic substances. The former are considered as non-threshold substances i.e. there is no safe threshold of exposure below which the probability of developing cancer is zero. It is well known that this assumption is not strictly correct from the scientific point of view. There is evidence that some carcinogens display threshold effects and perhaps only genotoxic ones can be safely deemed non-threshold. However, for risk assessment, a discipline placed at the interface between science and decision-making, the aforementioned dichotomy (carcinogenic vs toxic) adds methodological clarity. The conversion from exposure to health effect (probability of cancer) comes from the socalled cancer potency slope (CPS) which is a slope of dose-response curve [3]. In the case of carcinogenic substances the individual risk understood as an excess lifetime risk, is calculated by multiplying cancer potency slope and a dose:
Risk = CPS ⋅ Dose For the purpose of public health, individual risk may be converted into a population risk (giving an expected number of people at risk). Different strategies are used in the assessment of chronic exposures to non-carcinogenic toxic substances [4]. The dose-response relationship for such substances suggests the existence of a threshold value of exposure below which
SIMULATIONS IN HEALTH RISK ASSESSMENT
the substance is harmless. This threshold value is quantified as the no observed adverse effect level (NOAEL). Then the reference dose (RfD) is derived taking into account possible uncertainties (including interspecies extrapolation and reliability of database). Having introduced the reference dose, one compares actual exposure (dose) with the reference dose by calculating the hazard quotient (HQ):
HQ=
Dose ⋅ RfD
Hazard quotient greater than 1 provides evidence that a potential health risk associated with chronic exposure to a given substance does exist. Otherwise, it is assumed that the risk is at acceptable level. In every case of risk assessment, uncertainty must be discussed in order to incorporate the results into the context, to add integrity to analysis, and to manage future data collection. Poor description of uncertainties as well as inappropriate use of exposure factors or computational errors (including inappropriate round-off procedure) may misrepresent the risk, and thus make the results meaningless. One of the sources of uncertainty is variability i.e. the interpersonal differences in the exposure levels and in the magnitude of individual response. This particular source of uncertainty can be reliably modeled by applying Monte Carlo simulations. In this paper, the population living in the vicinity of the “Łubna” waste site was taken as an example to illustrate this method. Some particular problems associated with the use of historical data in the risk assessment process are also emphasized.
MATERIALS AND METHODS The material used to present some of the practical aspects of risk assessment has been gathered under the project on the health status assessment of the population living in the vicinity of the “Łubna” waste site [5]. Again we stress that somewhat lengthy description of the study site and of the current or historical data available is indispensable to illustrate the difficulties one encountered in “real life” and to emphasize the discrepancy in percep-
ORIGINAL PAPERS
tion often noticed among researchers who approach the same problem from different perspectives (i.e. environmental vs health risk assessment). The “Łubna” waste site (surface area of 180,000 m2) lies south of Warsaw at the top of the hill, 1–2 km away from four local communities (Łubna, Baniocha, Brześce, Kawęczyn). The site is surrounded by meadows and woods. Annual dumping yield is about 2 million cubic meters of waste, 98% of which is declared as municipal waste. Since the capital does not have a separate dumping site for hazardous wastes, one can expect that some unspecified and actually hard to estimate fraction of wastes is hazardous to the environment and human health. The history of the site dates back to the 1960s when the waste site grew out of a local landfill. In consequence, the waste site has never been properly isolated from the ground. Growing public concern about possible health effects of the site has prompted the local authorities to take appropriate steps. Among other initiatives the Department of Epidemiology of the Institute of Occupational Medicine and Environmental Health in Sosnowiec has been invited to perform a survey on the health status of the local population. Health status assessment was based on a questionnaire, standard medical examination and laboratory tests. The survey was based on voluntary participation. More detailed description of the study design and the results is published elsewhere [6]. Taking this opportunity an attempt was made to perform the health risk assessment and to use the current and historical environmental data gathered by independent agencies. However, due to the shortage of adequate information on air pollution, water quality soil and food chain the data available were practically useless from the point of view of integrated quantitative health risk assessment. Risk assessment methodology The comprehensive health risk assessment would demand a quantitative analysis of the following exposure routes: ingestion of drinking water and contaminated food, inhalation of particulates and volatile organic compounds released from the site and contaminated soil, accidental soil ingestion, and dermal absorption of water and soil. In
IJOMEH, Vol. 14, No. 4, 2001
399
ORIGINAL PAPERS
M. BIESIADA
view of the aforementioned difficulties the risk assessment was mostly descriptive (qualitative). However, an attempt was made to apply risk assessment procedure with respect to heavy metal contamination of underground water [7]. The data were based on measurements of heavy metal concentrations taken in the test wells surrounding the landfill. Therefore, the scenario adapted, which assumes the same concentration of heavy metals in drinking water as measured in test wells should be regarded as ultra conservative (the worst possible scenario). In the conventional “point” approach to risk assessment single values of variables are used. They usually contain: exposure (point concentrations of a pollutant in a given medium), individual factors (inhalation rate, ingestion rate, body weight, body surface area, etc.), or parameters (time weighted factors such as contact frequency, contact duration or lifetime exposure), appropriately defined (as mean values or the 95th percentiles) for a “typical” population. Then the results are reported as the “number” and the “measure of uncertainty”. However, in most cases the audience pays attention only to the numerical value of risk and ignores the uncertainty factor. Even if one does attempt to take account of the uncertainty, one does not know a priori what is the probability of getting a given risk value within the specified range of uncertainty. Monte Carlo simulations, one of the most widely used techniques for modeling natural, social or economic phenomena helps to overcome this problem. It is therefore not surprising that this method is also used in risk assessment. In the probabilistic approach, based upon the Monte Carlo techniques, all variables and parameters used in risk assessment may be regarded as distributions (more precisely as random variables characterized by probability distribution functions) throughout the analysis [8]. During the process of repeated simulations, the estimated quantity (a risk or hazard quotient) is calculated many times (usually 10,000 or more) with randomly chosen values of variables and parameters covering their range of variability and reproducing the assumed distribution density. The final result is given in the form of a probability distribution for a given risk. This is inherently a more informative way of presenting the results allowing to capture rigorously the uncertain-
400
IJOMEH, Vol. 14, No. 4, 2001
ties related to interpersonal variability in biological factors and to dynamical character of contaminated media. For the purpose of Monte Carlo simulations, water concentrations (C) of Cr, Zn, Cd, Ni were modeled with triangular distribution. This choice, recommended in the literature [9,10], is a standard whenever the actual distribution function is not known and the only data available are the mean, maximum and minimum values of a given variable. Body weight (BW) was modeled with truncated normal distribution N (70 kg, 20 kg) restricted to the range of 40–120 kg [6]. In a similar manner, water consumption rate distribution in the population was modeled with truncated normal distribution N (2 l/d; 0.5 l/d) restricted to the range 1–5 l/d [6]. Finally, the reference doses for the above listed heavy metals were taken from the table provided by the US EPA Region III [11].
RESULTS The results of the simulations are shown in Fig. 1 and summarized in Table 1. Fig. 1 displays the simulated probability distribution function for the total hazard quotient HQ = HQ(Cr) + HQ(Zn) + HQ(Cd) + HQ(Ni), whereas Table 1 shows the contribution of each contaminant to the total HQ. It turned out that even 90th percentiles of simulated hazard quotients were much below unity. In the light of ultra conservative assumption that quality of drinking water is the same as the water from test wells, this means that exposure to heavy metals in drinking water associated with the waste site is negligible. Of course, nothing can be said about other routes of exposure or other substances in water. Table 1. Statistical parameters of the total hazard quotient HQ and hazard quotients HQCr, HQZn, HQCd , HQNi for each substance simulated by using the Monte Carlo technique HQ
HQCr
HQZn
HQCd
HQNi
Mean
0.22
0.16
0.03
0.01
0.02
Standard deviation
0.06
0.05
0.01
