Retrospective Exposure Assessment and Quality ...

Ann. occup. Hyg., Vol. 47, No. 1, pp. 37–47, 2003 © 2003 British Occupational Hygiene Society Published by Oxford University Press DOI: 10.1093/annhyg/mef094

Retrospective Exposure Assessment and Quality Control in an International Multi-centre Case–Control Study H. TINNERBERG1*, P. HEIKKILÄ2, A. HUICI-MONTAGUD3, F. BERNAL3, A. FORNI4, S. WANDERS5, H. WELINDER1, P. WILHARDT6, U. STRÖMBERG1, H. NORPPA2, L. KNUDSEN7, S. BONASSI8 and L. HAGMAR1 1Department

Received 1 July 2002; in final form 19 September 2002 The paper presents the exposure assessment method and quality control procedure used in an international, multi-centre case–control study within a joint Nordic and Italian cohort. This study was conducted to evaluate whether occupational exposure to carcinogens influenced the predictivity of high frequency of chromosomal aberrations (CA) in peripheral lymphocytes for increased cancer risk. Occupational hygienists assessed exposures in each participating country: Denmark, Finland, Italy, Norway and Sweden. The exposure status to a carcinogen or a clastogen was coded in the cohort according to the original CA studies at the time of CA testing, but not for the whole work life. An independent occupational hygienist coordinated harmonization of the assessment criteria and the quality control procedure. The reliability of the exposure assessments was calculated as deviation from the majority of the assessors, as Cohen’s κ and as overall proportion of the agreements. The reassessment of the exposures changed the exposure statuses significantly, when compared with the original cohort. Harmonization of the exposure criteria increased the conformity of the assessments. The prevalence of exposure was higher among the original assessors (the assessor from the same country as the subject) than the average prevalence assessed by the other four in the quality control round. The original assessors classified more job situations as exposed than the others. Several reasons for this are plausible: real country-specific differences, differences in information available to the home assessor and the others and misunderstandings or difficulties in translation of information. To ensure the consistency of exposure assessments in international retrospective case– control studies it is important to have a well-planned study protocol. Due to country-specific environments a hygienist from each participating country is necessary. A quality control study is recommended, to be performed as described, combined with round-table meetings to minimize information bias between the assessors. Keywords: case–control; exposure assessment; international; reliability; retrospective

tasks is to assess exposure retrospectively (Kauppinen, 1996). To increase the credibility of retrospective case–control studies, thorough presentation of the exposure assessment procedure and measures of intra- and inter-rater reliability is recommended (Stewart and Stewart, 1994). Reliability has been

INTRODUCTION

In occupational epidemiology one of the most crucial *Author to whom correspondence should be addressed. Fax: +46 46143702; e-mail: [email protected]

37

Downloaded from http://annhyg.oxfordjournals.org/ by guest on June 7, 2013

of Occupational and Environmental Medicine, Lund University Hospital, SE-221 85 Lund, Sweden; 2Finnish Institute of Occupational Health, Helsinki, Finland; 3Instituto Nacional de Seguridad e Higiene en el Trabajo, Barcelona, Spain; 4Dipartimento di Medicina del Lavoro, Clinica del Lavoro ‘L. Devoto’, Milan, Italy; 5Department of Occupational Medicine, Telemark Central Hospital, Skien, Norway; 6National Institute of Occupational Health, Copenhagen, Denmark; 7Institute of Public Health, University of Copenhagen, Copenhagen, Denmark; 8Department of Environmental Epidemiology, Istituto Nazionale per la Recerca sul Cancro, Genova, Italy

38

H. Tinnerberg et al.

MATERIALS AND METHODS

The exposure assessment procedure Five occupational hygienists assessed exposures in each participating country: Denmark, Finland, Italy, Norway and Sweden. There was also an independent occupational hygienist team (Spain) to verify the reliability of the exposure assessment procedure. The exercise was divided into six steps: (i) a review of the literature from the original cytogenetic studies, selection of relevant exposure indices (agents or activities)

and establishment of the criteria for exposure classification; (ii) identification of subjects; (iii) data collection for the exposures (occupational exposure, smoking status, cytostatics, radiotherapy) for all cases and controls; (iv) harmonization of assessment criteria among occupational hygienists; (v) individual exposure assessments; (vi) analysis of the reliability of the assessments. During the procedure the whole group of occupational hygienists met five times. Moreover, some of the group met several times and there was also frequent use of e-mail to facilitate the work. The ethical aspects of the study protocol were approved by the local Ethics Committees or legal authorities in all the participating countries. Selection of exposure indices Scientific publications on chromosomal aberrations originating from the original cytogenetic studies comprising the basis for the pooled cohort (Hagmar et al., 1994; Bonassi et al., 1995) were compiled and carefully scrutinized. Relevant exposure agents or occupational activities, which were the reason for the original cytogenetic testing, were identified and listed. The main criteria in the selection of exposures with a potential impact on the association between CA and cancer risk were clear evidence or a strong suspicion of their carcinogenic or clastogenic properties. A matrix with 23 categorized occupational exposure indices was constructed (Table 1). Some subjects had not only been exposed to the agents included in the original cytogenetic studies, but also to others classified as class I carcinogens by the International Agency for Research on Cancer. A further category was created summarizing exposure for these agents. Exposure assessments were performed semiquantitatively at four levels (none, low, medium and high) for 18 of the agents, at three levels (none, low and high) for two agents and qualitatively for the remaining four agents. Cut-off limits between the different exposure levels were carefully defined and set in three different ways. For two agents they were based on absorbed dose and for a further 12 they were based on 8 h time-weighted average (TWA) air concentrations, expressed as either absolute values or as a percentage of the corresponding threshold limit value (TLV) [American Conference of Governmental Industrial Hygienists (ACGIH, 1997)]. For the remaining six agents the limits were based on either the frequency of exposure or the type of activity. An even distribution of the subjects in each exposure level category was taken into account to set quantitative cut-off values. Background exposures were defined for seven ubiquitous agents to which the general population can be exposed at low levels (Table 1).


evaluated in some studies (Goldberg et al., 1986; De Cock et al., 1996; Siemiatycki et al., 1997; Benke et al., 1997, 2001; Tinnerberg et al., 2001), but such data are, to our knowledge, sparse when the exposure assessment has been performed in more than one country. An international epidemiological study aiming to answer the question of whether occupational exposure to carcinogens modified the association between chromosomal aberrations (CA) in peripheral lymphocytes (PBL) and cancer gave the opportunity to evaluate methods for exposure assessment. It had earlier been shown that a high frequency of CA in healthy subjects predicts an increased cancer risk (Hagmar et al., 1994; Bonassi et al., 1995). The results from a European collaborative study [The European Study Group on Cytogenetic Biomarkers and Health (ESCH)] convincingly supported the earlier studies and also showed that the cancer predictivity of CA was not modified by either country, gender, age at cytogenetic testing or time since testing (Hagmar et al., 1998). However, whether the predictivity of CA for cancer was dependent or independent of exposure to carcinogens remained to be evaluated. To clarify this a nested case–control study was conducted within the ESCH cohort (Bonassi et al., 2000). The results from that study showed that neither exposure to occupational carcinogens nor smoking modified the risk predicted by CA. The subjects in the Nordic and Italian cohorts (Hagmar et al., 1998) were originally selected to participate in cross-sectional cytogenetic studies because of their potential occupational exposure to clastogens or carcinogens or because they were unexposed. CA in PBL have been used as a biomarker in order to survey genotoxic or carcinogenic exposures in workplaces since the 1960s. The exposure information for the cohort members was limited to the period of CA testing in the original studies. The nested case–control study provided the opportunity to collect information on lifetime occupational histories and other relevant factors. The aim of this paper is to describe the exposure assessment method used in an international multicentre case–control study and to evaluate the quality control procedure performed.

Retrospective exposure assessment

39

Table 1. The exposure matrix with the definitions, number of subjects by agent and by level of exposure for the 582 subjects included in the case–control study Exposure

Background

Low

Medium

High

Organic solventsa (% of TLV) (n)

10–25 (52)

>25 (50)

1–20 (85)

>20–50 (44)

>50 (26)

Polycyclic aromatic hydrocarbons (n) Non-exposed (491) e.g. drivers (68)

e.g. miners (26)

e.g. coke oven workers (4)

Benzene (p.p.m.) (n)

1–2.5 (22)

>2.5 (31)

Other IARC class 1 agents (n)

Non-exposed (509) Exposed (73)

c

c

Ionizing radiation (mSv/y) (n)

5–10 (19)

>10 (16)

Asbestos (f/ml) (n)

0.21 (15)

>1 (4)

Lead (µg/100 ml blood) (n)

35–60 (16)

>60 (40)

Man-made mineral fibres (n)


c

c

Aromatic hydrocarbonsb (p.p.m.) (n) 50 (6)

Non-exposed (547) 50 (20)

Styrene (p.p.m.) (n)


Formaldehyde (mg/m3) (n)

1 (14)

Vinyl chloride (ppm) (n)


Rubber chemicals (n)

Non-exposed (558) Low (13)

d

High (18)

Anticancer agents (AC) (n)

Non-exposed (562) Handling AC 1 time/month (9)

Regular manipulation of AC (11)

Welding fumes, stainless steel (n)

Non-exposed (563) Work in a welding MIG, TIG welding area (6) (15)

MMA welding (7)

Ethylene oxide (p.p.m.) (n)


Herbicides (n)

Non-exposed (571) 40 h/year, low level >40 h/year, high level (5) (1)

Cadmium (µg/m3) (n)

Non-exposed (572) 40 h/year, high level (1) (2)

Propylene oxide (p.p.m.) (n)

Non-exposed (576) 2.5 (6)

Tannery work (n)


c

c


c

c

0.15–1 (18)

>50 (7)

Each subject can contribute to more than one level of exposure. The non-exposed are classified as not exposed during the entire assessment period. aOrganic solvents not including aromatic hydrocarbons. bAromatic hydrocarbon solvents not including benzene or styrene. cFor these agents only the exposure status (unexposed/exposed) was assessed. dFor these agents the exposure was assessed for three levels only (unexposed, low and high).

Study population The cohort of subjects examined as adults for CA in PBL between 1965 and 1988 comprised 3541 subjects (1968 from 10 laboratories in the Nordic countries and 1573 from 10 laboratories in Italy). In total, the Nordic cohorts comprised 93 incident cancer cases and the Italian cohort comprised 62 deceased cancer cases during a follow-up period that ended between 1993 and 1996 for the participating countries. For the nested case–control study, the 155 cases and their corresponding controls gave 582 subjects (Bonassi et al., 2000), which was the population in the present study. Of the 582 subjects, 159 had been classified as unexposed in the original cytogenetic studies, 401 as exposed to one of the 17 original exposures also assessed in this study and 22 were classified as

exposed in the original studies but not to one of the exposure categories assessed in this study (Table 2). Data collection The responsible occupational hygienist in each country gathered data on occupational exposures, smoking, radiotherapy and cytostatic treatments. The subjects were approached with a postal questionnaire and asked to provide a lifelong working history by listing the name and location of company for each job held, main job tasks, smoking habit and the year of receiving cytostatics or radiotherapy. The inquiry was sent to cases and controls or, if deceased, to next of kin (only widows or widowers, children, parents or siblings were accepted). Together with the questionnaire, an explanatory covering letter was sent. When the questionnaire had been returned, the occupational hygienist performed a partly structured telephone


Non-exposed (546) 3%. Furthermore, κ for multiple ratings according to Fleiss (1981) for the 4 × 4 matrix and for a 2 × 2 matrix (non-exposed/exposed) and prevalence were calculated. The overall proportion of agreement (Fleiss, 1981) was calculated when the 10 pairwise matrices were summed up to one matrix. RESULTS

Data collection The information sources used in the five different countries are shown in Fig. 1. For eight subjects (1.4%) there was no information available and for another seven (1.2%) the data were so scarce that the exposure status remained uncertain. For 142 subjects (24%) there was only one source of information and for 210 (36%) there were at least three different sources of information. The total number of person-years under observation was 24 176. Data were missing for 1131 personyears (4.7%) and for 384 person-years (1.6%) data were so scarce that only uncertain assessments could be performed. Prevalence of exposure The distribution of the exposure prevalences is displayed in Table 1. The highest exposure prevalences were found for agents such as organic solvents that have been widely used. High exposure prevalences were also seen for agents that were common in the original studies. Among the 73 subjects that were classified as exposed to other IARC class 1 chemicals the most common exposures were silica (45%), wood dust (11%), mineral oil (11%) and working in a shoe factory (10%). The exposure statuses of the subjects in the case– control study according to the original cytogenetic studies on the cohort are presented in Table 2. As a result of the present study only 99 of the 159 subjects considered as unexposed in the original cytogenetic studies were classified as unexposed during their


Deviations from the majority In the calculations of deviations from the majority of assessors only the exposure status (nonexposed/exposed) was taken into account. The deviations are presented in three ways: (i) for each specific exposure; (ii) for all exposures by an assessor; (iii) for the main classification used in the epidemiological study.

Main classification used in the epidemiological study. In the epidemiological study the exposures were collapsed into three groups: (i) medium or high exposures to agents evaluated in the original cytogenetic studies and classified by the IARC as human carcinogens (class 1); (ii) all other agents in the matrix; (iii) non-exposed. In our calculations of the deviations from the majority of the assessors we used the time window from 5 years before the time of CA test until the test.

Retrospective exposure assessment

the hygienists assessed that 281 of these 390 subjects were exposed for the original exposure and 22 for another exposure.

whole work history. Work histories were missing for five subjects and 55 were assessed as exposed for at least one time period. Of these 55 subjects, 25 were also coded as exposed at the time of CA testing. Organic solvents and polycyclic aromatic hydrocarbons (PAH) were the most common exposures among this group. Of the 423 subjects that had been classified as exposed in the original cytogenetic studies, 23 were reclassified as unexposed by the five raters, data were missing for 10 and 390 subjects were assessed as exposed. At the time of CA testing

Other assessors Average

Range

35.3

17.6

0–32.4

9.0

7.6

6.5–10.5

Norway

13.6

8.5

6.2–10.0

Finland

9.2

8.2

5.8–11.3

Sweden

12.3

10.0

8.8–10.5

Denmark Italy

Table 4. Distribution of the exposure status assessments by the exposure indices in the quality control round: classified as unexposed or exposed by all the assessors or when one or two deviated from the majority Agent

All nonexposed

One assessor Two deviating assessors deviating

All exposed Total

Expected deviations on own assessments

Observed deviations on own assessments

Organic solvents

68

17

19

13

117

11

10

Aromatic hydrocarbons

61

22

8

9

100

8

11

Polycyclic aromatic hydrocarbons

64

5

11

6

86

5

5

Benzene

64

9

10

5

88

6

3

Other IARC class 1 agents

65

15

4

1

85

5

7

Ionizing radiation

53

6

4

5

68

3

1

Asbestos

60

16

5

4

85

5

9

Lead

58

3

2

4

67

1

2

Man-made mineral fibres

57

3

3

0

63

2

4

ChromiumVI

54

5

3

0

62

2

3

Nickel

55

4

2

3

64

2

2

Styrene

55

1

1

4

61

1

0

Formaldehyde

56

1

3

0

60

1

2

Rubber chemicals

55

2

0

2

59

0

0

Welding fumes

55

1

3

0

59

1

1

Herbicides

56

1

0

2

59

0

0

Cadmium

55

1

0

1

57

0

0

Insecticides

55

1

0

1

57

0

0

Tannery

56

1

0

0

57

0

0

276

0

0

12

288

0

0

1378

114

78

72

1642

53

60

Other agentsa Total

aOther agents include anticancer agents, propylene oxide, vinyl chloride, ethylene oxide and fungicides. The assessors agreed on the exposure status for the three first agents in 3–5 periods and for the rest of the periods the status was unexposed; all the periods for the last two agents were coded as unexposed.


Quality control The exposure prevalences for the 55 subjects assessed by the original assessors from the five different countries were compared with the prevalences assessed by the other occupational hygienists (Table 3). For all five countries, the prevalences assessed by the original assessor were higher than the average prevalence assessed by the other four. For three of the five countries the prevalence assessed by the original assessor was the highest assessed prevalence for that country. The distribution of deviating assessments for each exposure index is displayed in Table 4. The numbers of deviating periods for an agent are strongly correlated with the total numbers of periods for that agent (Spearman’s ρ = 0.89, P < 0.01). The proportion of full agreement on the exposure status for the 1642 periods was 88%, the majority (84%) of the periods being coded unexposed. Thus, one or two raters disagreed on the exposure status for 12% of the periods. The number of observed deviations on own

Table 3. The prevalence of exposure in the quality control round, as assessed by the original assessor and by the other assessors Original assessor

43

44

H. Tinnerberg et al.

assessments was higher than or equal to the expected deviations on own assessments for all agents except four: organic solvents, benzene, styrene and ionizing radiation. The total number of observed deviations on own assessments was higher than expected (60 versus 53). Of the 192 deviating periods, the minority of assessors regarded 132 (69%) of them as exposed. For the 60 deviations on own assessments as many as 56 (93%) belonged to the assessments that were in a minority (Table 4). The numbers of expected and observed deviating periods on own assessments for all periods are displayed in Table 5 with respect to the five assessors. For four of five assessors the numbers of

observed deviating periods are higher or equal to the expected ones. When considering the classification of exposure applied in the epidemiological study 34 of the 55 subjects were classified in the same exposure category independently of the assessor. All three levels of exposure were assigned to only two subjects. Regarding the remaining 19 subjects, one or two assessors’ classification differed from the majority and nine of these were on own assessments. The calculated κ values, range and overall proportion of agreement are presented in Table 6. Strong negative correlations were seen between the prevalences and reliability of exposure assessment,

Denmark

Italy

Norway

Finland

Sweden

Total

Total number of periods

34

524

339

346

399

1642

Number of exposed periods

13

70

58

54

69

264

Number of expected deviations in exposed periods

13

72

59

55

71

270

Number of expected deviating periods on own assessments

3

14

12

11

14

54

Number of observed deviating periods on own assessments

4

14

17

9

16

60

Table 6. The prevalence, κ statistics and overall proportion of agreement of the assessments performed by the five assessors in the randomized selection Exposure agenta

Periodsb (n)

Prevalencec (%)

Range of κe

κd 4×4 matrix

2×2 matrix

Overall proportion of agreement (%)f 4×4 matrix

2×2 matrix

Organic solvents

117

24.1

0.45

0.57

0.26–0.69

78.5

84.4

Aromatic hydrocarbons

100

18.8

0.50

0.57

0.40–0.58

84.1

87.2

Polycyclic aromatic hydrocarbons

86

15.6

0.49

0.60

0.38–0.73

85.8

89.5

Benzene

88

13.9

0.45

0.54

0.31–0.60

86.4

89.1

Other IARC class 1 agents

85

8.9

–g

0.39

0.14–0.78

–g

90.3

Ionizing radiation

69

15.6

0.55

0.74

0.34–0.77

87.4

93.0

Asbestos

85

14.3

0.43

0.55

0.25–0.60

85.4

88.9

Lead

67

8.4

0.72

0.77

0.56–0.88

95.5

96.4

Nickel

64

9.4

0.63

0.74

0.43–0.86

93.4

95.6

Styrene

61

8.9

0.87

0.90

0.79–1.0

97.9

98.4

Vinyl chloride

61

6.6

0.95

1.0

0.87–1.0

99.3

1.0

Rubber

58

5.2

0.76

0.86

0.48–1.0

97.6

98.6

Anticancer agents

57

5.2

0.86

1.0

0.83–1.0

98.6

1.0

Herbicides

59

4.7

0.80

0.92

0.59–1.0

98.1

99.3

Propylene oxide

60

8.3

0.77

1.0

0.58–1.0

96.3

1.0

aFor

the remaining nine exposure agents the prevalences for at least one assessor were