Article Text

Original article
Comparison of exposure estimates in the Finnish job-exposure matrix FINJEM with a JEM derived from expert assessments performed in Montreal
  1. Jérôme Lavoué1,
  2. Javier Pintos1,
  3. Martie Van Tongeren2,
  4. Laurel Kincl3,4,
  5. Lesley Richardson1,
  6. T Kauppinen5,
  7. Elisabeth Cardis4,
  8. Jack Siemiatycki1
  1. 1University of Montreal Hospital Research Center (CRCHUM), Montréal, Canada
  2. 2Institute of Occupational Medicine, Edinburgh, UK
  3. 3Oregon State University, College of Public Health and Human Sciences, Corvallis, Oregon, USA
  4. 4Centre for Research in Environmental Epidemiology (CREAL), Hospital del Mar Research Institute (IMIM), CIBER Epidemiologia y Salud Pública (CIBERESP), Barcelona, Spain
  5. 5Department of Occupational Health, Finnish Institute of Occupational Health, Helsinki, Finland
  1. Correspondence to Dr Jérôme Lavoué, Centre de recherche du CHUM, 3875 rue st urbain, 3ieme étage, bureau 316, Montréal, QC H2W1V1, Canada; jerome.lavoue{at}umontreal.ca

Abstract

Context Retrospective exposure assessment in population-based case–control studies poses a major challenge due to the wide range of occupations and industries involved. The FINJEM is a generic job-exposure matrix (JEM) developed in Finland, which represents a potentially cost-effective exposure assessment tool. While FINJEM has been used in several studies outside Finland, little is known of its applicability in other countries.

Methods We compared prevalence and intensity of exposure in FINJEM with a JEM developed from expert assessments of occupational histories obtained in a population-based case–control study in Montreal. Agreement for prevalence of exposure was measured by weighted κ coefficients between prevalence categories. Agreement for exposure intensity was measured by Spearman correlation coefficients between cells with non-null exposure.

Results The comparison involved 27 chemicals, the time period 1945–1995 and included 4743 jobs initially assessed by the Montreal experts. 4293 combinations of agent, occupational title and period were available for comparison of prevalence. Agent-specific prevalence was consistently higher in the Montreal JEM (median difference 1.7%). Agent-specific κ values between prevalence categories varied from 0.89 (welding fumes) to 0.07 (flour dust). The comparison of exposure levels involved 14 agents and 198 cells with non-null exposure in both sources. Agent-specific Spearman correlation varied from 0.89 (flour dust) to −0.35 (benzo(a)pyrene).

Conclusion Our observations suggest that information concerning several agents (eg, metals, welding fumes) can be successfully transported from Finland to Canada and probably other countries. However, for other agents, there was considerable disagreement, and hence, transportability of FINJEM cannot be assumed by default.

  • Case–control study
  • occupational exposure assessment
  • job-exposure matrix
  • validity and reliability
  • formaldehyde
  • retrospective exposure assessment
  • exposure assessment
  • statistics
  • toxicology
  • cancer
  • hygiene/occupational hygiene
  • wood dust
  • epidemiology
  • electromagnetic fields
  • non-ionising radiation
  • ionising radiation
  • public health
  • risk assessment
  • mathematical models
  • health surveillance
  • health screening
  • gender
  • Bayesian statistics
  • benzene
  • asbestos
  • alcohol
  • air pollution
  • respiratory
  • occupational asthma
  • mesothelioma
  • leukaemia

Statistics from Altmetric.com

What this paper adds

  • FINJEM is a generic job-exposure matrix created in Finland that can provide exposure information for community-based case–control studies but has not been extensively evaluated for use outside Finland.

  • Compared with occupational exposure data based on Montreal workers' description of their jobs and exposure assessments by local Montreal experts, FINJEM performed reasonably well in general.

  • FINJEM exposure estimates can be used in urban Northern American areas for assessing occupational exposure to some agents (eg, welding fumes, iron), but its transportability for other agents or settings cannot be assumed by default.

Introduction

The assessment of retrospective occupational exposure poses major challenges in population-based case–control studies. Exposure has to be evaluated for a wide spectrum of occupations and industries for the lifetime occupational histories of study subjects, representing hundreds or even thousands of different exposure scenarios. The main existing approaches include self-reported exposure, job-exposure matrices (JEMs) and expert review of individual work histories.1–4 Self-assessment of exposure is typically based on questionnaire checklists and subjects' knowledge and perception about their own exposure.5 JEMs are tools that automatically assign exposure to subjects on the basis of their occupational title.6 The expert assessment approach is based on detailed exposure-related information, obtained from the subjects for each job ever held. The resulting descriptions are translated into an assessment of exposure by a panel of hygienists/chemists.7 8

Notwithstanding the inherent inability of JEMs to account for exposure variations within occupations, using such a tool is an attractive option for exposure assessment because it can be implemented at low cost compared with the individual expert assessment approach, and it does not depend on error-prone self-reports of exposure. Therefore, the availability of a valid and generalisable JEM would be of great benefit in many circumstances.

Among currently available JEMs, the one that is perhaps most widely used is the FINJEM developed at the Finnish Institute of Occupational Health in the late 1990s9 and regularly updated since then.10 By contrast with most other JEMs, FINJEM is based on an extensive national occupational exposure database, and the rationale for each estimate is reported in detail. The FINJEM has been used frequently to elucidate relationships between occupational exposures and adverse outcomes in Finland and in other countries.9–11 To our knowledge, the only research efforts attempting to evaluate the applicability of FINJEM to countries other than Finland are studies by Benke et al12 (Australia) and Kauppinen et al10 (other Scandinavian countries). Benke et al concluded that FINJEM gave satisfactory results in Australia for some of the chemicals studied, underlining the need to further evaluate the transportability of this JEM to other settings. Kauppinen et al identified few major differences between the five Nordic countries studied but underlined the homogeneity of their economic structures.

The main purpose of this study was to assess the transportability of FINJEM estimates outside Finland within the framework of the INTEROCC project, a multicenter case–control study examining the relationship between occupational exposures and brain cancer. We compared estimates of intensity and prevalence of exposure in FINJEM with a JEM derived from expert reviews of individual occupational histories obtained in a population-based case–control study on lung cancer in Montreal.

Methods

The FINJEM matrix

A detailed description of the original FINJEM is available elsewhere.9 10 The 2008 version of FINJEM, purchased for the INTEROCC project, was used in this report. Briefly, the FINJEM has three dimensions: agent, occupation and time period. There were 83 agents in the version of FINJEM used in this study, of which 52 are chemical agents. The occupation axis of FINJEM includes 311 occupational titles from a Finnish classification system. The time axis includes six periods spanning 1945–2003. For each combination of these three dimensions, FINJEM provides two metrics: a prevalence of exposure denoted as P, expressed as a percentage, and a quantitative estimate of exposure intensity denoted as L, expressed in units of concentration, representing the yearly average exposure level among the exposed workers. A given occupational category is considered unexposed to a given agent if P is <5% or if the L does not exceed non-occupational background levels. The estimates in FINJEM were derived by a team of over 20 experts from the Finnish Institute of Occupational Health. Their sources of information included, among others, over 80 000 exposure measurements contained in the Finnish occupational exposure databank.11

The Montreal expert evaluations

From 1996 to 2004, a case–control study was undertaken in Montreal to explore the relationship between lung cancer and several risk factors, including occupational agents.13 14 The study included 2716 subjects who had held in total 13 817 jobs. The exposure assessment for occupational risk factors has been described in detail elsewhere.15–18 Briefly, during a face-to-face interview, each subject provided a detailed description of all jobs ever held, including information on tasks, working environment and protective measures. An expert team of chemists and industrial hygienists, blinded to case–control status, reviewed the information collected and translated each job into potential exposures from a list of 294 possible substances. Exposure assignment was performed by at least two experts and checked for consistency. For each substance considered present in each job, the experts noted three dimensions of information: the likelihood that the exposure had actually occurred (possible, probable, definite), the frequency of exposure (percentage of the number of hours per week exposed) and the relative exposure level of the agent (low, medium, high). Non-exposure was interpreted as exposure up to the level that can be found in the general environment.

Creation of the Montreal JEM

The Montreal JEM was created by aggregating the exposure information from all individual jobs held only by controls in the study.

Definition of the Montreal JEM axes

The lists of agents in the two databases (FINJEM and Montreal) were not identical, and sometimes, the definition and demarcation of a given agent differed between the databases. Twenty-seven agents were common to the two databases and defined similarly enough that we could include them in this comparison. Some had to be combined to achieve comparability: welding fumes in FINJEM corresponded to the combination of arc welding fumes and gas welding fumes in Montreal data and chlorinated solvents in FINJEM corresponded to chlorinated alkanes and chlorinated alkenes in Montreal data. For these combined agents, exposure level was equated to the maximal exposure level of the two component agents. The Montreal jobs were already coded to multiple occupation classification systems, including the International Labour Organisation's International Standard Classification of Occupations (ISCO, version 1968), used within the INTEROCC project. Therefore, ISCO'68 was selected as the occupational axis for the Montreal JEM. An expert-based crosswalk between the ISCO'68 and the Finnish classification was created for jobs reported within the INTEROCC study.19 Because 96% of the Montreal jobs that could be linked to FINJEM through the occupational crosswalk were within the three FINJEM periods 1945–1959, 1960–1984 and 1985–1995, we selected these time periods for comparison.

Derivation of the prevalence estimates for each cell of the Montreal JEM

Prevalence was estimated as the number of jobs associated with exposure divided by the total number of jobs in the cell (ie, a combination of agent, ISCO'68 code and period) of interest. In order to be included in the denominator, a job had to have at least 1 year in the time period corresponding to the cell. Cells with <10 jobs fulfilling this criterion were excluded from the comparison. A job was regarded as exposed if it was assigned exposure at least equivalent to low intensity for 2 h a week with a likelihood rating of probable or definite. In order to mimic the FINJEM interpretation of no exposure, cells with prevalence values <5% were considered unexposed.

Derivation of the exposure level estimates for each cell of the Montreal JEM

The Montreal ordinal classification of exposure (low, medium, high) was converted into a quantitative index using exposure weights of 1 for low, 5 for medium and 25 for high exposure. This choice represents what we believe to approximately mimic the way the Montreal experts used the three semiquantitative levels. Moreover, we combined the obtained indices of exposure with the frequency of exposure (expressed as percentage of number of hours per week exposed) to obtain a 40 h time-weighted exposure index for each job, hereafter referred to as Montreal exposure level. For example, assuming a typical 40 h work per week, a job evaluated as 20 h of medium exposure per week would yield a level of 5×20/40=2.5. The exposure level for a cell of the Montreal JEM was defined as the arithmetic average of the 40 h time-weighted indices of all exposed jobs corresponding to that cell. Cells for which there were less than five exposed jobs in the Montreal data set were excluded from the analysis. This threshold was preferred to 10 used for the estimation of prevalence because of the generally low prevalence of exposure in our data.

Comparison of the FINJEM and the Montreal JEM

Agreement between the prevalence estimates in both sources was measured by weighted κ (quadratic weights)20 between prevalence levels in all cells categorised as follows: 0%, 5%–15%, 15%–40%, ≥40%. The 15% and 40% values represent tertiles of the non-null prevalence estimates for all cells included in the comparison across the Montreal and FINJEM data. Overall prevalence values for each agent across occupations and time periods were calculated by taking the mean of the Montreal and FINJEM-based cell-specific estimates weighted by the number of jobs among Montreal controls in each occupational category (within an occupation, the same weight was given to all periods). The resulting estimates are therefore directly relevant to the Montreal general population. Agreement between exposure levels in the Montreal JEM and FINJEM was assessed for each agent for all cells with exposure in both FINJEM and the Montreal JEM using the Spearman correlation coefficient.21 In order to estimate agreement across agents, exposure levels in the Montreal JEM and FINJEM were standardised (divided) by their respective agent-specific 90th percentiles of exposure levels in the comparison data set.

Sensitivity analyses were conducted to evaluate the influence of several assumptions made for the comparison. Analyses common to prevalence and intensity included: (1) using an alternative definition to declare a given job ‘exposed’ in the Montreal data set: at least a 40 h time weighted index equivalent to 2 h per week where the experts coded the concentration level as medium or higher (ie, excluding the low category); (2) using no minimum exposure likelihood criterion; and (3) using the FINJEM classification of occupations as a basis for comparison instead of ISCO'68 codes. Additional analyses specific to exposure intensity included: (4) using other scales for the transformation of Montreal exposure categories into quantitative levels (1-2-3, 1-3-9 and 1-10-100 instead of 1-5-25); (5) using a Montreal exposure level not weighted by exposure frequency, that is, the original exposure estimate; (6) performing the comparison between the product of prevalence and level of exposure (in both data sets) instead of the level of exposure only and (7) using the median (instead of mean) of job-specific levels to estimate the intensity of exposure within a cell.

Results

Coverage of the comparison

The crosswalk between the ISCO'68 and FINJEM classification included 769 ISCO'68 codes, 108 being three-digit instead of five-digit codes. A total of 114 FINJEM codes were associated with at least two different ISCO'68 codes. Restriction of the Montreal evaluations to those pertaining to controls, occupations included in the crosswalk and the time window studied (1945–1995) yielded a database of 4743 jobs (1945–1959, n=1382; 1960–1984, n=2499; 1985–1995, n=853), of which 2669 were associated with probable exposure to at least one of the 27 selected agents. Restrictions for the comparison of prevalence (≥10 jobs per cell) led to the inclusion of 84 ISCO'68 codes (linked to 56 FINJEM codes), covering 64% of the Montreal study population.

Comparison of exposure prevalence

There were 159 combinations of occupation and time period, which had at least 10 jobs in the Montreal data. With 27 agents, this corresponded to 4293 comparison units. Table 1 presents the distribution of prevalence estimates based on the Montreal data and FINJEM. Bold values in table 1 highlight instances of strong disagreement between the two sources (at least three categories in the table). The 116 combinations of ISCO'68 and agent corresponding to strong disagreement (Montreal > FINJEM for 106, FINJEM>Montreal for 10) are listed in appendix 1 (see online supplementary file).

Table 1

Bivariate distribution of ISCO'68 and agent-specific estimates of prevalence of exposure (%) in FINJEM and the Montreal data

Overall weighted κ for prevalence categories based on prevalence tertiles was 0.42. There was an increase in agreement from 1945–1959 (weighted κ =0.33) to 1960–1984 (0.43) and to 1985–1995 (0.48). Agreement was higher for generic (eg, welding fumes) versus specific agents (0.47 vs 0.33, agents in the ‘specific’ category are identified in table 2). When stratifying by ISCO'68 divisions (large occupational groups), weighted κ was 0.59 for ‘7-8-9 Production and related workers, transport equipment operators and labourers’ and lower than 0.1 for the six other divisions. Table 2 presents prevalence values for FINJEM and the Montreal data as well as weighted κ coefficients by agent. Table 2 also presents the number of cells with non-null exposure in both data sets and the proportion of these cells for which FINJEM and Montreal agreed. Agent-specific weighted κ coefficients varied from 0.07 (flour dust) to 0.89 (welding fumes). Restricting the Montreal definition of exposure to a minimum of 1 h per week at medium or higher consistently reduced agent-specific prevalence values (median 1.8%) and increased the overall weighted κ to 0.51. None of the other sensitivity analyses performed showed any discernible effect on agreement.

Table 2

Comparison between FINJEM and Montreal data in prevalence of exposure

Comparison of exposure levels

The comparison of assigned exposure levels was conducted on the 198 combinations of ISCO'68 code, time period and agent (JEM cells) that had at least five jobs with exposure in the Montreal data and could be linked to a FINJEM non-null estimate. The overall Spearman correlation coefficient between Montreal and FINJEM standardised exposure levels was 0.37. There was an increase in agreement over time (from 0.29 in 1945–1959 to 0.32 in 1960–1984 and to 0.47 in 1985–1995). Agreement was higher for generic versus specific agents (0.44 vs 0.04). It was not possible to stratify the comparison by ISCO'68 division due to the small number of cells with non-null exposure outside the ‘7-8-9’ division. Table 3 provides summary statistics of the FINJEM and Montreal exposure estimates separately for each agent, along with correlation coefficients and p values. Table 3 is limited to the 14 agents that had at least five cells with non-null estimates in both data sets. Agreement varied from −0.35 (benzo(a)pyrene) to 0.89 (flour dust). Table 4 presents results of the sensitivity analyses conducted to assess the effect of various assumptions made to perform the comparison. Although some of the sensitivity analyses had a significant effect on agreement for some agents (eg, correlation coefficient from 0.47 to 0.05 for lead when restricting the Montreal data to medium or high exposure intensity), no common pattern was apparent across all agents.

Table 3

Comparison between FINJEM and Montreal data in exposure level

Table 4

Influence of different assumptions on the observed Spearman correlation coefficient between FINJEM and Montreal exposure estimates

Discussion

The present work represents to our knowledge the most comprehensive comparison effort between two multi-occupation sources of exposure information. FINJEM provides a more detailed and more documented population-based assessment than other generic JEMs that have been presented in the literature. The Montreal exposure database includes almost 14 000 jobs for which exposure to hundreds of chemicals was assessed using state-of-the-art expert-based exposure assessment. This quantity of data allowed us to perform the comparison by first agglomerating the Montreal evaluations into a JEM format, covering the majority of occupations prevalent in the general population. Whereas most previous studies have used an ‘exposed/not exposed’ dichotomy, we were able to compare prevalence and exposure levels. This design allowed us to address the question: do these two sources of information provide the same overall message on occupational exposure for the past 50 years? Not unexpectedly, our results suggest a mixed answer to this question.

The majority of marked disagreements in prevalence categories for each JEM cell corresponded to Montreal estimates being higher than FINJEM estimates. Prevalence estimates across occupations and agents confirmed this trend of more frequent assignment of exposure by the Montreal JEM. On an agent-by-agent basis, agreement in prevalence estimates varied from poor (flour dust, weighted κ =0.07) to good (welding fumes, weighted κ =0.89). Despite some variability, agreement was in general highest for metals (including welding fumes) and lowest for solvents, gases and dust. Our results are compatible with recent observations by Mester et al,22 who compared an industry-specific JEM to expert assessment within a cohort in the car manufacturing industry. They reported that experts tended to overrule the JEM-based assessment more often for solvents than for metals. When comparing individual expert assessments from a group of experts within an international, multicentre case–control study of lung cancer, ‘tMannetje et al23 also observed low agreement for ‘organic solvents’ and ‘chlorinated solvents’. However, agreement for metals ranged from poor to good. We observed a better agreement for agents with a generic (eg, welding fumes) rather than specific (eg, toluene) title. These observations are similar to results reported by ‘tMannetje et al.23 Agreement between FINJEM and Montreal prevalence estimates were higher for the most recent time period in the comparison. While this might reflect the better availability of industrial hygiene data available to both groups of experts, the fact that prevalence was significantly lower in the Montreal data for this period (results not shown) is a likely explanation. After stratification by broad occupational group, most of the agreement between FINJEM and the Montreal JEM appeared concentrated in traditional ‘blue-collar’ occupations (ISCO'68 divisions 7-8-9). This might be explained not only by the usually very low prevalence of exposure in FINJEM for other categories but also by the fact that occupations in this group have traditionally been the focus of industrial hygiene studies.

Due to the low prevalence of exposure to most agents studied, the comparison of exposure intensities between FINJEM and the Montreal JEM was limited to fewer agents and occupations than the prevalence comparison and was therefore subject to more variability. Spearman correlation coefficients varied substantially between substances, ranging from poor (benzo(a)pyrene, −0.35) to very good (flour dust, 0.89), with rankings generally similar to the prevalence results. Notable exceptions included flour dust and formaldehyde (κ=∼0.1 vs ρ∼0.6–0.9) and benzo(a)pyrene (κ=0.52 vs ρ=−0.35). In accordance to the prevalence comparison, agreement for exposure intensity was higher for recent versus older periods and for generic versus specific agents.

There are several potential sources of disagreement between FINJEM and the Montreal JEM.

First, differences in true exposure conditions are likely to exist between Finland and Canada, or more specifically, Finland and the region of greater Montreal. These differences are not easy to predict without an extensive comparative study of the economic, industrial and legal settings in each location.

Second, because the FINJEM occupation classification is less precise than ISCO'68, it is plausible that observed disagreements could be partly due to comparison being based on ISCO'68 job codes. Changing the occupational classification from ISCO'68 to the Finnish job codes did not increase agreement for prevalence or intensity estimates, suggesting little influence of the job classification system on the comparison itself. However, a little over one-third of the FINJEM occupational codes involved in the comparison were linked to more than two ISCO'68 codes. Moreover, for 65 combinations of agent and FINJEM codes (16 unique FINJEM codes), ISCO-specific exposure estimates from the Montreal JEM differed significantly within one FINJEM code (results not shown). Our observations suggest that using non ‘one-to-one’ crosswalks between occupational classification systems constitutes an added potential for misclassification in the broader scope of assigning exposure estimates for an epidemiological analysis. However, it remains unclear to what extent this adds to any misclassification from the use of classification systems not designed for exposure assessment purposes.

Third, the exposure metrics used in both systems were very different: FINJEM provides an estimate of yearly average concentration based on measurements complemented with expert assessment, and the Montreal JEM estimate was derived by averaging an ordinal exposure classification across jobs weighted by exposure frequency to arrive at an index representing an average weekly level. The results of the sensitivity analyses we performed suggest that our results actually reflect how similarly both systems rank occupations with regard to exposure.

Fourth, while the experts from both teams operationally define their criterion for the ‘exposed’ status as concentrations above environmental background (FINJEM provides an explicit quantitative threshold for most agents), it is plausible that their appraisal of this threshold would be different, explaining the higher prevalence observed in the Montreal data. Indeed, using a stricter threshold for Montreal (by excluding the lowest exposure category) increased agreement on prevalence estimation. ‘tMannetje et al24 observed a similar pattern of lower prevalence in FINJEM than in a JEM built from individual expert evaluations in a European multicentre case–control study of lung cancer. Interestingly, Benke et al12 reported higher prevalence in FINJEM than in individual assessments by a team of three experts for three out of five agents tested. In particular, they obtained 11% prevalence for formaldehyde, whereas we observed 0.6%, both based on FINJEM. While we estimated overall prevalence by weighting occupation-specific estimates to reflect our study population, Benke et al12 defined an individual job as exposed according to FINJEM if it belonged to a combination of period and occupation with P>5% in FINJEM and then calculated prevalence by counting individual jobs exposed according to this criterion. Using such a low threshold most probably increased the resulting prevalence estimates by a significant amount. Indeed, we found 7.3% using their approach. Benke et al's results would have therefore probably shown similar patterns as ours (expert prevalence > FINJEM prevalence) had they estimated prevalence by weighting occupation-specific estimates. Because we have no gold standard, it is difficult to say that FINJEM missed exposure or that the Montreal (and other) experts overestimated exposure. Based on the main rationale in each approach (process knowledge and task descriptions for Montreal and exposure measurement database for FINJEM), it is possible that the Montreal experts picked up exposures that would not seem ‘interesting’ for hygienists to monitor. This hypothesis has been mentioned by Fritschi et al25 but remains speculative. The results in online appendix 1 illustrate the differences in exposure threshold used by both approaches.

Lastly, differential validity of the assessments provided by the two sources of information is another potential reason for observing discrepancies. FINJEM is more advanced and more transparent than many of the JEMs mentioned in the available reviews, both in terms of the exposure indices provided and of the knowledge database used for its development. Insight about the validity of FINJEM has mostly come from studies identifying well-known risk factors using FINJEM estimates.10 11 Regarding the Montreal expert assessment database, evaluations based on a previous study in Montreal with the same approach and the same experts in our study also identified well-established carcinogens and showed weighted κ values around 0.70 in inter- and intrarater agreement evaluation.15 26

It bears emphasis that this was not a comparison of the Montreal expert-based assessment of individuals with a FINJEM-based assessment of individual subjects; rather, it was a comparison after transforming the Montreal data into a FINJEM-like format and more pertinently addresses the question of transportability of FINJEM, rather than of the performance of individual assessment versus JEM-based assessment.

Finally, restricting the creation of the Montreal JEM to information from the controls resulted in the use of only ∼60% of the Montreal exposure database. This was motivated by the concern that cases are not necessarily representative of the study base, especially in regard to exposure to carcinogens. Among recent uses of experts evaluations from past case–control studies to create JEMs, Peters et al27 excluded cases but ‘tMannetje et al24 included them. None of the authors provided empirical evidence supporting one approach or the other. We believe the question of including data from cases and controls, while not the focus of the present work, raises an important bias/precision trade-off issue, which should be addressed empirically in future studies through detailed comparisons of case and controls exposure estimates.

In conclusion, although there was a substantial potential for observing no agreement, we found at least moderate agreement between FINJEM and the Montreal JEM for more than half of the agents studied, well in the range of published intra- and inter-rater studies available in the literature. Welding fumes and iron were associated with good agreement consistently across comparisons based on prevalence, exposure intensity and in sensitivity analyses. These observations, while formally only representative of the Montreal area, offer optimistic insight on the transportability of FINJEM to similar settings for these agents, that is, urban Northern American areas. For other agents or settings, depending on the availability of local expertise, international users could use FINJEM as a starting point for developing region specific JEMs, such as was done for Scandinavian countries and New Zealand.10 24 Other extensive comparison efforts involving FINJEM should be performed to increase knowledge about its applicability in various settings. FINJEM seems to have a higher threshold for assigning exposure compared with the Montreal JEM as well as other JEMs based on individual evaluations. While the impact of this characteristic on dose–response relationships estimation might be minor when it is only important to have a well-defined ‘high exposed’ group, other applications such as the estimation of number workers exposed will be affected, especially for agents occurring infrequently within occupations.

Acknowledgments

The authors acknowledge and appreciate the contributions of several colleagues and assistants. Marie-Elise Parent of the INRS-Institut Armand-Frappier was partly responsible for directing the lung cancer study. The coding of occupations in the Montreal data was done by Ramzan Lakhani, BenoîtLatreille and Louise Nadon, who also contributed greatly to the development of the exposure assessment methods used in Montreal. Jordi Figuerola of the Centre for Research in Environmental Epidemiology (CREAL) in Barcelona provided the files necessary to link the Montreal data with FINJEM. Aude Lacourt provided helpful comments on the manuscript.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

Footnotes

  • Funding The INTEROCC study was supported by a grant from the National Institutes of Health (USA). The lung study was funded by grants from the Canadian Institutes for Health Research. The team was supported by the Guzzo-SRC Chair in Environment and Cancer. JL was supported by the Canadian Cancer Society Research Institutes and the Fond de la Recherche en Santé du Québec.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.