Article Text

Original article
Comparison of exposure assessment methods in a lung cancer case-control study: performance of a lifelong task-based questionnaire for asbestos and PAHs
  1. Eve Bourgkard1,
  2. Pascal Wild1,2,
  3. Maria Gonzalez3,
  4. Joëlle Févotte4,
  5. Emmanuelle Penven2,
  6. Christophe Paris2,5
  1. 1Département Epidémiologie en Entreprise, Institut National de Recherche et de Sécurité (INRS), Vandœuvre-lès-Nancy, France
  2. 2INGRES, EA 7298, Université de Lorraine, Vandœuvre-lès-Nancy, France
  3. 3Hôpitaux Universitaires de Strasbourg, Centre de consultation de pathologie professionnelle, Strasbourg, France
  4. 4Unité mixte de recherche épidémiologique et de surveillance en transport, travail et environnement—Umrestte (UCB Lyon 1/Inrets), Université Claude Bernard Lyon 1 8, Lyon, France
  5. 5Centre Hospitalier Universitaire de Nancy, Centre de consultation de pathologie professionnelle, Vandœuvre-lès-Nancy, France
  1. Correspondence to Dr Eve Bourgkard, Département Epidémiologie en Entreprise, INRS, 1 rue du Morvan, CS60027, Vandœuvre-lès-Nancy 54519, France; eve.bourgkard{at}


Objective To describe the performance of a lifelong task-based questionnaire (TBQ) in estimating exposures compared with other approaches in the context of a case-control study.

Methods A sample of 93 subjects was randomly selected from a lung cancer case-control study corresponding to 497 jobs. For each job, exposure assessments for asbestos and polycyclic aromatic hydrocarbons (PAHs) were obtained by expertise (TBQ expertise) and by algorithm using the TBQ (TBQ algorithm) as well as by expert appraisals based on all available occupational data (REFERENCE expertise) considered to be the gold standard. Additionally, a Job Exposure Matrix (JEM)-based evaluation for asbestos was also obtained. On the 497 jobs, the various evaluations were contrasted using Cohen's κ coefficient of agreement. Additionally, on the total case-control population, the asbestos dose-response relationship based on the TBQ algorithm was compared with the JEM-based assessment.

Results Regarding asbestos, the TBQ-exposure estimates agreed well with the REFERENCE estimate (TBQ expertise: level-weighted κ (lwk)=0.68; TBQ algorithm: lwk=0.61) but less so with the JEM estimate (TBQ expertise: lwk=0.31; TBQ algorithm: lwk=0.26). Regarding PAHs, the agreements between REFERENCE expertise and TBQ were less good (TBQ expertise: lwk=0.43; TBQ algorithm: lwk=0.36). In the case-control study analysis, the dose-response relationship between lung cancer and cumulative asbestos based on the JEM is less steep than with the TBQ-algorithm exposure assessment and statistically non-significant.

Conclusions Asbestos-exposure estimates based on the TBQ were consistent with the REFERENCE expertise and yielded a steeper dose-response relationship than the JEM. For PAHs, results were less clear.

  • Lung cancer < Organ system, disease, disease type
  • Task-based exposure assessment < Methodology, speciality

Statistics from

What this paper adds

  • Our study suggests that a short questionnaire based on a limited list of tasks, used in a case-control study, has allowed an exposure assessment consistent with a reference expertise.

  • The potential of the task-based questionnaire to extract the information depends on the nature of the exposure to evaluate. Better results can be obtained for agents that can be easily recognised by the participant of the study, for example, asbestos.

  • The dose-response relationship between the cumulative exposure to asbestos based on the task questionnaire (algorithmic evaluation) and the lung cancer incidence, shows that this type of questionnaire may offer a satisfactory alternative to the job exposure matrix for occupational exposure assessment in case-control studies.

  • The paper proposes a list of 19 questions which allowed a quantification of asbestos exposure consistent with a thorough expert assessment.


Good assessment of occupational exposure in epidemiological studies, and especially in case-control studies on cancer, is essential to limit misclassification of exposure. Even if such misclassification is non-differential, it can lead to a bias towards the absence of relationship between disease and exposure when exposure is dichotomous, or it can lead to a bias in any direction resulting in distorted exposure-disease relationships when a continuous exposure is categorised. This retrospective evaluation remains difficult to achieve and several methods are typically used, such as occupational histories, job exposure matrix (JEM), self-reports, expert assessments, measurement data, or a combination of these methods.1–6

The expert assessment is usually considered as the reference method.1 ,3 It uses all occupational data collected during the study, especially occupational histories with jobs held, sectors of activity, and period of activity as well as data from questionnaires for which the quality and quantity of information gathered can be, however, very different from one study to another. The better the experts and the greater the quantity of data, the better the expertise will be. However, expert assessment has its limitations notably in terms of cost and expert-time availability.1 ,3 Thus, a balance is needed between the amount and the quality of the collected data. In what follows, we distinguish between two types of questionnaire: the first type relies on specific questionnaires adapted to jobs or professional sectors used in several studies.7–12 In this approach, the study subject fills in several questionnaires, triggered for each job mentioned in his curriculum laboris, with a large number of items requiring a significant investment in time and fatigue which can lead to incomplete interviews for seriously ill cases.

Another approach relies on a questionnaire detecting exposing tasks performed during the entire job history.13–15 Based on this idea, one of us (CP) developed a task-based questionnaire (TBQ) which can be more concise than the preceding job-specific ones. Thus, a set of questions is asked once for the whole career, through establishing periods of homogenous tasks according to their frequency and period of time. Therefore, this type of questionnaire has the advantage of being shorter and thus potentially more acceptable to study participants, especially those with health problems.

In order to use the information obtained by using these questionnaires (specific and TBQ) a posthoc estimation of exposures must be carried out which can either be based on an expert assessment or on an automatic method using an algorithm based on all the closed-form questions.16 ,17 Many authors have highlighted the limitations of the approach by expert judgment because of its complexity and its cost in time and human or financial resources especially for large studies,1 ,3 and also for reasons of its intrinsic validity.18 ,19 The use of an algorithm assigning exposure automatically would save time and reduce logistical constraints but it is not clear how critical the information loss through this procedure is.

The aim of the present paper is to compare exposure estimates based on a TBQ (expertise or algorithm) to those obtained through applying traditional approaches: a REFERENCE expertise based on all available exposure information considered as the gold standard, and a general population JEM. These comparisons involved two agents: asbestos and polycyclic aromatic hydrocarbons (PAHs).


The case-control study

Study design

A population-based lung cancer case-control study was conducted in the northern part of the French Lorraine region, characterised by a past of heavy industries (iron mining, coal mining, steel-producing factories and a petrochemical complex). The study protocol and details on the selection of cases and controls have been previously reported.20 The study included male cases and controls aged 40–80 years, living in the study region and giving written statements of informed consent. The cases had histologically confirmed lung cancer and were recruited in the hospitals of the area. The controls were selected by a random digit dialling procedure and stratified by age-classes, administrative districts and seven large socioeconomic classes according to the last held job. All cases and controls filled in a series of questionnaires in a face-to-face interview on non-occupational factors, lifetime job history and occupational exposures. The case-control study comprised 246 cases and 531 controls.

The study protocol was approved by the ethics committee of the French national data protection agency devoted to human studies.

Occupational questionnaires

A lifetime job history was obtained for all jobs (job title, starting and finishing dates and company name) held for at least 3 months. For each job, the participant described the main tasks performed and the company's activity. These jobs were coded using the ISCO-68 code and the NAF 2000 (the French activities coding system similar to the European NACE coding scheme).

For each job, a general exposure questionnaire and possibly one or more specific exposure questionnaires (we refer to as job-specific questionnaires) were applied. The use of these specific questionnaires was triggered (or not) by the job name or activity mentioned in the job history. Altogether, 20 questionnaires were used, corresponding to jobs or occupational sectors. These specific questionnaires had been developed for the ICARE study12 by JF. They consist of a series of questions concerning specific agents or describing the tasks carried out for this job by the worker himself or the neighbouring workers, the corresponding frequencies and the use of personal or collective protection devices.

In addition to the specific questionnaires, another questionnaire (developed by CP, which we refer to as the TBQ) was used, which asked about 47 tasks of the whole work history regardless of each job. These tasks were identified as the ones potentially exposed to recognised or suspected pulmonary carcinogens. Of these 47 tasks, 15 concerned exposure to asbestos (see online supplementary appendix 1) and five to PAHs (see online supplementary appendix 2). For each question or task, the whole job history was divided into homogeneous exposure periods with respect to the task and the task frequency. Thus, each exposure period could group together several jobs, or alternatively, consist of only a fraction of a job if the task frequencies had changed over the time within this job. For each of these exposure periods, dates of beginning and end, as well as a task frequency code (percentage of working time: <1%, 1–9%, 10–29%, 30–69%, >69%) were obtained from the subject.

Exposure assessment validation study

Study sample

All subjects (134 cases and 530 controls) interviewed after 3 years of inclusion were selected, then randomly sorted. Exposures were assessed for the first 93 subjects only (41 cases and 52 controls) since experts spent their time devoted to the study.

For the purpose of comparing the various occupational exposure assessments for asbestos and PAHs, all jobs performed by each subject were gathered to obtain a group of 497 jobs. During the occupational exposure assessment process, some jobs were divided into several periods according to time trends of exposure levels.

Exposure assessments

The occupational exposures to asbestos and PAH were assessed for the 497 jobs using the TBQ in the form of (1) an expert-based assessment (TBQ expertise) or (2) an algorithm-based assessment (TBQ algorithm) and were compared with those of (3) an expert based on all job questionnaires (REFERENCE expertise) considered as the gold standard, and (4) a general population asbestos JEM. The TBQ algorithm was the only assessment method used in Wild et al.20

During the various exposure assessment strategies, experts categorised each job-period into the following groups 0: non-exposed, 1: <1, 2: 1–10, 3: ≥10 f/mL for asbestos, and 0: non-exposed, 1: low, 2: medium, 3: high for PAHs.

1-Expert assessment based on TBQ (TBQ expertise)

The exposure assessment was conducted in two stages. In a first step, two experts, professors of occupational medicine (MG, CP) independently assessed exposures based on the job history, the TBQ, and for each job, the description of the main task performed, and the company's activity. In a second step, the exposure codes assigned by the two experts were compared. Thus, job-periods whose probability was coded ‘certain’ by at least one of two experts or ‘probable’ by two experts were considered as exposed, and the other job-periods were deemed ‘unexposed’. For job-periods considered as exposed with no difference of more than one class on the level or frequency between the two experts, the lowest assessments were assigned. For job-periods considered to be exposed, expert opinions were discordant if a difference of more than one class was observed. These discrepancies were analysed job-period by job-period at plenary consensus meetings, 1 or 2 years apart from the first expertise (see online supplementary 1). Three experts (MG, CP, and one senior occupational epidemiologist (EB)) obtained a consensus exposure code for all discordant exposure levels.

2-Algorithmic assessment based on TBQ (TBQ algorithm)

Two experts (CP and EB) set up an algorithm which assigned an exposure level to each of the tasks identified by the 15 asbestos questions and the 5 PAH questions.

3-Expert assessment based on all job questionnaires (REFERENCE expertise)

First, for each job, the past occupational exposures were independently coded by two occupational hygienists, differing from the previous TBQ expertise, based on all available information, that is, the lifelong job history and for each job, a description of the main tasks performed by the worker and a description of the company's activity as well as the responses from the study participants to the job-specific questionnaires and to the TBQ. During the expertise, some jobs were divided by the experts into several periods to take into account time trends of exposure levels. The same judgment criteria were given to the two experts at the beginning of the study and, to avoid a drift in assessment, regular checks were organised during the coding process. Second, the discordances between the two evaluations were analysed by the same second group of three experts as for the TBQ expertise and a consensus exposure code was obtained, independently of the first consensus expertise.

4-Asbestos JEM

A general population JEM has been developed by the Department of Occupational Health of the French Institute for public health surveillance with assessments for several occupational exposures including asbestos (Matgéné program),21 rating all combinations of job and activity codes as assessed in the ISCO68 and NAF coding system which were split into several periods of time in order to take into account the variation in exposure over time. Asbestos was assessed using two items: exposure probability and exposure intensity (0: non-exposed, 1: 0.0001–0.01, 2: 0.01–0.1, 3: 0.1–1, 4: 1–10, 5: ≥10 f/mL). We only kept exposures coded in the JEM as either probable or certain. In our study, the job-periods were merged based on ISCO and NAF codes with the intensity items of the JEM. These items were simplified by combining levels to obtain the same scale as with the other methods (0: non-exposed, 1: <1, 2: 1–10, 3: ≥10 f/mL).

Unfortunately, no comparable PAH matrix was as yet available to estimate exposures via this type of approach.

Comparison of asbestos-exposure estimates in the case-control study

Exposures of each subject of the case-control study (246 cases and 531 controls) were assessed using TBQ algorithm and JEM, the only methods able to estimate exposures of all subjects.

Statistical analysis

All analyses were conducted using Stata V.12 (StataCorp, College Station, Texas, USA). Since the objective was to assess the performance of the TBQ, the main analysis compared the exposures assessed from the TBQ (expertise and algorithm) with the REFERENCE expertise, which was taken as a gold standard. Sensitivity and specificity were computed by grouping together the semiquantitative levels of exposures into two groups (non-exposed: 0; exposed: 1–3).22 Furthermore, percentages of exact agreement and Cohen's κ statistic of agreement were calculated (Stata command: kap). Exposure was also studied in the form of the semiquantitative variable (4 levels). In this case, the measurement of intermethod agreement was calculated using Cohen's κ and its weighted version (w-κ) (Stata command: kap wgt(w)) and 95% CIs (Stata command: kapci).23 ,24 Weighted κ gives also some weight to observations that are close, but do not agree exactly. We adopted a weighting scheme that for the ijth cell of an r×r agreement table is defined by: wij=1 − |i − j|/(r − 1). Thus, a single category difference in agreement was given a weight of 0.66, and a two category difference in agreement was given a weight of 0.33.

A second objective was to compare the TBQ assessments with the JEM-based exposure estimates. This was done for asbestos in two series of analyses. The first contrasted the TBQ estimates (expertise and algorithm) with the JEM-exposure estimates in the subsample of 497 jobs as previously. The second compared the TBQ-algorithm asbestos-estimates with the JEM-asbestos estimates in the model applied to all subjects of the case-control study discussed in the previous publication.20 Briefly, for the purpose of the analysis of the case-control studies, cumulative exposure indices were computed for asbestos, PAHs, and silica by first assigning to each job-period a quantitative exposure level (obtained from the highest semiquantitative assessment with respect to tasks of this job obtained applying the TBQ-algorithmic assessment) which was multiplied by the estimated frequency of this highest exposed task and the duration in years of the period. The subject-specific cumulative exposure was then computed as the sum of the corresponding job-period specific cumulative exposures. The resulting index was log-transformed after adding the constant 1, so that non-exposed subjects had a zero exposure. The logarithmic transformation is justified by the fact that the indicative boundaries of the exposure levels were increasing geometrically for asbestos, 1: <0.1; 2: 0.1–1; 3: 1–10; 4: >10 f/mL. Since the same exposure levels as well as frequency codes were assigned by the JEM, we computed a corresponding JEM-based cumulative score. Two logistic regression analyses were performed for each asbestos cumulative estimate: The first was adjusted only on non-occupational confounders (including smoking) and the second one, also on occupational confounders, that is, cumulative exposures to PAHs and crystalline silica as well as ever exposure to diesel motor exhaust.


The proportion of asbestos-exposed job-periods assessed using the TBQ expertise is slightly lower than using the TBQ algorithm, but are quite similar to those obtained with the REFERENCE expertise or the asbestos JEM (table 1). In the case of PAHs, these proportions are lower when using the TBQ (expertise or algorithm) compared with the REFERENCE expertise.

Table 1

Percentages of job-periods considered exposed for asbestos and PAH

When comparing asbestos exposures assessed using the TBQ expertise to those obtained using the REFERENCE expertise (table 2A), 431 job-periods were classified in the same exposed (levels 1–3)/non-exposed (level 0) groups (Agreement: 90.0%, κ: 0.70). The specificity of this TBQ expertise (96%) was better than the sensitivity (69%). With respect to the exposure levels, the two methods were 86.4% in agreement (w-κ: 0.68) and no discordance was greater than one level. Compared to the REFERENCE expertise, the computerised algorithm based on the TBQ (table 2B) classified 410 job-periods in the same exposed/non-exposed groups (Agreement: 90.4%, κ: 0.73), with higher specificity (95%) than sensitivity (77%). As regards the levels, 81.7% agreement was observed between the two methods (w-κ: 0.61) with one or two levels of discordance. Similar results were observed when comparing assessments of the TBQ algorithm to the TBQ expertise (see online supplementary 2).

Table 2

Asbestos-exposure estimates determined by the TBQ (expertise, algorithm) compared with the REFERENCE expertise

When excluding the non-exposed job-periods according to the REFERENCE expertise, the weighted-κ were lower (TBQ expertise/REFERENCE expertise: w-k 0.29; TBQ algorithm/REFERENCE expertise: w-k 0.21), but Spearman's rank correlation remained highly statistically significant (p<0.0001) (data not shown).

With respect to PAHs (table 3), 391 job-periods were classified with the TBQ expertise (table 3A) in the same exposed/non-exposed groups as the REFERENCE expertise (Agreement: 84.5%, κ: 0.47), with higher specificity (94%) than sensitivity (48%). As regards the levels, the results of the two expert appraisals agreed in 78.5% (w-κ: 0.43) with up to 2-level discordances (2-level discordance: 20 job-periods, 4%).

Table 3

Polycyclic aromatic hydrocarbon exposure estimates determined by the TBQ (expertise, algorithm) compared with the REFERENCE expertise

Of the 502 jobs-periods, the REFERENCE expertise and the TBQ algorithm (table 3B) classified 396 job-periods in the same exposed/non-exposed groups (Agreement: 83.2%, κ: 0.40) with much lower sensitivity (39%) than specificity (95%). Regarding the levels, the two methods were 78.9% in agreement (w-κ: 0.36), with up to 3-level discordances (2-level discordance: 17 job-periods, 3.4% and 3-level discordance: 12 job-periods, 2.4%).

Based on the same TBQ (see online supplementary 3), the algorithm and the expertise classified 418 job-periods in the same exposed/non-exposed groups (Agreement: 89.2%, κ: 0.54). With respect to exposure levels, the agreement is lower with 83.3% (w-κ: 0.37), with up to 3-level discordances (2-level discordance: 27 job-periods, 5.4% and 3-level discordance: 12 job-periods, 2.4%).

When excluding the non-exposed job-periods according to the REFERENCE expertise, the weighted-κ were lower (TBQ expertise/REFERENCE expertise: w-k 0.13; TBQ algorithm/REFERENCE expertise: w-k 0.14) with statistically significant Spearman's rank correlation (p<0.01) (data not shown).

The agreement observed between the JEM and the REFERENCE expertise (see online supplementary 4) was lower than that observed between the TBQ (expertise or algorithm) and the REFERENCE expertise (table 2). Compared to the JEM, the TBQ expertise (table 4A) classified 78.7% (n=386) of the job-periods in the same exposed/non-exposed groups (κ: 0.36) and 76.1% in the same levels (w-κ: 0.31). When comparing the two computerised approaches, JEM and TBQ algorithm (table 4B), 362 job-periods were classified in the same exposed/non-exposed groups (Agreement: 77.3%, κ: 0.36) and 70.9% were classified in the same level groups (w-κ: 0.26).

Table 4

Asbestos-exposure estimates determined by the TBQ (expertise, algorithm) compared with the general population JEM

In the case-control study analysis, the dose-response relationship between lung cancer and asbestos-exposure estimates based on the TBQ algorithm was steeper (OR1: 1.27, 95% CI 1.15 to 1.40) than using exposure estimates based on the JEM (OR1: 1.10, 95% CI 0.98 to 1.24) and statistically significant (table 5). This was particularly apparent when adjusting for occupational confounders. In this case, the JEM-based asbestos dose-response relationship disappeared altogether (OR2JEM: 1.02, 95% CI 0.91 to 1.16).

Table 5

Adjusted ORs for lung cancer according to JEM-based and TBQ asbestos-exposure assessments in the case-control study


The present study shows that the exposure assessment based on a limited list of tasks known or suspected to constitute risks for lung cancer (TBQ) shows good consistency with a REFERENCE exposure assessment with respect to asbestos. For PAHs, this consistency is still acceptable but less good. The consistency between the asbestos-exposure levels as assessed by the JEM and those based on the TBQ assessment is still high but less so than between the various questionnaire-based assessments (TBQ expertise and algorithm vs REFERENCE expertise). The dose-responses with cumulative-asbestos based on the JEM are less steep and not significant compared with the TBQ algorithm-based assessment.

This paper shows, with an example, that our TBQ has the potential to extract most of the information with respect to occupational exposure. A similar result was obtained in a case-control study in which exposures were estimated using, among other methods, a checklist of silica-specific jobs and tasks.13 Arguments why this could be expected have been put forward by Teschke et al.1 They argue in particular that better responses are obtained by placing the subject in a known work environment, and by questioning the study participants about their actual tasks, irrespective of the context. One could also argue that in the context of case-control studies, objective questions with respect to the tasks carried out are less sensitive to differential recall bias. Finally, a TBQ may detect exposing tasks which are not typically performed in a given job and which would thus possibly be missed in specific questionnaires. On the other hand, the efficiency of TBQ is based on the fact that the exposure depends mostly on the actual task. The tasks proposed as choices must also carry enough information for a posteriori rating of the exposure25 which has to be balanced with the length of the questionnaire. Given the results shown above, one could argue that for asbestos, but possibly not yet for PAHs, the right balance has been reached. For other agents however, tasks are not the dominant predictor of exposure. This might, for instance, be the case for diesel exhaust exposure which is usually not directly related to the task carried out.

Although our results seem to indicate that a TBQ approach is a promising way forward in obtaining efficient and reliable exposure information, much depends on the list of tasks included and the exposure considered. In what follows, we discuss in more detail the TBQ we used. Regarding asbestos, the specificity of the TBQ expertise was very high indicating that the tasks identified were confirmed using the expert assessment based on the full exposure information (REFERENCE expertise). The slightly lower sensitivity indicated that our TBQ did not identify some exposures. Listing the exposures missed by the TBQ, we identified tasks not present in our questionnaire: demolition of ovens, chimneys or coatings containing asbestos, asbestos-exposure activities in industrial maintenance in the construction sector for whom we propose additional questions (see online supplementary appendix 1). With respect to PAHs, the results were less consistent, in particular with low sensitivity indicating that the five questions, despite being relatively general, missed a substantial part of the exposing tasks. Job-specific questionnaires asking more specific questions could better identify some exposing circumstances. This applies for questions about the use of grease, used oils, fumes or dust from asphalt/bitumen, combustion soot, heated or sprayed oils, while in our TBQ, there is only a single question on the use of cutting oils. There is also an intrinsic difficulty in assessing exposure to PAHs. PAHs are a complex family of products often generated by neo-formation, and whose presence is highly variable depending on the period of activity, temperature of implementation and working conditions. This heterogeneity in exposure may explain the difficulty, even for experienced experts, of encoding this type of pollution, especially for exposure levels.25 ,26 Similarly to asbestos, we were able to identify six other questions (see online supplementary appendix 2) by exploring the exposing situations missed by our TBQ, and also building on another questionnaire used for identifying bladder carcinogens.27

In what precedes, we considered the expert assessment based on all data (REFERENCE expertise) as the gold standard. Although many authors,1 ,25 including us, tend to consider expert-assessed as the best especially in the context of case-control studies, whether an expert-assessed exposure can be considered as a gold standard is a disputed topic28 ,29 and the evidence is conflicting. When comparing expert assessment with measurements in biological samples or in the work environment, studies show quite different outcomes, from good correlations to none at all.1 ,6 ,18 ,30 ,31 Moreover, studies examining agreement between experts’ ratings show κs or intraclass correlation-coefficients ranging from 0 to 1.0 with a median of about 0.6.1 However expert assessment can only be as good as the experts and data provided to them.32 In our study, experts were given detailed and full information coming from job histories as well as both series of questionnaires. Nevertheless, differences in assessments were observed between the two initial experts. This was mostly due to differences in experience and depended on the type of agent coded, asbestos or PAH. In the case of asbestos, occupations or work situations exposed to asbestos are fairly well known in the literature. The agreement between the two experts is fair to good (non-exposed and exposed groups: agreement=88.0%, κ=0.60, data not shown). On the other hand, the PAH levels were more discordant (non-exposed and exposed groups: agreement=84.3%, κ=0.39, data not shown) owing to the already mentioned difficulties in assessing PAH exposure. However, we are quite confident that the second round of expertise, by consensus of a group of three more senior experts, enabled these discordances to be resolved successfully.

In comparison with the traditional approach, collecting descriptive information about all tasks, environments, processes for each job, we concede that in contrast with the TBQ, it allows experts to make inference about virtually any chemical as has been shown in the numerous papers based on expert information in the Montreal case-control studies.33–35

As an alternative to this REFERENCE expertise, we also considered a JEM-based assessment. This comparison was restricted to asbestos as no PAH-JEM was available using the French activity coding scheme. As expected, the agreement between the TBQ and the JEM-based assessments was less good than the corresponding agreement between TBQ and REFERENCE expertise. The same exposure assessment is attributed to all workers with the same job without taking into account the local conditions.1 ,29

None of the preceding comparisons can actually state which assessment is more relevant. Assuming that there is a dose-response relationship between cumulative asbestos exposure and lung cancer incidence, a more accurate exposure estimation should lead to a steeper and significant dose-response relationship.36 Thus, the statistical significance of this dose-response relationship and, to a lesser degree, the estimated slope in the analysis of the case-control data can be considered as external information in discriminating between the various assessments of the dose. The steeper slope found in our study favours the TBQ-algorithm approach in comparison with the JEM approach. An interesting feature is the fact that the JEM-based index seems to be more sensitive to confounding by other occupational carcinogens. The TBQ-based slope decreased from 1.27 to 1.18 after adjustment and remained significant, while the JEM-based slope decreased from 1.10 to 1.02, that is, it virtually disappeared. This could be due to the lack of specificity of the JEM in the jobs which might (or might not) be exposed to several carcinogens, these jobs being better discriminated based on the tasks performed.

To ensure the exposure assessments were independent, two different teams of experts encoded exposures, one for the REFERENCE expertise (NB, GR) and the other one, for the TBQ expertise (MG, CP). However, the analyses of discordances between experts for the REFERENCE and the TBQ-expert appraisals were not fully independent because they were conducted by the same team of experts (MG, CP, and EB), 2 months apart (see online supplementary 1). It should be noted that these discrepancies concerned few jobs (REFERENCE expertise: 45 job-periods for asbestos and 91 job-periods for PAH; TBQ expertise: 34 job-periods for asbestos and 46 job-periods for PAH) and were not necessarily the same between the two expert appraisals. However, discrepancy analysis with the same team of experts, meaning with the same experience and the same knowledge, would probably avoid bias when comparing assessments between the two methods. Indeed had two different teams assessed the different sources of information, the differences would have been confounded with differences between experts.

A final discussion point is whether the differences depended on the case-control status. Assuming that the higher exposure level would be easier to identify, the agreement between different assessments might have been expected to be better among cases as they are more exposed than the controls. In our limited dataset of 93 cases and controls, we observed the reverse (data not shown). This might be due to a differential recall bias whereby cases try harder to recall occupational exposures. Thus, questions about tasks that do not mention the exposing agents might be considered better. We must admit that this does not apply to all the questions in our TBQ which explicitly mentions asbestos exposure in some questions. Another explanation could be that a better agreement can be easily obtained in the case of low or short exposures, or non-exposure.

As a conclusion, we consider that the present comparisons show that a TBQ is a promising way to assess exposure to carcinogens in case-control studies.


The REFERENCE expertises were expertly performed by two industrial hygienists, Nicolas Bonnet (NB) and Grégory Rollin (GR). We would like to thank them for their availability. We thank Pierre Goutet, a senior chemical engineer, for his help in quantifying exposure levels to PAH, based on literature and numerous measurements performed in many companies. We acknowledge the French Institute for Public Health Surveillance (InVS) for providing us with the asbestos Job Exposure Matrix (Matgéné program). This study would not have been possible without the able participation of interviewers and data managers: Aurélie Bannay, Christine Bertrand, Julie Corvisier, Mathieu Dziurla, Gaëlle Feicht, Maryvonne Fournier and Monique Veillé. We thank the study subjects who had the patience to answer our questionnaires. We would like to thank the reviewers who have significantly improved our paper.


Supplementary materials


  • Contributors EB and PW drafted the manuscript. EB analysed the data. CP and PW designed the study. CP, EB, MG and PW participated in the analysis plan. CP developed the task-based questionnaire. JF developed the job-specific questionnaires. CP, EB and MG managed the exposure assessment discordances. All coauthors contributed to the interpretation of the results and discussion, and revised the paper.

  • Funding The study interviewers were funded by the French ANSES funding agency (grant number EST 07-17)

  • Ethics approval The French national data protection agency devoted to human studies.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval Ethics committee of the French national data protection agency devoted to human studies.

  • 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.