Retrospective assessment of asbestos exposure--II. At the job level: complementarity of job-specific questionnaire and job exposure matrices.

Abstract

Retrospective assessment of asbestos exposure--II. At the job level: Complementarity of job-specific questionnaire and job exposure matrices. International Journal of Epidemiology 1993; 22 (Suppl. 2): S96-S105. The assessments of asbestos exposure by two a priori job exposure matrices (JEM) and by a job-specific questionnaire (SQ) are compared at job level. The data used for the comparison were generated by an ongoing case-control study on lung cancer in a region of northern Germany with a relatively high past prevalence of asbestos exposure. Among job periods assessed as unexposed by either JEM, 96% are recognized as such by the SQ. Discrepancies between the SQ and JEM were observed in jobs rated potentially exposed by the JEM. Despite varying estimates, the JEM and SQ were consistent as regards the relative classification of job periods by probability of exposure. The concordance of the methods, estimated by Kappa statistics, was stronger for the two JEM than for either of the JEM and the SQ. The identification of specific occupation/industry combinations in which discrepancies were most frequent and the comparison with expert ratings in some jobs yield insights into the sources of the disagreement between the methods. The misclassification of exposure by the JEM usually results in an overestimation of exposure. This is essentially related to loss of information due to the use of job codes as surrogates for job task descriptions and to the insufficiency of published data on asbestos exposure in different industries. As regards the SQ, two main sources of potential loss of sensitivity were identified: 1) possible omission of indirect sources of exposure by this method, 2) possible incompleteness of the SQ. The present comparison of methods of asbestos exposure assessment does not allow any one approach to be considered superior to another. Indeed, as proposed by Ahrens et al. in Part I of the study, both should be used to ensure optimal epidemiological performance.