Abstract
The ability to define dose-response relationships in occupational epidemiology depends critically upon the measure(s) of exposure used. If statistical measures are employed, then the lognormal distribution provides a convenient mechanism for categorizing exposure within and between individuals. Preliminary analysis of exposure data obtained from 31 groups of workers in 9 industrial processes indicates that both components of exposure variability can be large. Thus, significant misclassification can arise either from systematic differences in exposure between individuals within a given job category (e.g., measuring the wrong people) or from changes in exposure which occur over time (e.g., measuring the wrong days).
Regarding the relative merits of the mean versus the “peak” exposure, a toxicokinetic model indicates that cumulative exposure (the mean exposure times the duration) should be a valid predictor of damage as long as the rates of elimination and repair are first order. Even when the relationship between burden and damage is nonlinear (as represented by a threshold burden), the amount of damage is unlikely to be significantly affected by “peak” exposures as long as the mean exposure is less than one-fourth to one-eighth of that which gives rise to the threshold burden.
It is concluded that the goals of exposure assessment in occupational epidemiology are twofold. First, it is necessary to obtain unbiased estimates of the mean exposure (not the geometric mean or median) received by each group under investigation. Second, it is desirable to derive information concerning the homogeneity of exposure within each group. These goals can be achieved through prospective sampling but are only rarely met in retrospective studies owing to the generally poor quality of the historical database. Rappaport, S.M.: Selection of the Measures of Exposure for Epidemiology Studies. Appl. Occup. Environ. Hyg. 6:448–457; 1991.