Validity of exposure in one job as a surrogate for exposure in a cohort study

Frequently, information pertaining to only one job is available or used to evaluate risk estimates of disease in occupational epidemiologic research. The amount of misclassification that such a practice could create has not, however, been examined. We used data from a mortality study of workers employed in 10 formaldehyde-producing or -using plants to address how closely several parameters of exposure based on the first, longest, or last job held in a company compared with those based on the worker's entire employment history at the plant. The best predictor for cumulative formaldehyde exposure at the plant was the longest job at that plant, with a correlation coefficient (r) of 0.70. The correlation with average exposure over the worker's employment was 0.77 for the first job and 0.74 for the longest and last jobs. Peak exposures and highest exposure levels experienced in the plant were more closely related to the first job (r = 0.72 and r = 0.74). The highest correlation with any of the measures was never with the last job. Variation between plants for each of these comparisons, however, was wide. These findings indicate that the use of a single job as a surrogate for exposure received at a particular worksite can result in different degrees of misclassification for different exposure measures. Even though the correlations were generally high, the associated misclassification of exposure could lead to a substantial underestimation of the relative risks in some situations. In this report two hypothetical examples show what effect the misclassification rates could have on estimates of disease risks.