The wide variety of jobs encountered in population-based studies makes retrospective exposure assessment challenging in occupational epidemiology. In this analysis, two methods for estimating exposure intensity to chlorinated solvents are compared: rated (assigned by an expert rater) and modeled (assigned using statistical models). Estimates of rated and modeled intensities were compared for jobs held by mothers participating in the National Birth Defects Prevention Study with possible exposure to six chlorinated solvents: carbon tetrachloride, chloroform, methylene chloride, perchloroethylene, 1,1,1-trichloroethane, and trichloroethylene. For each possibly exposed job, an industrial hygienist assigned (1) an exposure intensity (rated intensity) and (2) determinants of exposure to be used in a statistical model of exposure intensity (modeled intensity). Of 12,326 reported jobs, between 31 (0.3%) and 746 (6%) jobs were rated as possibly exposed to each of the six solvents. Agreement between rated and modeled intensities was low overall (Spearman correlation coefficient range: -0.09 to 0.28; kappa range: -0.23 to 0.43). Although no air measurements were available to determine if rated or modeled estimates were more accurate, review of participants' job titles showed that modeled estimates were often unexpectedly high given the low-exposure tasks found in these jobs. Differences between the high-exposure jobs used to create the statistical models (obtained from air measurements in the published literature) and the low-exposure jobs in the actual study population is a potential explanation for the disagreement between the two methods. Investigators should be aware that statistical models estimating exposure intensity using existing data from one type of worker population might not be generalizable to all populations of workers.
Keywords: Agreement; chlorinated solvents; expert raters; exposure assessment; statistical models.