An experimental design approach to retrospective exposure assessment

There are several methods currently in use for retrospective estimation of quantitative exposure levels in occupational and environmental epidemiologic studies. The most popular is a job-exposure matrix approach using a combination of existing data and professional judgment. Another method is the use of statistical models based on available exposure data. The authors present an alternative approach using an experimental design in which several factors thought to affect exposure levels are identified and set at specific levels in a cross-classified design. This approach was used to estimate historical exposures to formaldehyde in a mortality study of embalmers. Exposures were estimated as a function of solution concentration, air exchange rate, and autopsied versus intact body. There were 12 combinations involving these 3 factors and a total of 25 embalming procedures (approximately 2 replicates of each combination) performed at a college of mortuary science. In addition to these design factors several covariates such as temperature, humidity, and the occurrence of spills were considered in an analysis of covariance statistical model. The results of the model prediction were validated against published measurements, and field samples were taken in several funeral homes. The overall accuracy of the model predictions was comparable to the variation found in replicate measurements of identical embalming procedures.