STATISTICAL-MODEL FOR PREDICTION OF RETROSPECTIVE EXPOSURE TO ETHYLENE-OXIDE IN AN OCCUPATIONAL MORTALITY STUDY

Since direct measures of individual exposure seldom exist for the enti re period of an occupational mortality study, retrospective exposure e stimates are necessary. This is often done in a subjective manner invo lving a consensus of opinion from a panel of epidemiologists and indus trial hygienists. An alternative method utilizing a statistical model provides a more objective procedure for retrospective exposure assessm ent. The development of a weighted multiple regression model is presen ted for estimation of exposure levels to ethylene oxide (ETO) for incl usion in a cohort mortality study of workers in the sterilization indu stry. Three steps in development of the model are described: (1) data acquisition and assessment, (2) model building, and (3) evaluation of the model. The final model explained a remarkable 85% of the variabili ty in 205 average measurements of ETO levels. Exposure factors include d in the model were exposure category, product type, size of the steri lization unit, selected engineering controls, days after sterilization , and calendar year. The model was evaluated in two ways: against a se t of measurement data not used to develop the model and a panel of 11 industrial hygienists representing the sterilization industry. The mod el predicted ETO exposures within 1.1 ppm of the validation data set w ith a standard deviation of 3.7 ppm. The arithmetic and geometric mean s of the 46 measurements in the validation data set were 4.6 and 2.2 p pm, respectively. The model also outperformed the panel of industrial hygienists relative to the validation data in terms of both bias and p recision. (C) 1994 Wiley-Liss, Inc.