From expert-based to quantitative retrospective exposure assessment at a Soderberg aluminum smelter.

OBJECTIVES Expert judgement of exposure levels is often only poorly or moderately correlated with directly measured levels. For a follow-up of a historical cohort study at a Söderberg aluminum smelter we updated an expert-based semiquantitative job exposure matrix of coal tar pitch volatiles (CTPV) to quantitative estimates of CTPV and benzo(a)pyrene (BaP). METHODS Mixed effects models to predict exposure for potroom operation and maintenance jobs were constructed from personal CTPV and BaP measurements. Mean exposures of jobs in non-potroom locations were directly calculated when measurements were available. Exposure estimates for jobs/time periods with no measurements were based on proportion of time spent in exposed areas compared to jobs where exposure was modeled or measured. For pre-1977, the original expert exposure assignments were calibrated using the updated 1977 estimates. RESULTS The rate of change in exposure levels varied by time period and was accounted for in mixed models with a linear spline time trend. Other variables significant in the models were job, potroom group and season as fixed effects, and worker as a random effect. The models for potroom operations explained 45 and 27% of the variability in the CTPV and BaP measurements, respectively. The models for maintenance jobs explained 40 and 19% of the variability in the CTPV and BaP measurements, respectively. For 1977-2000 model estimates, direct calculation of means and extrapolation from modeled/measured exposures accounted for 57, 6 and 37% of the exposed person-years, respectively. CONCLUSIONS The above methodology maximized the use of exposure measurements and largely replaced the original expert-based estimates. Finer discrimination between exposure levels was possible with the updated exposure assessment. The new estimates are expected to reduce exposure misclassification and help better assess the exposure-response relationships.