Estimating serum polychlorinated biphenyl levels in highly exposed workers: an empirical model

A regression model estimating high-homolog polychlorinated biphenyl (PCB) serum concentration on the basis of job exposure categorizations was developed. The model assumes first-order kinetics with a half-life determined empirically and uses variables that incorporate both intensity and duration of exposure over a 30-yr period. In order to compare the efficiency of these regression-based exposure estimates relative to often-used epidemiological parameters, models with dichotomized, ordinal, and continuous exposure surrogates were also investigated. Among the alternative exposure categorizations the most straightforward measure, ever versus never direct, was a particularly poor predictor of serum PCB level (r2 = .01). Nearly all of the candidate exposure measures we tried predicted serum levels poorly. The best of these after the fact was with total months employed in direct-exposure jobs (r2 = .43). None of the logical deductive models approached the predictability of the empirical model developed here (r2 = .69).