Empirical analysis of the variability of the maximum likelihood estimates of benzene cancer risks

This paper presents results from an on-going study of the carcinogenicity o f benzene at low doses. It focuses on the statistical variability of the pa rametric maximum likelihood estimator (MLE) of the parameters of the linear ized multistage (LMS) cancer dose-response model. This paper describes: 1) how pharmacokinetic conversions from animal to human were made to convert a dministered benzene to estimated internal doses of its metabolites; 2) the exact form of the LMS; 3) likelihood surfaces for fitting the exact form to data; and 4) results of bootstrap and Monte Carlo simulations that assess the variability of the parameters of the dose-response model. This paper su ggests that the cubic form of the LMS dose-response for animal tumors is ro bust to sampling variability. For environmental policy, the hypothesis that linearity governs low dose benzene carcinogenicity is probably incorrect. (C) 1999 Elsevier Science Ltd.