Reassessing benzene risks using internal doses and Monte-Carlo uncertainty analysis

Human cancer risks from benzene have been estimated from epidemiological data, with supporting evidence from animal bioassay data. This article reexamines the animal-based risk assessments using physiologically based pharmacokinetic (PBPK) models of benzene metabolism in animals and humans. Internal doses (total benzene metabolites) from oral gavage experiments in mice are well predicted by the PBPK model. Both the data and the PBPK model outputs are also well described by a simple nonlinear (Michaelis-Menten) regression model, as previously used by Bailer and Heel [Metabolite-based internal doses used in risk assessment of benzene. Environ Health Perspect 82:177-184 (1989)]. Refitting the multistage model family to internal doses changes the maximum-likelihood estimate (MLE) dose-response curve for mice from linear-quadratic to purely cubic, so that low-dose risk estimates are smaller than in previous risk assessments. In contrast to Bailer and Heel's findings using interspecies dose conversion, the use of internal dose estimates for humans from a PBPK model reduces estimated human risks at low doses. Sensitivity analyses suggest that the finding of a nonlinear MLE dose-response curve at low doses is robust to changes in internal dose definitions and more consistent with epidemiological data than earlier risk models. A Monte-Carlo uncertainty analysis based on maximum-entropy probabilities and Bayesian conditioning is used to develop an entire probability distribution for the true but unknown dose-response function. This allows the probability of a positive low-dose slope to be quantified: it is about 10%. An upper 95% confidence limit on the low-dose slope of excess risk is also obtained directly from the posterior distribution and is similar to previous q(1)* values. This approach suggests that the excess risk due to benzene exposure may be nonexistent (or even negative) at sufficiently low doses. Two types of biological information about benzene effects-pharmacokinetic and hematotoxic-are examined to test the plausibility of this finding. A framework for incorporating causally relevant biological information into benzene risk assessment is introduced, and it is shown that both pharmacokinetic and hematotoxic models appear to be consistent with the hypothesis that sufficiently low concentrations of inhaled benzene do not create an excess risk.