Applicability of alternative models of revertant variance to Ames-test data for 121 mutagenic carcinogens

Abstract

Models of sampling variance in replicate revertant scores play a role in analyses of Ames-test data on mutagenicity in Salmonella, both in modeling the dose-response relation and in estimating initial dose-response slope or ‘potency’, e.g., for use in correlating mutagenic and carcinogenic potencies among different chemicals. Both generalized Poisson (GP) and negative binomial (NB) models of revertant variance have been applied in this way, but their empirical applicability has only been assessed using Ames-test data on a few chemicals. The applicability of these and related variance models was therefore assessed for 1905 such data sets pertaining to 121 putatively mutagenic carcinogens. Only ∼ 50% of the data sets analyzed were found to involve a significantly positively correlated dose-response, and < 50% were found to exhibit a plausibly heterogeneous response variance regardless of dose-response correlation. Among data sets with plausibly heterogeneous variance, < 60% were found to exhibit significantly extra-Poisson variability. Among the significantly extra-Poisson data sets, most (> 75% among dose-response correlated data sets) were found to exhibit revertant variance consistent with both the GP and NB models; while the GP model was found to be somewhat more consistent with these data, the NB model more often gave a nominally better fit when both models were consistent. Implications of these results for the design of methods used to analyze Ames-test data are discussed.