Analysis of Dose-Response Uncertainty using Benchmark Dose Modeling

Analysis of Dose-Response Uncertainty using Benchmark Dose Modeling Jeff Swartout U.S. Environmental Protection Agency Office of Research and Development National Center for Environmental Assessment October 11, 2007 I. Introduction and Methods I have assessed the uncertainty in modeling the test data sets using the U.S. EPA BMDS (Ver. 1.4.1) software (U.S. EPA, 2007), with a secondary verification using a parametric ® ®bootstrap algorithm in S-PLUS (Ver. 6.2 for Windows ). I present only the analyses of the (mythical) frambozadrine, nectorine and persimonate data sets, as they offer the most interesting issues for combining results. The issue of combining response data is first addressed by determining if the individual data sets are similar conceptually. That is, can they be considered to belong to the same population? The statistical aspect of pooling the data is addressed by assessing the net deviance (-2 x optimized log-likelihood ratio) between the dose-response model fits to the pooled data set and the individual data sets, assuming a specific dose-response family (Stiteler et al., 1993). The test statistic is D – ΕD , where D is the deviance of the pooled data model fit and ΕD is the pool i pool isum of the individual model fit deviances from the same fitted model. Data are pooled by combining all dose groups into one data set, keeping each group intact, including controls (i.e., not by summing animals and responders into single ...