Statistical Methods In Toxicology

Environmental Health Sciences

Investigators

Linked publications & trials

Abstract

The goal of this project is to develop improved statistical methods for toxicology studies. Work has proceeded in two areas: (1) the analysis of multi-site tumor data from 2-year bioassays, and (2) risk assessment and testing in toxicology studies that measure multiple endpoints. Two-year bioassay studies are routinely conducted by the NTP to assess the tumorigenic potential of test agents. Current methods consider each tumor site separately. We have developed an approach for joint analysis of data from multiple tumor sites. This approach accounts for within-animal dependency, survival differences between groups and tumor lethality, while also allowing for the incorporation of historical control information. Multiple endpoints are often measured in reproductive and developmental toxicity studies. We have developed methods for assessing overall toxic effects in reproductive experiments when data include both the number of subunits per dam (litter size, number of implants) and multiple outcomes on each subunit (low birth weight, malformation). We have also developed a general framework for modeling of multivariate clustered data that enables joint estimation of effects on disparate outcomes such as the number of implantation sites per animal, the proportion of dead fetuses per dam, the proportion of malformed fetuses per dam, and birth weight. Another approach we have developed uses a novel dynamic latent variable structure within the framework of a Bayesian hierarchical model to characterize neurotoxic effects of chemicals.