The goal of this project is to develop improved statistical methods for toxicology studies. Work has proceeded in two areas: (1) the analysis of data from skin painting studies, with emphasis on transgenic mouse studies, and (2) risk assessment and testing in reproductive toxicity studies. Skin painting studies on transgenic mice are used to rapidly identify carcinogens and to explore cancer mechanisms. Analysis is complicated by within-animal and serial correlations in the tumor counts, non-linear trends, tumor regression, and survival differences between animals. We have developed three new models for analyzing skin tumor data: (1) one that assumes increasing counts, (2) one that allows global and separate testing of effects on tumor onset, multiplicity, and regression, and (3) one that enables testing for global effects in sparse data sets. In reproductive toxicology studies multiple endpoints are measured on each of multiple subunits within each study subject. I have developed a method for quantitative risk assessment and testing when endpoints include both the number of subunits per subject (litter size, number of implants) and multiple binary outcomes on each subunit (implant resorbed, fetus dies, fetus malformed). I have also developed a general framework for modeling multivariate data from reproductive toxicology studies. This enables one to jointly estimate 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.
Showing the most recent 10 out of 18 publications
