The purpose of this proposal is to address statistical issues related to the design, analysis, and interpretation of animal bioassays for both cancer and developmental toxicity. These types of animal studies play an essential role in the risk assessment process for evaluating poten-tially hazardous chemical compounds and other environmental agent. Developmental toxicity studies evaluate adverse effects of exposure on the developing fetus. Increased sensitivity to exposure during certain periods of the gestational cycle makes the timing and the duration of exposures particularly important. A major component of the proposed research involves developing statistical methods for incorporating duration and timing of exposure into the risk assessment process for developmental toxicity. In particular the concept of a Bench-mark Dose will be extended to account for exposure duration. Improved methods for study design of developmental toxicity experiments which address exposure level, duration, and timing will also be developed. The assessment of developmental effects often relies on multiple endpoints, such as prenatal death or viability, malformations of various types and low birth weight. The statistical analysis of such data must therefore address both correlations among offspring for the same end-point (i.e., the litter effect) and correlations among the multiple endpoints. The proposed research addresses several issues related to the assessment of multiple outcomes in developmental toxicity studies. First, the issue of testing for exposure effects on multiple outcomes will be addressed when not all outcomes can be observed on each offspring, either due to logistical or economic constraints. Secondly, statistical methods for assessing the effect of exposure on multiple ordinal outcomes will be considered. For both cancer and developmental toxicology, much recent interest has focused on incorporating additional biological information into dose-response models for risk assessment. One component of the proposed research concerns the development of statistical methods for assessing non-linearities in dose-response and their relationship with biological and chemical characteristics, such as mutagenicity, molecular structure, and chemical activity. Flexible dose response models will first be fit to a large existing body of rodent bioassay data. Meta-analysis techniques will be developed to account for the variability in estimated dose-response patterns for a given chemical across differing sex/species combinations and across multiple endpoint (tumor sites or developmental outcomes).
