| Concern over the occurrence of birth defects and developmental abnormalities that may occur as a result of prenatal exposure to drugs, chemicals, and other environmental factors has led to the development of federal testing programs which focus on making risk assessment statements about exposure and potential detrimental effects to the unborn fetus. Of interest is to identify an acceptably 'safe' dose, often referred to as a reference dose (RfD). The RfD is determined by making a low dose extrapolation from an effective dose derived from experimental data involving animals exposed to high doses. Developmental toxicity studies play an important regulatory role in establishing a RfD for substances that are potentially damaging to the developing fetus. In efforts to conduct risk assessment, the benchmark dose (BMD), the effective dose corresponding to an acceptably small increase in risk, has gained attention as the basis for the RfD. Of interest is to identify the BMD when risk is based on based on multiple outcomes that are observed in a single fetus. To estimate the BMD, joint statistical models have been developed. Due to a shared maternal environment, fetal responses are clustered by litter, and statistical approaches must account for overdispersion that results from clustering.; This thesis is organized in two parts. In Part 1, Chapters 1 and 2 describe two likelihood approaches for the analysis of developmental toxicological data. In Chapter 1, a likelihood approach is developed to analyze a mix of binary and continuous outcomes, in which the binary outcome is viewed as a less precise counterpart to the continuous outcome. The model is discussed in the context of abnormality of fetal eye development and fetal eye size. In Chapter 2, a random effects approach is developed for the analysis of malformation and fetal weight. A partial likelihood approach is presented for analyzing data in which malformation status is known but fetal weight is observed in a non-representative subsample. Valid estimates are obtained via inverse sampling weights to correct for bias when sampling is nonrandom. Results from simulation and data analyses are presented. In Part 2, Chapter 3 addresses design considerations in developmental toxicology when the outcome of interest is fetal death and the BMD is the parameter of interest. A simulation study is conducted to identify optimal allocation of animals to dose groups when risk of fetal death is related to the number of implantations. Simulation results indicate that improvement over an equal allocation design is possible. Designs are found to be sensitive to implant size via its effect on the underlying dose response curve. |
