- Main
Designing Efficient Toxicology Studies with Metaheuristics
- Gertsch, William Walter
- Advisor(s): Wong, Weng Kee
Abstract
Dose-response toxicology studies provide valuable information about the potential risks of harmful substances. These experiments inform public health policy and help identify safe doses for patients in clinical trials. However, these studies can often be poorly designed, leading to increased uncertainty around safe thresholds, wasting of resources, and unacceptable harm to subjects. The primary contribution of this dissertation is to optimize two types of experiment. The first type of study is for estimating the benchmark dose (BMD), a widely used measure for determining safety limits. Single and two-stage adaptive designs for BMD analysis are considered. To make these designs more accessible to toxicologists, a web-based tool for finding designs is introduced. The second type of experiment is optimized assuming a J-shaped non-monotonic dose-response, also known as hormesis. Because sample sizes are typically small in toxicology, exact optimal designs are found for hormesis. Optimizing these experiments requires advanced optimization approaches including mixed-integer programming and swarm intelligence methods (metaheuristics) such as differential evolution (DE) and particle swarm optimization (PSO). These swarm-based optimizers are very flexible and can be applied to many problems in statistics. For example, an additional application to subgroup analysis for an HIV prevention study is presented at the end of the work. This research paves the way for more effective and cost-efficient experiments that minimize harm.