Bayesian modeling of combined endpoints for sequentially adaptive design and comfirmatory trial planning
Recent scientific advances in biomedical research have rapidly increased the number of promising new cancer treatments available for clinical evaluation. Yet, current drug development strategies have proved to be quite inefficient. Agents are screened one-at-a-time in a sequential process that is characterized by low cost effectiveness and low specificity, as it often fails to identify those agents for which further development should be stopped. The vast majority of failures occurs late in the drug development process, phase III trials. The high failure rate in phase III may be due to the deficiency of phase II with improper endpoints. My dissertation is to explore multi-arm clinical trial design for phase II that uses Bayesian hierarchical modeling to discern and leverage the relationships between short-term tumor response and long-term survival endpoints. We evaluate the extent to which combining endpoints may help facilitating simultaneous sequential screening of multiple competing therapies for the purpose of efficiently and effectively identifying the most promising therapies for subsequent confirmatory evaluation in phase III. We explain how to use phase II data to properly predict the probability of success in a future phase III as a function of the planned sample size, as well as how to use predictive probability of success and predicted utility to decide whether to conduct a phase III trial. Simulation studies demonstrate that the proposed design greatly improves the efficiency of the trial by reducing total sample size by 40% to 50%, and achieves significantly greater power to identify the efficacious arm(s) while maintaining comparable family wise type I error rate when compared to the conventional sequential two-armed designs. The reduction in total sample size is more profound when there is(are) efficacious treatment arm(s). Our proposed design also outperforms a comparator Bayesian multi-arm design without ordering of the long-term survival for the ordinal short-term response categories by achieving significant higher posterior probability to identify the superior experimental arm(s) and significantly higher predictive probability of success in future phase III. When using expected utility and predictive probability of success combining both endpoints we enhance/optimize the expected payoff of a future phase III for a given sample size and utility. ^
Wei, Caimiao, "Bayesian modeling of combined endpoints for sequentially adaptive design and comfirmatory trial planning" (2015). Texas Medical Center Dissertations (via ProQuest). AAI3721417.