Innovative Bayesian Designs for Clinical Trials

Abstract

Traditional clinical trial designs are generally based on the doctrine of studying one drug for one disease at a time, which may be slow and inefficient. With a high failure rate in drug development, there is a great need to speed up the process of drug development and minimize the cost. Novel trial designs have been proposed, such as the master protocol approach, which has expanded the trial design horizon to umbrella, basket, and platform trials. Compared to traditional clinical protocols, the master protocol enables investigators to evaluate multiple drugs and diverse disease populations simultaneously in a single protocol with the capacity to modify the protocol based on the observed trial data and new drugs. While many statistical methods for trial designs have been proposed for umbrella, basket, and platform trials in the literature, most of the designs are based on a binary or continuous endpoint. However, in the context of oncology trials, there is a great need to develop novel methods for survival endpoints. In this dissertation, we propose three novel Bayesian statistical methods for three distinctive trial design problems, respectively: 1) an optimal Bayesian design for platform trials with multiple endpoints; 2) a novel Bayesian design for basket trials with survival outcomes; 3) an adaptive Bayesian design for seamless phase II/III platform trials with survival endpoints. Extensive simulation studies are performed to evaluate the operating characteristics of the proposed designs under various scenarios.