Bayesian methodological extensions for comparative effectiveness, dose-response, and cluster randomized trials

Abstract

In this dissertation, we explored three Bayesian methodological extensions, including an adaptive Bayesian design featuring participant reuse for comparative effectiveness clinical trials, an innovative Bayesian dose-response EMAX model for a mixture of normal distributions, and a Bayesian analysis of weight loss for a cluster randomized clinical trial. We first developed an adaptive Bayesian clinical trial design in the setting of comparative effectiveness clinical research where multiple treatments are of interest and the accrual rate is slow. Our proposed design mimics the real-world clinical practice that allows patients to switch treatments when the desired outcome is not achieved. As a result, each participant can have more than one observation, and hence it is possible to control for participant-specific variability which in turn results in a reduced number of participants needed. Additionally, response adaptive randomization is employed to improve trial efficiency by allocating more participants to the promising arms. We also developed an innovative Bayesian dose-response EMAX mixture model incorporating finite mixture distributions into the EMAX framework. It is the first time that an EMAX model being extended to a finite mixture distribution. The model was motivated by a proposal investigating the dose effect of DHA supplementation on preterm birth rate (< 37 weeks of gestation), where gestational age was analyzed as continuous with a normal mixture distribution. We compared our proposed EMAX mixture model with an EMAX logistic model and an independent doses logistic model for a dichotomized endpoint using extensive simulations. Across the scenarios under consideration, the EMAX mixture model achieved higher power in detecting the effect of DHA supplementation on the PTB rate. It also resulted in smaller mean squared errors (MSE) in PTB rate estimates. Lastly, we reanalyzed the percent weight loss data from Rural Engagement in Primary Care for Optimizing Weight Reduction (REPOWER), a cluster randomized clinical trial, using a Bayesian hierarchical model. We showed that the Bayesian approach can derive probability estimates of direct clinical interest and can provide additional insights into data interpretation by utilizing posterior distributions for parameters of interest. We also demonstrated that the Bayesian approach can easily handle complex problems using the same statistical framework.