Ph.D. Candidate in Biostatistics, Lillian Haine, will present:

“Bayesian Methods for the Incorporation of Real World Data Into the Design and Analysis of Randomized Controlled Trials”

Ph.D. Advisers: Dr. Thomas Murray and Dr. Joseph Koopmeiners

Abstract: Randomized controlled trials (RCTs) are the most rigorous form of clinical evidence for regulatory decision making and are the gold-standard for studying causal relationships. Though RCTs have major benefits, they also have challenges. One major challenge is that they are expensive, requiring substantial amounts of time, money, and participants, and often struggle to enroll an adequate number of patients. One potential way to increase efficiency of RCTs would be to leverage external data on one or both trial arms in the design or the analysis of the RCT. However, incorporating external data must be done carefully to avoid undermining the internal validity of the RCT. Here, we investigate Bayesian approaches to incorporating external real world data into the design and analysis of randomized controlled trials. We aim to improve trial efficiency by decreasing necessary trial sample size, improving trial power, alongside improving statistical precision of secondary trial analyses that often are underpowered. We first introduce a Bayesian approach for incorporating observational data into the analysis of a clinical trial through a Semi-Supervised Mixture distribution and Multisource Exchangeability Model (SS-MIX-MEM). We then apply the SS-MIX-MEM to the context of binary outcomes, when many of the same baseline covariates are measured between the RCT and the observational study, and there is no missingness through an application to an influenza RCT incorporating observational data. We then introduce a trial design that facilitates borrowing during an interim futility analysis using the SS-MIX-MEM approach. We also find the optimal stopping point for this trial design, which aims to leverage external data to improve trial efficiency by stopping ineffective treatments earlier, with fewer trial participants without inflating type 1 error as we only allow early stopping and borrowing in an interim futility analysis. Finally, we apply the SS-MIX-MEM approach to borrow from electronic health records in the COVID-19 setting for a secondary analysis of a recently conducted trial to increase effective trial sample size and power.