Bayesian Divide-and-Conquer Propensity Score–Based Approaches for Leveraging Real-World Data in Randomized Control Trials

• Activity Number: 374 - Bayesian Clinical Trial Designs with Heterogeneous Patient Subgroups
• Type: Topic Contributed
• Date/Time: Wednesday, August 10, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Eric Baron* and Jian Zhu and Sammi Tang and Ming-Hui Chen
• Companies: University of Connecticut and Servier Pharmaceuticals and Servier Pharmaceuticals and University of Connecticut
• Keywords: borrowing-by-parts prior; compatibility indexes; covariate balance; propensity score matching; propensity score stratification; real-world data

Abstract

There has been an increasing interest in the usage of real world data (RWD) to supplement trial data in the medical and statistical literature. Propensity score methods such as stratification have been used to balance baseline characteristics and prognostic factors between external control patients and current trial’s control patients to improve the estimation of a treatment effect size. This paper merges propensity score methodology and Bayesian inference to estimate a treatment effect size in the presence of substantial historical control data as follows: (i) match external control patients and a current trial’s control patients by strata using the percentiles of the current control and treatment patients’ propensity scores, (ii) apply a prior within each stratum to leverage RWD to then estimate the stratum-specific treatment effects, and (iii) then use a weighted average to combine the stratum-specific treatment effect estimates to estimate the overall trial’s treatment effect size. In stage (ii), an extension of the borrowing-by-parts prior is used. A simulation study is carried out to evaluate the performance of the proposed method.