Online Program

Incorporating Non-Randomized Data with Randomized Clinical Trials Using Commensurate Priors

*Hong Zhao, Division of Biostatistics,University of Minnesota School of Public Health 
Brian P. Hobbs, The University of Texas MD Anderson Cancer Center 
Haijun Ma, Amgen Inc. 
Qi Jiang, Amgen Inc. 
Brad Carlin, University of Minnesota 

Keywords: Bayesian analysis; commensurate priors; Markov chain Monte Carlo (MCMC); observational studies (OSs); propensity score matching; randomized clinical trials (RCTs)

Historically, randomized clinical trials (RCTs) have been recognized as the gold standard for evaluating the efficacy or safety of a therapeutic intervention. Although RCTs have reliable internal validity, they often are restricted to a specific group compared to observational studies (OSs), which may have better generalizability due to a broader subject pool. However, OSs may suffer from selection bias without proper adjustment for potential confounders. Therefore, combining RCTs and OSs is often criticized due to the limitations of OSs. Recent research suggests research synthesis of treatment effects should not be restricted to specific study types. In this work, we develop hierarchical Bayesian approaches to combine data from all sources simultaneously while explicitly acknowledging the differences in designs. Using commensurate priors, we can adaptively borrow from the propensity score--matched non-randomized data when justified by its commensurability with the RCT data. We apply the proposed framework to a recent HIV trial and investigate its properties via simulation. Our findings elucidate the extent to which OSs can complement RCTs for improved clinical decision making.