Semi-Supervised Mixture Multi-Source Exchangeability Model Approach for Incorporating Real World Data into Randomized Controlled Trial Analyses

• Activity Number: 331 - ASA Biopharmaceutical Section Student Paper Award Competition
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Lillian Moran Fitzmorris Haine* and Thomas Murray and Raquel Nahra and Giota Touloumi and Eduardo Fernández-Cruz and Kathy Petoumenos and Joseph Koopmeiners
• Companies: University of Minnesota and University of Minnesota and Cooper Medical School of Rowan University and Cooper University Health Care and Department of Hygiene, Epidemiology Medical Statistics, Medical School, National and Hospital General Universitario Gregorio Marañón, Servicio de Immunología Clínica, Madrid and The Kirby Institute, University of New South Wales, Sydney, Australia and University of Minnesota - Minneapolis, MN
• Keywords: Bayesian Model Averaging; causal inference; influenza; propensity scores; real world data

Abstract

The traditional trial paradigm is criticized as being inefficient and costly. Statistical approaches that leverage external trial data have emerged to make trials more efficient by augmenting the sample size. However, they assume that external data are from previous trials, leaving a rich source of untapped real world data that cannot be effectively leveraged. We propose a semi-supervised mixture multisource exchangeability model; a flexible, two-step Bayesian approach for incorporating real world data into randomized controlled trial analyses. The first step is a semi-supervised mixture model on a modified propensity score and the second step is a multisource exchangeability model. The first step targets a representative subgroup of individuals from the trial population and the second step avoids borrowing when there are differences in outcomes among the trial and the representative subgroup. When comparing the proposed approach to competing borrowing approaches, we find that our approach borrows efficiently when the trial and real world data are consistent, while mitigating bias when the trial and external data differ on either measured or unmeasured covariates.