Integrated Propensity Score-Power Prior Approach for Augmenting the Control Arm of a Randomized Controlled Trial

• Activity Number: 385 - Biomarkers, Endpoint Validation and Other Topics
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yeonil Kim* and Erina Paul and Santosh Sutradhar
• Companies: Merck & Co., Inc. and Merck & Co., Inc. and Merck & Co., Inc.
• Keywords: Historical control; Power prior; Bayesian; Propensity score; Random forest; Gradient boosting

Abstract

In recent years, the real-world data are accessible due to high-volume research in biomedical sciences. This type of historical data can be incorporated for augmenting the control arm in a randomized control trial to reduce sample size, duration, and cost of the trial. We present how Bayesian method based on integrated propensity score power prior is helpful to use historical data effectively for greater benefit in study design by augmenting the control arm with matched observations from existing historical data. To this end, the integrated propensity scores (PS) are estimated by averages of PS from random forest and gradient boosting methods for minimizing selection bias of historical controls. We consider Bayesian method since it captures both heterogeneity and modeling uncertainty, and propose to use power prior approach that weighs down the likelihood of the historical data. A simulation study is conducted for binary outcomes to evaluate the performance of the proposed method by comparing with other traditional methods and the corresponding results are presented.