Incorporating propensity scores for evidence synthesis under bayesian framework: review and recommendations for clinical studies.

Abstract

The amount of real-world data (RWD) available from sources other than randomized-controlled trials (RCTs) has grown ultra-rapidly in recent years. It provides the impetus for generating substantial evidence of effectiveness and safety from both RCTs and RWD to accelerate medical product development. Especially in the areas of unmet needs, the conduct of fully powered RCTs is generally infeasible because of their sizes, duration, cost, or ethical constraints. The unique challenges in such areas include a small patient population, heterogeneity in disease presentation, and a lack of established endpoints. However, merging information from disparate sources is an intricate task. The value of the Bayesian framework has gained more recognition due to its flexibility in calibrating uncertainty and handling data heterogeneity, and its inherent updating process ideal for synthesizing information. Meanwhile, propensity score, as a powerful tool in causal inference, can be used in various ways to adjust for confounders. As a newly emerging data borrowing strategy in a regulatory setting, integrating propensity scores in a Bayesian setting not only utilizes the strengths from Bayesian models but also minimizes bias from external data borrowing. These methods potentially allow information sharing among data sources, provide more reliable estimates when the sample size is small, and improve the efficiency of treatment effect estimation. In this paper, we will review the recent development of methods incorporating propensity score for evidence synthesis under the Bayesian framework, and discuss different examples of incorporating external data with or without RCTs, as well as the recommendations for reporting in clinical studies.