Integrating Randomized Controlled Trial and External Control Data Using Balancing Weights: A Comparison of Estimands and Estimators (2502.13871v1)

Abstract: Randomized controlled trials (RCTs) face inherent limitations, such as ethical or resource constraints, which lead to a limited number of study participants. To address these limitations, recent research endeavors have sought to incorporate external control (EC) data, such as historical trial data or real-world data, with RCT data in treatment effect evaluation. This integration introduces unique questions regarding target population specification, causal estimand, and optimality of pooled estimators. Balancing weights have emerged as valuable tools to ensure comparability in patient characteristics, but there remains a gap in implementing them with ECs. In this study, we elucidate potential estimands of interest and propose corresponding balancing-weight-based estimators. We provide statistical and clinical definitions and interpretations of the estimands. Our extensive simulations show that different causal estimands perform differently with respect to bias and efficiency, based on the level of similarity between RCT and EC.