バランスウェイトを用いた外部対照データによる試験の増強：推定対象と推定量の比較

As the availability of real-world data has expanded in recent years, studies leveraging external controls (ECs) have emerged as potential alternatives to conventional randomized controlled trials (RCTs). While RCTs remain the gold standard for estimating treatment effects, in some contexts, such as rare diseases, it is infeasible or unethical to enroll and randomize a sufficient number of patients. In such cases, it may be justified to conduct a hybrid or single-arm trial that incorporates EC data. This integration first depends on the harmonization of inclusion criteria, covariates, and outcomes, and then requires appropriate statistical methods that account for differences between the trial and EC patients. The propensity score (PS) has emerged as a valuable tool to summarize such differences in patient characteristics. The PS may contribute to a range of estimation methods, including balancing weights, augmented estimators, and Bayesian methods for dynamic borrowing. However, in the context of trials with ECs, the interpretation of the PS and associated causal estimands is often unclear. Motivated by the LIMIT-JIA trial of juvenile idiopathic arthritis, we elucidate potential estimands and evaluate the performance of PS-weighted estimators. We show that some estimands are easier to estimate than others, with potentially greater feasibility for small studies like LIMIT-JIA.