Combining the Target Trial and Estimand Frameworks to Define the Causal Estimand: An Application Using Real-World Data to Contextualize a Single-Arm Trial

Abstract

Single-arm trials (SATs) may be used to support regulatory submissions in settings where there is a high unmet medical need and highly promising early efficacy data undermine the equipoise needed for randomization. In this context, patient-level real-world data (RWD) may be used to create an external control arm (ECA) to contextualize the SAT results. However, naive comparisons of the SAT with its ECA will yield biased estimates of causal effects if groups are imbalanced with regards to (un)measured prognostic factors. Several methods are available to adjust for measured confounding, but the interpretation of such analyses is challenging unless the causal question of interest is clearly defined, and the estimator is aligned with the estimand. Additional complications arise when patients in the ECA meet the inclusion/exclusion criteria for the SAT at multiple timepoints. In this article, we use a case-study of a pivotal SAT to illustrate how a combination of the target trial and the ICH E9(R1) estimand frameworks can be used to define the target estimand and avoid common methodological pitfalls. We also propose an approach to address the challenge of how to define an appropriate time zero for external controls who meet the SAT inclusion/exclusion criteria at several timepoints.