External comparators and estimands

Abstract

The estimand framework as defined by the ICH E9(R1) addendum aims to clearly define “the treatment effect reflecting the clinical question posed by the trial objective”. It intends to achieve this goal of a clear definition by specifying the 5 estimand attributes: treatment conditions, population, endpoints, handling of intercurrent events (IEs), and population-level summary. However, hybrid clinical/observational research like External Comparators (ECs) leads to new reflections on existing attributes and to considerations for additional ones. Specifically, treatment conditions and exposure may be more difficult to handle in the EC, and especially Standard of Care (SoC) treatment needs detailed attention. The external population typically cannot be based on the classical Intention-to-treat population and constitutes also an approximation only. Endpoints may not be comparable across cohorts, and IEs may be more different than in an RCT setting, such that the hypothetical treatment policy according to the ICH E9(R1) addendum may become of greater interest especially for long-term endpoints. Finally, the necessary assumptions for some population-level summaries (e.g., the proportional hazards assumption) can become more fragile when joining data from different sources due to induced heterogeneity. Finally, it is shown that the baseline definition and the marginal estimator are candidates for additional estimand attributes in case the estimand framework is revised to account for observational study needs.

Application of the target trial emulation framework to external comparator studies

“…In the EC setting, treatment strategies must be well-defined and relevant to clinical and policy decision making (Rippin, 2024). The description of a treatment strategy for the comparator should be made as unambiguous as possible to avoid violating the causal consistency assumption, which requires that all versions of a treatment strategy with respect to a specific dose, frequency, duration, and route of administration would have the same effect (Cole and Frangakis, 2009).…”

“…The PP effect in an EC study is analogous to the PP effect in an RCT and accordingly requires a detailed description of the comparator treatment strategy (see section 2.2) and the use of the appropriate methods for baseline and postbaseline confounding adjustment (Toh et al, 2010; Frontiers in Drug Safety and Regulation frontiersin.org Hernández-Díaz, 2012;Maringe et al, 2020). For more detail on estimands in EC studies, refer to Rippin (Rippin, 2024). Another consideration is that the EC study sample represents a case-mix of at least two populations with different selection mechanisms (i.e., referent trial and RW).…”