Causal mediation framework for understanding and communicating estimands and analysis strategies in clinical trials

Abstract

Estimands, introduced in the ICH E9(R1) Addendum, are deeply rooted in the causal inference framework, its widespread implementation and the engagement of functional stakeholders in a clinical trial are in progress. Nevertheless, communicating the estimands to various stakeholders in an easy-to-understand way poses some challenges, the tools used in the causal inference such as counterfactual outcome, directed acyclic graph (DAG), and single-world intervention graph (SWIG) have started to play a valuable role in defining, identifying, and communicating estimands across functional stakeholders. However, a pragmatic question remains: what type of (treatment) effect that the estimand is endeavoring to estimate in handling an intercurrent event? The causal mediation framework sheds light on establishing the connection between estimand and type of effect, and provides an instructively streamlined decision workflow to select the proper analysis strategy for an interested ICE. We elaborate on this connection in detail and hope it can further enhance and facilitate the understanding of estimands and the communication with different stakeholders in a clinical trial.