Applying the estimand framework to safety: aligning estimators with treatment effects in the presence of intercurrent events

Understanding the risks associated with healthcare interventions is essential for clinical decision-making and for the regulatory evaluation of new drugs. Yet, safety analyses in randomised controlled trials are often conducted without clearly specifying (a) the estimand of interest in the presence of intercurrent events (IEs) (ie, events such as treatment discontinuation or use of rescue medication that occur after randomisation and complicate outcome interpretation) or (b) the estimand implicitly targeted by the statistical estimator used in the analysis. This lack of clarity produces ambiguous results and complicates their interpretation.

This paper provides practical guidance on applying the estimand framework from the ICH E9(R1) addendum to safety outcomes in phase III randomised trials conducted for regulatory submissions. We describe how IEs influence the interpretation of treatment effects in safety assessment and how commonly used safety estimators align with different estimands, making explicit both the clinical question and the assumptions required to estimate it.

Although the primary emphasis is on regulatory trials, where these decisions are especially consequential, the principles outlined are equally applicable to other randomised interventions and non-interventional studies. As safety results from pivotal regulatory trials are increasingly published using estimand terminology, these concepts are essential not only for researchers such as statisticians, clinicians and regulators involved in clinical development but also for clinicians who prescribe newly approved drugs and need to interpret safety results communicated using estimand terminology to inform their clinical decision-making. This paper aims to make these concepts accessible to a broader medical audience.