There is a saying among statisticians that behind every analysis there is an estimand, whether explicit or not. The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) E9 addendum formalised this intuition in 2019, calling for prospective definition of the treatment effect of interest before any analytical decisions are made. An ISPOR 2026 issue panel made it clear that while regulatory adoption has been swift, the practical discipline the framework demands remains underutilised, especially for patient-reported outcomes and tolerability endpoints.1 This matters because these endpoints may determine how regulators and health technology assessment (HTA) bodies differentiate treatments of similar clinical efficacy.
Regulatory and HTA bodies ask fundamentally different questions of the same data. A regulator asks whether treatment is safe and effective; HTA asks whether it represents sufficient value to reimburse. These questions may require genuinely different estimands; for example, different intercurrent event strategies or time horizons. While trials may not be designed with both of these priorities in mind, the estimand framework, at a minimum, gives us a system to meet these goals. As one panellist noted, a single analysis cannot answer every question, rather, the discipline lies in knowing which question you are answering and for whom.1
From a post-hoc analysis perspective, the consequences of poorly defined or misaligned estimands affect cross-trial comparability and interpretability. When data are inherited with estimands that were defined for regulatory purposes alone, analysts have little choice in whether the original estimand is fit for the question now being asked. Patient-reported outcomes are particularly vulnerable to this. Missing data strategies, assessment schedules and assumptions about what happens after treatment discontinuation vary widely across trials and without a prospective estimand anchored to a specific analytical purpose, conclusions may become uncertain.
This ambiguity compounds when evidence is synthesised across trials. A concurrent workshop on multivariate methods demonstrated the analytical ambition now possible, such as composite treatment-ranking frameworks weighting outcomes by clinical importance.2 Extending these methods to incorporate patient-reported outcomes and tolerability alongside traditional efficacy endpoints is conceptually achievable. But it rests on a precondition that does not exist in practice: estimand alignment across the synthesised trials. Heterogeneous estimands manifest as hidden between-trial heterogeneity that these methods are not designed to account for.
The estimand framework and multivariate synthesis methods have progressed and their intersection is important for regulatory submission and value recognition.1, 2 Prospective estimand discipline applied to patient-reported outcomes and tolerability supports the use of methods to synthesise patient-relevant outcomes reliably alongside efficacy endpoints, with the goal of ultimately supporting a more complete, patient-centered basis for treatment comparisons. Whether those conditions can be created upstream, in trial design and in the shared expectations of sponsors, regulators, and HTA bodies, remains an open question. However, the sessions at ISPOR this year suggest the field at least has the language to start answering it.
References
If you would like any further information on the summary presented above, please get in touch, or visit our Statistics page to learn how our expertise can benefit you. Aidan Franklin (Senior Statistician) created this article on behalf of Costello Medical. The views/opinions expressed are their own and do not necessarily reflect those of Costello Medical’s clients or affiliated partners.
