Estimands and intercurrent events (ICH E9(R1))

Two trials can measure the same endpoint and answer different questions. If a control patient crosses over to the experimental drug at progression, does the overall survival comparison ask what happens under the policy of assigning the drug (whatever patients later receive), or what would have happened had nobody crossed over? If a patient stops the drug for toxicity, is their later progression counted against the drug, or is the question about the effect while on treatment? These things that happen after randomisation and change what is measured are called intercurrent events, and the estimand framework in the 2019 addendum ICH E9(R1) requires a trial to name, for each one, the strategy it will use: treatment policy (count everything that happens, the classic intention-to-treat approach), hypothetical (estimate what would have happened without the event, as crossover adjustment does), composite (treat the event as part of the outcome), while-on-treatment, or principal stratum (restrict to the patients in whom the event would not occur under either arm).

The framework turns familiar arguments into explicit choices. The crossover debate over VISION and PSMAfore, where most control patients received the radioligand at progression, is a disagreement about estimands: the treatment policy estimand showed little overall survival difference, and the hypothetical estimand estimated by crossover-adjustment methods showed a larger one, and each answers a legitimate but different question. Intention-to-treat and per-protocol analyses in a non-inferiority trial such as PERSEPHONE are two estimands for the same data. Trials that add a new drug on top of a control that patients may later receive anyway, as in many adjuvant designs, are asking a treatment-policy question whether they say so or not.

For readers the practical use is a checklist: what population, what variable, what happens to the number when patients switch, stop or are lost, and what summary (a hazard ratio, a difference in proportions at a landmark) is reported. A trial that states its estimand in the protocol and reports sensitivity analyses under the alternatives is harder to spin than one that picks the most flattering analysis after the fact. Regulators now expect the estimand to be specified for the primary endpoint of every confirmatory trial.