Estimands for time to event endpoints


The current discussion on estimands primarily grew out of the wish to get more clarity on assumptions made when analyzing longitudinal or clustered data with missing values. This will likely also lead to a repositioning of the role of sensitivity analyses. Using case examples from recent Phase 3 clinical trials, we will illustrate that for time-to-event endpoints such as progression-free survival, often the estimand is not clearly specified either and thus results may be subject to ambiguity and not answer the question that scientist were asking. In one of these case studies, we discuss that Health Authority requests for additional complicated sensitivity analyses during a filing, with potential to delay approval, are often not clearly focused on getting more information on a clearly defined quantity but rather simply vary some assumptions to pressure-test robustness of results. We advocate embedding such questions in the estimand framework in order to get a clear understanding of the value of such additional analyses and why they are needed.

Berlin, Germany