All Times EDT
Keywords: Estimands, Causal Inference, Principal Stratification, Intercurrent Events
Estimation of treatment effect in patient subpopulations determined by intercurrent events (ICEs) that occur post-treatment initiation is often of interest in randomized clinical trials. One such example is the subpopulation (or stratum) of patients who would adhere to the treatment regimen specified in the protocol, that is, those who would not discontinue treatment or initiate a rescue medication that is not envisioned as part of the treatment regimen. Additionally, a stratum of interest could comprise patients who would not experience ICEs such as adverse events on experimental treatment regardless of their actual treatment received. The ICH E9 (R1) suggests that Principal Stratification (PS) is an important strategy to handle ICEs. However, there are still apprehensions surrounding the applications of PS. For instance, some are under the impression that PS is essentially subgroup analysis, which does not use potential outcomes or is often not a valid causal estimand. Others contest that looking at subpopulations of patients who can adhere to at least the experimental arm constitutes an invalid causal inference since the missing data imputed is not a valid potential outcome due to the non-existence of the outcome itself. This roundtable is aimed at having a discussion about PS: what constitutes a valid causal inference under this framework, potential applications of PS and past successful case studies.