A Pragmatic Implementation Framework for Phase 2 Trials with Invasive Endpoints: Estimands, Missing Outcomes, and...

Phase 2 trials that rely on invasive or high-burden follow-up outcomes (e.g., biopsy or endoscopy) can lose a substantial proportion of endpoint observations even when the underlying clinical outcome still exists. The estimand itself is predefined by the clinical question; it is not determined by the analysis set. However, restricting estimation to participants with an observed in-window invasive assessment can produce an estimator that is no longer well aligned with the target estimand and is vulnerable to post-randomization selection. Using a recent 24-week phase 2 trial with liver histology as an operational example, we distinguish intercurrent events from failures of outcome observation and show why this distinction matters for estimation. We propose an implementation framework, rather than a new statistical doctrine, that applies ICH E9(R1) and complementary trial guidance to settings where outcome ascertainment is burdensome, sparse, and potentially incomplete. The framework separates the clinical question from the observation process, maps each intercurrent event to an appropriate estimand strategy, specifies a primary missing-data approach supported by sensitivity analyses spanning plausible MAR and MNAR departures, and integrates inferential robustness with prespecified phase 2 decision criteria and ascertainment feasibility. An invasive-endpoint reporting supplement emphasizes procedure-specific denominators, reasons for missingness, technical adequacy, auxiliary non-invasive measurements, and feasibility for later-phase development.