Presentation (CTI)

OBJECTIVES: Clinical trials may report the same endpoint label yet define or measure such endpoints differently. How frequently such definitions differ, and how they differ, is not well characterized. This study examined definitions of endpoints within a defined set of clinical trials used in FDA approvals for drugs in Non-Small Cell Lung Cancer (NSCLC).
METHODS: The study protocol was prospectively registered on Open Science Framework. Data were from a previously constructed dataset of 58 FDA NSCLC approval actions from 2015 to 2024 that extracted endpoint specifications from trial registry, protocol, primary publication, and FDA review documents. Endpoints were characterized by labels, response criteria, assessor, modality, scan cadence, central review, and blinding. I evaluated definition characteristics across the corpus including overall-survival (OS) censoring rules and the handling of intercurrent events with reference to the ICH E9(R1) estimand framework.
RESULTS: Seven endpoint labels (including OS, PFS, and objective response rate, ORR) corresponded to 819 distinct specifications across 3,115 instances drawn from all documentary sources. OS censoring was expressed in 24 distinct formulations reducible to four rules. Trial scan cadence for PFS assessment ranged from 6 wks (24), 7 wks (1), 8 wks (10), to 9 wks (2). The handling of intercurrent events demonstrated divergence with 10 trials censoring events occurring while on treatment, 6 adopting a treatment-policy approach counting all events, and 5 conforming to no ICH E9(R1) strategy.
CONCLUSIONS: Endpoints sharing a common label do not necessarily share a common definition. Such variation remains undetectable when endpoints are assessed for reporting presence alone, yet may undermine indirect comparison, pooling, and other evidence synthesis activities.