Advancing outcome measurement in oncology clinical trials: Optimising validating and innovating endpoints

Abstract

Robust outcome assessment underpins therapeutic evaluation in oncology trials. Although overall survival (OS) remains the definitive endpoint, interpretation is complicated by crossover, subsequent therapies, non-proportional hazards, large sample requirements, and attrition with extended follow-up. To address this, surrogate endpoints are used to provide earlier readouts and reduce post-progression confounding. Progression-free survival (PFS) and objective response rate (ORR) are central to regulatory decisions, yet their validity as surrogates for OS remains uncertain in contemporary practice.

To define these limitations, I examined informative censoring in ovarian cancer maintenance trials. Using reconstructed patient-level data, censored patients were reclassified as progressors to model non-random censoring, showing that even modest discontinuation can inflate apparent PFS benefit. Landmark survival rates have been proposed to capture durable immunotherapy benefit. Using reconstructed lung cancer immunotherapy data, I showed that late landmark survival is highly sensitive to censoring assumptions and small numbers at risk.

To identify more reliable measures, I validated progression-free survival-2 (PFS-2), demonstrating stronger correlation with OS than PFS or ORR, with only a modest increase in follow-up or sample size. Analyses of duration and depth of response in NSCLC confirmed prognostic value but did not meet criteria for surrogacy.

Recognising the limits of conventional endpoints, I explored artificial intelligence to enhance outcome assessment and trial design.

This work traces a continuum from identifying bias in existing endpoints to validating alternatives and applying AI to adaptive, data-driven frameworks. Surrogates may improve efficiency and interpretability, but none replace OS. Progress depends on integrating statistical rigour, clinical insight, and computational intelligence to develop clinically meaningful measures of benefit.