Overall survival as a safety guardrail in cancer drug trials

Surrogate endpoints have accelerated access to oncology drugs but often leave uncertainty about net patient benefit. We propose a pragmatic framework that recentres overall survival (OS) as both the definitive patient-centred efficacy endpoint and a safety guardrail capable of detecting net harm that intermediate measures may miss. The framework comprises four pillars. First, designing for harm exclusion: protocols should prespecify a clinically meaningful OS harm margin, power event-driven follow-up to exclude or detect that margin with precision, and mandate independent data monitoring committee oversight. Analyses should match explicit estimands and pair hazard ratios with restricted mean survival time to accommodate non-proportional hazards. Second, handling crossover and post-progression therapy: when crossover is ethical or unavoidable, trials should adopt treatment-policy estimands for the primary question and prespecified causal sensitivity analyses. Protocols must map access to guideline-concordant post-progression care, record uptake and timing, and distinguish biological dilution from health-system scarcity. Third, using surrogates with discipline: when OS cannot feasibly be primary, sponsors should submit a disease-specific surrogate dossier summarising trial- and patient-level validation and quantifying expected translation to OS, while continuing to collect OS and assess it against the prespecified harm boundary. Fourth, preventing missingness-driven bias: continue outcome collection after treatment discontinuation, set triggers for asymmetric loss to follow-up, and perform structured tipping-point analyses. We recommend a regulatory traffic-light aligned to OS maturity, green (traditional approval), amber (time-limited with confirmatory obligations), red (standards unmet), and conditioning economic conclusions (QALYs, net monetary benefit) on OS harm exclusion. Operationalising OS as a safety guardrail protects patients and strengthens the credibility and value of cancer trials.