Surrogate endpoint validation: which stand-ins have earned trust

Validating a surrogate means more than showing it predicts prognosis. A patient whose tumour responds usually lives longer than one whose tumour does not, but that is a patient-level correlation and it holds for almost any marker of a less aggressive cancer. What matters for using a surrogate in place of survival is trial-level surrogacy: across many randomised trials, does the size of the treatment effect on the surrogate predict the size of the treatment effect on survival or on how patients feel and function? That requires a meta-analysis of completed trials in a given disease and drug class, and the answer can be yes in one setting and no in the next. Regulators keep a public table of surrogates that have supported approvals and distinguish validated surrogates, which can support full approval, from those reasonably likely to predict benefit, which support accelerated approval with a confirmatory trial owed.

The corpus shows the range. Pathological complete response after neoadjuvant therapy is strongly prognostic for the individual, and KEYNOTE-522 converted a higher rate into a later event-free and overall survival benefit, but pooled analyses across breast cancer trials have found only a weak trial-level association, so pCR supports accelerated approval and not more. Progression-free survival is accepted in some settings and has repeatedly failed to predict survival in others: TROPION-Breast01 met its progression endpoint with a hazard ratio of 0.63 while overall survival was 1.01, CodeBreaK 200 delayed progression without lengthening life, and DUO-O and BEAT-meso are similar stories. Disease-free and event-free survival in the adjuvant setting have a stronger record, and monarchE, NATALEE and ADAURA rest on them, with ADAURA's survival benefit arriving later. Measurable residual disease in myeloma was accepted by an FDA advisory committee in 2024 as an endpoint for accelerated approval after a meta-analysis of trial-level correlation, and PERSEUS and CAPTIVATE show how it now drives treatment decisions inside trials. Objective response rate underpinned the accelerated approvals of olaratumab and lurbinectedin, and the confirmatory trials ANNOUNCE and ATLANTIS then failed to show a survival benefit. In screening, UKCTOCS found cancers earlier without reducing deaths, the clearest warning that a stage shift is not a life saved.

The practical reading rule is to ask three questions of any trial that wins on a surrogate: has this surrogate been validated at trial level for this disease and this class of drug, was survival at least not harmed, and is a confirmatory trial under way? The bottleneck page on trial design and the idea of an independent programme to validate surrogate endpoints take the question further.