Surrogate Endpoint

A surrogate endpoint is a biomarker or intermediate outcome used in clinical trials as a substitute for a clinically meaningful final endpoint such as survival, disease progression, or functional status. The surrogate is measured earlier or more conveniently than the true endpoint and is assumed to predict eventual clinical benefit. Regulators accept surrogate endpoints in some settings: in the US, for example, a surrogate known to predict clinical benefit can support traditional approval, while accelerated approval can rest on a surrogate that is only reasonably likely to predict clinical benefit, with confirmatory trials required afterward (since the Food and Drug Omnibus Reform Act of 2022, the FDA can require that they be underway when it grants accelerated approval). However, the validity of a surrogate endpoint depends on the strength of evidence linking changes in the surrogate to changes in the clinical outcome across multiple trials and therapeutic contexts.

Clinical trials designed around final endpoints such as overall survival can require thousands of participants followed for many years, making them extremely expensive and slow. Surrogate endpoints allow researchers to evaluate treatment effects more quickly and with smaller sample sizes. This is especially important in diseases with long natural histories, such as cancer or cardiovascular disease, where waiting for final outcomes would delay access to potentially life saving therapies. However, surrogate endpoints can mislead if the assumed relationship with the clinical outcome does not hold. Several notable examples exist in which treatments improved a surrogate measure yet provided no clinical benefit or even caused harm. In the Cardiac Arrhythmia Suppression Trial (CAST), encainide and flecainide suppressed ventricular arrhythmias after myocardial infarction, yet total mortality was 7.7% (56 of 730 patients) with the drugs against 3.0% (22 of 725) with placebo. In the ILLUMINATE trial, torcetrapib raised HDL cholesterol by about 72% but increased the risk of death and cardiac events. Cases like these underscore the need for rigorous validation of any proposed surrogate.

A surrogate endpoint must satisfy specific criteria to be considered valid. At the individual level, the surrogate should correlate strongly with the clinical outcome. At the trial level, the treatment effect on the surrogate should reliably predict the treatment effect on the clinical outcome. An early, strict standard is Prentice's criterion (1989): the surrogate must capture the treatment's whole effect on the clinical outcome, so that once the surrogate is known, the treatment adds no further information about the outcome; this is rarely possible to verify in a single trial. Researchers therefore usually evaluate surrogates with meta analyses of multiple randomized trials, summarizing the individual level association between surrogate and outcome (individual level R squared) and how well the treatment effects on the surrogate predict the treatment effects on the clinical outcome across trials (trial level R squared). The proportion of the treatment effect explained by the surrogate is an older, related measure. A well validated surrogate, such as blood pressure reduction for cardiovascular events or viral load suppression for HIV clinical outcomes, can support regulatory decisions and accelerate therapeutic development. In the US, FDA publishes a Table of Surrogate Endpoints that were the basis of drug approval or licensure, which records whether each surrogate supported traditional or accelerated approval. A candidate surrogate that shows only individual level correlation without consistent trial level prediction remains unvalidated and should not be treated as a substitute for the final endpoint; where a regulator accepts such a surrogate, as under US accelerated approval, confirmatory trials of clinical benefit are still required.

Use surrogate endpoints when the final clinical outcome takes years to observe, when the disease is rare and accrual of clinical events is slow, or when ethical considerations make it inappropriate to withhold treatment long enough to observe final outcomes. Surrogate endpoints are also appropriate for dose finding studies, adaptive trial designs, and early phase efficacy assessments. They are most justified when prior meta analytic evidence has established a strong and consistent surrogate to outcome relationship across the relevant drug class and disease population.

A serious mistake is treating an unvalidated surrogate as if it were a proven predictor of clinical benefit. Researchers sometimes cite a single trial showing correlation between the surrogate and the outcome, but surrogate validation requires evidence across multiple studies. Another error is assuming that a validated surrogate in one drug class automatically applies to a different mechanism of action. The relationship between tumor shrinkage and survival, for example, varies substantially across cancer types and treatment modalities. Confusing a biomarker that predicts prognosis with one that predicts treatment effect is also problematic, because a prognostic marker may not capture the specific mechanism through which a therapy delivers benefit.

A cardiovascular outcomes trial uses LDL cholesterol reduction as a surrogate endpoint for major adverse cardiovascular events (MACE). A new lipid lowering agent reduces LDL cholesterol by 55% compared to placebo over 12 weeks (the surrogate endpoint), a result that, because LDL cholesterol is an established surrogate for lipid lowering drugs, can support regulatory approval for lowering LDL cholesterol. However, a subsequent large outcomes trial following patients for five years shows that the drug reduces MACE by only 8%, well below what would be predicted from the LDL reduction alone based on statin trial data. Investigators conclude that the surrogate relationship validated for statins does not fully transfer to this novel mechanism, reinforcing the need for confirmatory outcomes trials even after surrogate based approval.