BMJ Framework Puts Surrogate Endpoint Judgments on firmer ground
Surrogate endpoints can speed drug development, but they also can mislead. A laboratory value, imaging marker, or other substitute outcome may move in the right direction while the outcome patients actually care about—how they feel, function, or survive—fails to improve. The new REASSurE framework, short for Recommendations for Evaluating a Single Surrogate Endpoint, was built to make that tradeoff explicit rather than implicit.
The project arrives at a moment when surrogate use has become central to modern trials, regulatory decisions, and health technology assessment. Faster readouts can shorten development timelines and bring treatments to market sooner, yet the cost is uncertainty. When a surrogate has not been evaluated carefully, a favorable signal can overstate benefit, obscure harm, or produce a false sense of therapeutic promise.
One sentence captures the entire clinical risk: a surrogate can look convincing long before the patient outcome does.
Why surrogate evaluation needs more than one statistical test
The framework addresses a problem that has persisted across decades of methods work: no single statistical metric has won universal acceptance. Some approaches focus on the proportion of treatment effect explained, while others examine individual-level associations, trial-level correlations, surrogate threshold effects, causal predictiveness, individual causal association, sample size multipliers, or true endpoint advantage. Each method asks a slightly different question. None alone settles whether a surrogate is robust enough to stand in for a patient relevant outcome.
REASSurE organizes evaluation around three unresolved issues. First, which non-statistical factors matter most in judging a surrogate? Second, should numerical thresholds determine validity? Third, which statistical method or methods are best suited to the task? By bringing those questions into a single structure, the framework moves surrogate assessment away from ad hoc interpretation and toward a more disciplined review.
The distinction between individual level surrogacy and trial level surrogacy sits at the center of that effort. Individual level surrogacy asks whether, within a trial, patients with better surrogate values also tend to have better clinical outcomes. Trial level surrogacy asks whether treatment effects on the surrogate track treatment effects on the patient relevant outcome across studies. Those are not interchangeable concepts. A marker can predict risk within a patient group yet still fail to predict how a treatment will perform in a randomized trial.
The framework was built through consensus, not isolated method choice
REASSurE was developed in four stages: a two-round online Delphi survey, an international online consensus meeting, piloting of the draft framework, and finalization of the tool. The process drew on five stakeholder groups: statisticians, trials methodologists, regulators, health technology assessment experts, and clinicians. Their involvement matters because surrogate evaluation is not only a statistical exercise. It also influences regulatory approval, reimbursement decisions, trial design, and patient counseling.
The item set used for the Delphi process was assembled from three established surrogate evaluation frameworks, along with material from three additional multi-stakeholder consensus studies on surrogate assessment. Existing frameworks had already contributed important pieces of the puzzle, including hierarchy-of-evidence concepts, correlation-based approaches, reliability criteria, and domain-based evaluation schemas. What they did not fully resolve was the lack of a unified structure that covers both the scientific and practical dimensions of surrogate use.
REASSurE aims to fill that gap by listing mandatory and recommended items for evaluation and by giving guidance on how each item should be assessed. The goal is not to replace scientific judgment with a scoring formula. It is to make judgment transparent, reproducible, and easier to defend when a surrogate endpoint is used to support a clinical decision.
Patient relevance, informed consent, and the limits of speed
The bedside implications are straightforward. If a surrogate endpoint is inadequately evaluated, the uncertainty does not stay inside a statistical appendix. It reaches the patient who enrolls in a trial, the clinician who prescribes the eventual therapy, and the regulator or payer who decides whether the therapy should be available. The article explicitly links surrogate assessment to informed consent: trial participants should understand the risks they accept, and those risks rise when the endpoint itself has not been scrutinized carefully enough.
Biology still matters, and so does context. A surrogate may be more or less credible depending on the disease process, the intervention, and the strength of the evidence linking the surrogate to the endpoint that matters to patients. In some settings, the pathophysiologic pathway is direct and well mapped. In others, the surrogate may sit only loosely connected to downstream clinical benefit, leaving room for discordance between marker change and real-world outcome.
This is why a purely numerical approach can fail. A correlation estimate or threshold can be informative, but it cannot substitute for a holistic review of biological plausibility, clinical relevance, epidemiologic consistency, and statistical reliability. A favorable number without the right context can create overconfidence. A weaker number in a well-understood mechanism may deserve a different interpretation than the same number in a poorly understood one.
REASSurE is designed for that broader reading of evidence. Statisticians, trial methodologists, regulators, health technology assessment agencies, payers, and clinicians can apply the framework to judge whether a candidate surrogate endpoint is suitable for the decision being made. The framework does not claim to erase uncertainty. It makes uncertainty visible, which is the more defensible position when surrogate endpoints are used to accelerate development and approval.
For now, the important limitation is not hidden in the framework’s design but in the field it addresses: surrogate validation remains heterogeneous, and consensus is still lacking on the best metric, the right threshold, and the proper balance between scientific efficiency and patient-relevant certainty. The framework provides structure, but it does not remove the need for careful interpretation in every disease area and every trial program.