Surrogate Endpoints in Clinical Trials: The Hidden Risk
Tumour shrinkage is not survival. Blood pressure reduction is not stroke prevention. The use of surrogate endpoints in clinical trials has accelerated drug approval - and produced some of medicine's most expensive mistakes.
In 2004, rosiglitazone was one of the best-selling diabetes drugs in the world. It lowered HbA1c reliably - a surrogate endpoint for glycaemic control that regulators and physicians had come to trust as a proxy for the outcomes that actually matter to patients: cardiovascular events, microvascular disease, death. In 2007, a meta-analysis by Steven Nissen and Kathy Wolski in the New England Journal of Medicine found that rosiglitazone was associated with a significant increase in the risk of myocardial infarction. The surrogate had been positive. The clinical outcome was the opposite. The FDA ultimately imposed severe restrictions on the drug.
What a Surrogate Endpoint Is
A surrogate endpoint is a measure used in clinical trials as a substitute for a direct measure of clinical benefit - how patients feel, function, or survive. It is used because direct clinical endpoints often require large sample sizes and long follow-up to detect differences, making trials expensive and slow. A surrogate is justified when there is strong evidence that changes in the surrogate reliably predict changes in the clinical outcome of interest.
The operative word is "reliably." Not all biological associations translate to reliable prediction. A surrogate can be correlated with a clinical outcome without being on the causal pathway that the treatment affects - and in that case, improving the surrogate says nothing about whether the clinical outcome has improved.
The FDA Accelerated Approval Pathway
The FDA's accelerated approval pathway, established in 1992 in response to the HIV epidemic, allows drugs for serious conditions to be approved based on surrogate endpoints "reasonably likely to predict" clinical benefit, with post-market confirmatory trials required. The pathway has enabled faster access to drugs for diseases with unmet need - particularly in oncology - and it has also produced approvals that were later withdrawn when confirmatory trials failed to show clinical benefit.
The FDA's accelerated approval programme has faced sustained scrutiny for the pace at which confirmatory evidence has been generated. Congressional investigations and subsequent FDA reforms in 2023 gave the agency stronger authority to withdraw approvals when confirmatory trials fail to materialise or fail to show benefit - a recognition that the original pathway had created a structural incentive to defer the hard evidence question.
Torcetrapib and the Surrogate Trap
Torcetrapib was developed by Pfizer as a cholesterol-modifying agent. It raised HDL cholesterol substantially - an effect that the lipid hypothesis of cardiovascular disease predicted should be beneficial. The drug progressed through a full clinical development programme at enormous expense, reaching a phase III outcomes trial with more than 15,000 patients. The trial was stopped early not because of futility but because of harm: patients on torcetrapib had significantly higher cardiovascular mortality than those on placebo. HDL elevation by this mechanism did not translate to clinical benefit; it caused harm.
The torcetrapib story is not a failure of the surrogate per se - HDL is genuinely associated with cardiovascular outcomes in observational data. It is a failure of assuming that raising a surrogate by any mechanism would produce the same outcome improvement as the observational association implied. This is the surrogate trap: the surrogate is valid as a marker, but that does not make it valid as a target.
The Validation Standard
Rigorous validation of a surrogate endpoint requires demonstrating that the treatment effect on the surrogate reliably captures the treatment effect on the clinical outcome - a standard known as "surrogacy" in the statistical literature. This requires meta-analytic evidence across multiple trials showing that the treatment-surrogate relationship and the surrogate-outcome relationship together account for the treatment-outcome relationship. The Prentice criteria and more recent meta-analytic approaches provide frameworks, but few surrogates in common use meet the formal validation standard.
The practical implication is that a trial showing benefit on a surrogate endpoint is evidence of biological activity, not evidence of clinical benefit - and conflating the two has real consequences for patients who receive treatments based on surrogate-only evidence.
What This Means for Evidence Synthesis
For systematic reviewers and HTA analysts, the surrogate endpoint question is a frequent challenge. Many trials in common disease areas report only surrogate endpoints, and the decision about whether to include those trials - and how to weight their evidence - requires explicit judgement about surrogate validity. Reviews that pool trials using different outcome definitions, some measuring surrogates and some measuring clinical outcomes, without clearly distinguishing between them, can produce conclusions that look more confident than the underlying evidence supports.
At NousLab, the distinction between surrogate and clinical endpoints is a standard component of the evidence characterisation we produce for systematic reviews and HTA dossiers. Knowing where a given endpoint sits on the validation spectrum is not a technical nicety - it is central to how confident any conclusion can legitimately be. See how NousLab supports HTA evidence synthesis.
