A surrogate endpoint is a laboratory measurement or physical sign that a trial tracks in place of an outcome patients actually experience — LDL cholesterol instead of heart attacks, tumor shrinkage on a scan instead of survival, HbA1c instead of diabetic complications. Surrogates exist because the real outcomes often take years or decades, making definitive trials prohibitively long and expensive. The arrangement mostly works: when a surrogate is validated — meaning interventions that improve it have reliably improved the clinical outcome in rigorous trials — it lets medicine move faster at acceptable risk. The catch is in the word validated. History contains famous stand-ins that failed exactly when it mattered, and those failures define the questions worth asking whenever a headline celebrates an endpoint you cannot feel.
This article is information, not medical advice. Endpoint choice is a property of studies, not a property of your treatment plan; discuss any therapy decision with a clinician who can weigh the full evidence, including how directly it measured what matters.
The canonical failures
Three episodes anchor every serious discussion of the topic. Antiarrhythmic drugs in the 1980s: after a heart attack, certain abnormal beats predicted sudden death, and drugs that suppressed those beats were widely used — until the CAST trial showed the medications were increasing death itself, and the approach was abandoned, with some estimates attributing tens of thousands of premature deaths in the United States to the intervening years of surrogate-based practice. Bone mineral density: osteoporosis drugs were long assessed by scans, yet improving density does not reliably predict fracture reduction across all drug classes, and regulatory thinking has shifted toward fractures as the endpoint that counts. And in cardiology, the torpedo of torcetrapib — a drug that spectacularly raised HDL cholesterol, the surrogate it was designed to improve — increased mortality in trials and was abandoned in 2006, demonstrating that even the surrogate everyone knew could betray the outcome it stood for.
What makes a surrogate valid
Validation is a claim about a chain of evidence: the surrogate must correlate with the clinical outcome, and — the stronger, decisive requirement — interventions that improve the surrogate must have been shown to improve the outcome. The first condition is common; the second is what repeatedly fails, because a disease is not a single number and drugs can move the number without touching the disease. Regulatory science formalized this: the FDA's accelerated approval pathway explicitly permits surrogate endpoints “reasonably likely” to predict clinical benefit, but attaches a condition — confirmatory trials must verify the benefit afterward. The pathway has brought effective drugs to patients years early, and its critics, including researchers who have tracked confirmations, note that some confirmatory trials have been delayed, amended, or never completed, keeping therapies on the market on surrogate credit.
- Correlation is not validation: a biomarker can track outcome across a population while responding to drugs in ways that diverge from the disease process.
- Surrogates are disease-stage specific: HbA1c control meaningfully reduces microvascular complications in diabetes; the same number has a much weaker claim on macrovascular outcomes like heart attacks.
- Composite endpoints blend surrogates: when trials count “death, MI or rehospitalization,” the softer components often drive significance.
Related stories: What a p-Value Can and Cannot Tell You About a Health Headline · Cohort Study or Randomized Trial? Why the Design Behind a Finding Changes the Claim.
How to spot a surrogate in the wild
The press release usually buries the giveaway. Phrases to look for: “improved biomarker levels,” “slowed disease progression by imaging,” “reduced surrogate markers,” or a percentage improvement in a lab value rather than in a condition. Then ask the two-question test. First, what would the patient notice — death, fracture, dialysis, stroke, or only a number on a chart? Second, is there any direct clinical-outcome evidence for this intervention in this population, or does everything rest on the stand-in? For cholesterol drugs, the outcome evidence exists — large randomized trials showed statins reduce heart attacks and mortality, which is why the LDL surrogate is now among the most validated in medicine. For a new supplement claiming to “improve inflammatory markers,” the marker may be the only thing ever measured.
Why the system still chooses surrogates
The trade is not corruption but arithmetic. A survival endpoint in early cancer may require a decade and thousands of patients; a progression-free-survival endpoint needs months and hundreds. Patients with serious diseases rationally prefer earlier access on reasonable surrogates to waiting for certainty that may outlast them. The design question is therefore not whether to use surrogates but which ones, with what confirmatory obligations, and how honestly the certainty is communicated. Oncology's own experience argues for vigilance: analyses of progression-free survival as a surrogate for overall survival have found the link holds for some drugs and tumor types and not others — endpoint validity is a property of a specific drug-disease pair, not of the biomarker in general.
Cancer medicine offers the largest current stage for this debate. Progression-free survival — the time a tumor takes to grow on a scan — is the endpoint behind most oncology approvals of the past two decades, and researchers have repeatedly compared it against overall survival across drug classes, finding the surrogate holds well in some settings and weakly in others. Registries of confirmatory-trial status for accelerated approvals, maintained by watchdog researchers and cited in congressional reviews, give the public a way to check whether a drug's surrogate-based approval was ever verified. That transparency is the direction of travel: endpoints are becoming a public conversation rather than a footnote in a protocol.
What to do with a surrogate-based claim
When a headline, advertisement or even a treatment recommendation rests on a biomarker, ask whether clinical-outcome trials exist, and whether any accelerated approval's confirmatory evidence has arrived. For patients, the practical step is a direct question in the appointment: what outcome was this actually shown to improve? The answer — death and fractures versus a scan and a lab value — should be part of the decision, stated plainly rather than discovered later.
For more context, read FDA Approval vs 510(k) Clearance: Two Words That Sound Alike and Mean Different Things.
For more context, read relative vs absolute risk.
For more context, read cohort study vs randomized trial.