Ask Mallard

Surrogate outcomes

A surrogate outcome is something measured in place of the clinical outcome of interest, on the assumption that changing it implies changing the outcome.

Why they are used

The real outcome may take a decade to accrue, may be rare enough to demand tens of thousands of participants, or may be unethical to wait for. A surrogate can make an otherwise impossible trial possible, and in early-phase work it is often the only practical way to decide whether to proceed.

Why they fail

A treatment can affect the surrogate through one pathway and the outcome through several, some of them harmful. Suppressing ventricular ectopy, lowering blood glucose by particular mechanisms, raising HDL cholesterol, and increasing bone density have all been shown to move the marker in the right direction while doing nothing good or actual harm to patients. The marker being a good PREDICTOR of the outcome is entirely compatible with the treatment effect not transferring.

Why a marker can mislead, as a diagram

A causal diagram in which treatment changes a surrogate marker that predicts the clinical outcome, and treatment also reaches the clinical outcome through other pathways that bypass the marker, some of them harmful.Why a surrogate can mislead: the treatment has more than one pathwayMoves the markerMarker predicts outcomeOther pathways, someharmfulTreatmentSurrogate markerClinical outcomeA marker that predicts the outcome well is compatible with a treatment effect on the marker that never reachesthe outcome.
Conceptual diagram. It shows the structure of the idea, not data from any study.

What validation actually requires

  • The surrogate predicts the outcome, which is necessary and far from sufficient.
  • The treatment effect on the outcome is fully mediated by the effect on the surrogate, for THIS class of treatment.
  • Evidence from multiple trials that the treatment effect on the surrogate predicts the treatment effect on the outcome, which is the demanding trial-level criterion.
  • Validation is specific to the treatment mechanism: a surrogate validated for one drug class says nothing about another.
  • Where these have not been shown, the honest description is unvalidated, and the trial's conclusion should be stated in terms of the surrogate.

Reporting them honestly

State plainly that the primary outcome is a surrogate, say what is known about its validation for this treatment class, and do not translate the result into a claim about the clinical outcome in the abstract. Where regulatory approval rests on a surrogate, the confirmatory outcome trial is part of the evidence, not an optional extra.

What to watch out for

That a surrogate correlates with the outcome, even strongly, is not evidence that a treatment effect on the surrogate implies an effect on the outcome. This has been demonstrated repeatedly and expensively: antiarrhythmic drugs suppressed the arrhythmias that predict sudden death and INCREASED mortality. The requirement is that the treatment effect on the outcome is fully captured by its effect on the surrogate, which is a much stronger and rarely tested condition than correlation.

Clinical example

A trial reports improved progression-free survival. Whether that implies longer overall survival depends on trial-level evidence for this tumour type and this drug class, and in several settings the correlation between the two treatment effects has been found to be weak.

Recent studies that used this
Searching PubMed…
Related Learn pages
Sources & further reading
  1. Surrogate end points in clinical trials: are we being misled? (opens in a new tab) · Ann Intern Med · doi 10.7326/0003-4819-125-7-199610010-00011
  2. Surrogate End Points and Their Validation in Oncology Clinical Trials (opens in a new tab) · J Clin Oncol · doi 10.1200/JCO.2016.66.4581
Last source review: 2026-09-06
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.