A surrogate endpoint is a measurement used in place of the outcome that actually matters, usually because the real outcome takes years to appear.
A surrogate endpoint is a measurement used in place of the outcome that actually matters, usually because the real outcome takes years to appear. A marker that moves in weeks is cheaper and faster to study than an event that occurs in a decade, and that is the entire reason surrogates exist.
Using a surrogate assumes that changing the marker changes the outcome. That assumption is sometimes correct and sometimes not, and it is an empirical question rather than a definitional one. A treatment can move a marker in the right direction while leaving the outcome unchanged or worse.
A marker can be a consequence of the disease process rather than a step in it. Acting on the marker then changes a symptom of the mechanism without touching the mechanism, which looks like progress on every measurement the trial took.
Several well-known treatments improved their surrogate markers and were later found to worsen the outcomes those markers were standing in for. These cases are the reason regulators treat surrogate-based approvals more cautiously than they once did.
Ask what was actually counted. If the reported result is a change in a measured value rather than a count of events that happened to people, the endpoint is a surrogate, whatever the summary calls it.
Some markers have been shown across many trials to track outcomes reliably, and using them is reasonable. The validation is itself a body of evidence, and a paper relying on a surrogate should be able to point at it rather than assert the link.
Combining several outcomes into one count can let a common, minor component drive a result attributed to the whole. A composite dominated by its least serious element is reporting something different from what its name suggests.
A result on a surrogate is a reason to run a longer trial, not a conclusion about the outcome. Treating the two as equivalent is the most common way a technically accurate finding becomes a misleading claim.
The claims on this page about measuring a marker instead of an outcome, and when that substitution fails can be checked against sources outside this site: