Surrogate endpoints: when the stand-in and the real outcome diverge

A surrogate endpoint is a stand-in. Instead of measuring the outcome a patient actually cares about, dying, having a heart attack, losing function, a trial measures something easier or faster to track and assumes it moves in step: a lab value, a scan result, a symptom score. The hope is that if the surrogate improves, the real outcome will too. Often it does. Sometimes it does not, and the gap between the two is where some of medicine's worst mistakes have happened.

Regulators formalize this with a specific definition: a surrogate endpoint is used in a trial as a substitute for a direct measure of how a patient feels, functions, or survives, and it is only expected, not proven, to predict the real benefit. Some surrogates are well validated by decades of evidence. Others are candidates, chosen because they are convenient or cheap to measure or let a trial finish in months instead of years, long before anyone has confirmed that moving the surrogate actually helps patients.

Two real trials show what happens when that gap opens up. In the Cardiac Arrhythmia Suppression Trial, the surrogate was an abnormal heartbeat pattern after a heart attack, and drugs that suppressed it were assumed to prevent sudden death. Instead, encainide and flecainide caused 63 deaths among 755 treated patients, compared with 26 deaths among 743 on placebo, and the trial was stopped early. Cholesterol offers a second lesson: in a 15,067-patient trial, the drug torcetrapib raised protective HDL cholesterol by 72 percent, yet patients taking it died at a higher rate than those on placebo, a hazard ratio of 1.58. The surrogate moved the right way while patients got worse.

So when a headline reports that a drug improved some biological marker, ask what that marker is standing in for, and whether any trial has shown that moving it actually changes how long or how well people live. A surrogate backed by decades of confirming trials is a reasonable basis for a claim. A new or convenient one is not, and the only way to tell the difference is to look for a trial that measured death or disability directly, not just the stand-in.