Research

What a trial endpoint actually measures

An endpoint is the thing a trial measured. Whether it is the thing you care about is a separate question, and FDA keeps a public list of the substitutes it has accepted because the answer is so often no.

By Nora Castellan, Standards EditorUpdated September 2026

The endpoint is the question the trial asked

A clinical trial's endpoints are what it measured. Everything else about the study, including its size, its length and its statistics, exists to answer the question the endpoint poses.

FDA divides them into two kinds and defines both. Clinical outcomes directly measure whether people in a trial feel better, function better, or live longer. Surrogate endpoints are substitutes, used instead of a clinical outcome.

The distinction is the whole subject. A trial that measured a marker found out about the marker. Whether the marker predicts anything a person would notice is a question that has to be answered separately, with its own evidence.

Why substitutes exist, and what earns one

Surrogates are not a shortcut anyone invented to cut corners. FDA gives three reasons for using them. The clinical outcome might take a very long time to observe. The benefit of improving the marker might already be well understood. Or running a clinical endpoint study would be unethical.

The agency's worked example is blood pressure. Many trials of many different medications showed that reducing systolic blood pressure reduced the risk of stroke. Because that link was established across drugs, a reduction in blood pressure can now stand in for a reduction in strokes, and trials can be smaller and faster as a result.

That is what a validated surrogate is. FDA describes the process as requiring extensive accumulated evidence, from epidemiological studies and from clinical trials, showing that the marker can be relied on to predict or correlate with clinical benefit in a defined context.

The important word in that sentence is context. A surrogate is validated for a use, not in general. The same marker can be a good stand-in for one disease and a poor one for another.

The weaker tier, written into regulation

There is a second, explicitly lower standard, and it is in the Code of Federal Regulations rather than in anyone's marketing.

FDA may approve a product based on trials showing an effect on a surrogate endpoint that is reasonably likely to predict clinical benefit, judged on epidemiologic, therapeutic, pathophysiologic or other evidence. That is the accelerated approval pathway.

The regulation attaches a condition in the same sentence. Approval is subject to the requirement that the applicant study the drug further to verify and describe its clinical benefit, where there is uncertainty about the relation between the surrogate and the benefit.

So the law itself recognizes three positions: an outcome that was measured, a substitute that has been validated, and a substitute that is only reasonably likely to predict. A page that says a trial showed a benefit is not distinguishing among them.

FDA maintains a public table of the surrogate endpoints that have been the basis of an approval or a licensure. It exists because Congress required it, and it is the closest thing to an authoritative answer about whether a given marker has ever carried an approval.

How common substitutes are

FDA states the proportion itself. Between 2010 and 2012, it approved 45 percent of new drugs on the basis of a surrogate endpoint.

That number is worth holding onto in both directions. It means surrogate-based evidence is normal, accepted and everywhere in modern medicine, so a marker-based trial is not automatically weak.

It also means that reading the word approved tells you very little on its own about what was measured. Nearly half the time, the trial that supported an approval measured a stand-in.

A composite endpoint counts several things as one

The third common structure is a composite, where a trial counts the first occurrence of any event from a defined list.

The semaglutide labeling describes one precisely. In a cardiovascular outcomes trial, the primary endpoint was time to first occurrence of a three-part composite: cardiovascular death, non-fatal heart attack, and non-fatal stroke.

Two features of that design are worth understanding. It counts the first event only, so a patient who has a heart attack and later dies contributes one event. And it treats three events of very different severity as one countable thing.

Composites are legitimate and often necessary, because single serious events are rare enough that a trial measuring only one of them would need to be enormous. The reading rule is simply that a composite result is a result about the composite. Whether the effect sat mostly in one component is a separate question the headline number does not answer.

Primary, secondary and the difference that matters

A trial names its primary endpoint before it starts. That is the one it is designed and sized around, and it is the one the result belongs to.

Everything else is secondary or exploratory. Those measures are collected and reported, and they were not what the trial was built to answer. A secondary endpoint that moved is a reason to run another trial, not a finding on the same footing as the primary.

A trial can have more than one primary measure. The tirzepatide labeling describes a weight study whose primary efficacy parameters were the mean percent change in body weight and the proportion of patients reaching at least a five percent reduction. Two measures, both pre-specified, both primary.

The labeling is also explicit about what is not primary. For one study it notes that change from a particular time point was not a primary endpoint, which is exactly the kind of distinction that disappears when a result is summarized elsewhere.

A measured index is not the same as how someone feels

The sleep apnea studies in the tirzepatide labeling show the two kinds of measurement sitting side by side.

Eligibility and the main measure ran on the apnea-hypopnea index, which counts breathing events per hour on an overnight sleep study. The labeling also reports the Epworth Sleepiness Score, which is a questionnaire about how sleepy a person is during the day.

Both are real measurements and they answer different questions. One counts events on a machine. The other records what the person noticed.

That gap is where most disagreement about a treatment lives. A marker can move convincingly while the experience does not, and the experience can improve for reasons a marker does not capture. Knowing which one a trial measured is the difference between reading a result and inheriting someone's summary of it.

How to read an endpoint on a product page

Find the primary endpoint and read it literally. What was counted, in whom, over how long.

Ask whether it is an outcome or a substitute. Did people feel better, function better or live longer, or did a number move.

If it is a substitute, ask whether it has been validated for that use. FDA's public table of surrogate endpoints that supported approvals is the reference point, and a marker's absence from it is not proof of anything on its own.

If it is a composite, notice what is inside it and that only the first event counts.

Check whether the result being quoted was the primary endpoint at all. A secondary finding presented as the study's conclusion is the most common way a real trial gets misreported.

Then check the last thing, which is whether there is a trial. Most compounds sold in this market have no endpoint to read, because no controlled human study of the marketed use has been published.

Key takeaways

Frequently asked questions

What is the difference between a clinical outcome and a surrogate endpoint?

FDA defines clinical outcomes as measures of whether people in a trial feel better, function better or live longer. A surrogate endpoint is a substitute used in place of one, usually because the real outcome would take too long to observe or because a study measuring it directly would be unethical. A surrogate is accepted only for a defined context of use, after evidence has accumulated that it predicts or correlates with the clinical benefit it stands in for.

How often are drugs approved on a surrogate endpoint?

Frequently. FDA states that between 2010 and 2012 it approved 45 percent of new drugs on the basis of a surrogate endpoint. That cuts both ways for a reader. Marker-based evidence is normal and accepted rather than second-rate, and at the same time the word approved does not by itself tell you whether anyone measured how patients felt, functioned or fared.

What is accelerated approval based on a surrogate?

It is a lower evidentiary tier written into federal regulation. FDA may approve a product on trials showing an effect on a surrogate endpoint that is reasonably likely to predict clinical benefit, based on epidemiologic, therapeutic, pathophysiologic or other evidence. The same regulation requires the applicant to study the drug further to verify and describe the clinical benefit where the relationship is uncertain. Reasonably likely to predict is a weaker claim than validated.

What does a composite endpoint hide?

It counts several different events as one and stops at the first. The semaglutide cardiovascular trial used time to first occurrence of a three-part composite of cardiovascular death, non-fatal heart attack and non-fatal stroke. That design is often necessary, because single serious events are rare enough that measuring one alone would need an enormous trial. The reading rule is that a composite result describes the composite, and whether the effect sat mostly in one component is a separate question.

Why does it matter whether a result was the primary endpoint?

Because the primary endpoint is named before the trial starts and the study is designed and sized around it. Secondary and exploratory measures are collected as well, and a movement in one of them is a reason to run another trial rather than a finding on the same footing. Presenting a secondary result as the study's conclusion is the most common way an honest trial gets misreported downstream.

Sources

Each document below is named as it names itself, with the date printed on that document rather than the day it was read.

  1. Surrogate Endpoint Resources for Drug and Biologic Development — U.S. Food and Drug Administration, July 2018
  2. Table of Surrogate Endpoints That Were the Basis of Drug Approval or Licensure — U.S. Food and Drug Administration, April 2026
  3. 21 CFR 314.510 — Approval based on a surrogate endpoint or on an effect on a clinical endpoint other than survival or irreversible morbidity — Office of the Federal Register, Electronic Code of Federal Regulations, January 2026
  4. OZEMPIC (semaglutide) injection — FDA-approved prescribing information — National Library of Medicine, DailyMed, June 2026
  5. ZEPBOUND (tirzepatide) injection — FDA-approved prescribing information — National Library of Medicine, DailyMed, August 2026