Surrogate Endpoints vs Clinical Outcomes

A trial can show that a laboratory value, imaging measurement, or physiologic sign changed without directly showing that people felt better, functioned better, or lived longer. The distinction is not a criticism of the measurement. It identifies what the endpoint directly demonstrates and what it is being asked to predict.

FDA defines clinical outcomes as direct measures of how people feel, function, or survive. A surrogate endpoint is used in place of that direct measure and is expected to predict clinical benefit.[1] Whether FDA will accept a proposed surrogate is case by case and context dependent, including the disease, population, mechanism, and development program.[4]

Endpoint, outcome, biomarker, and assessment

A clinical-trial endpoint is an outcome measured in the trial.[1] It may directly measure clinical benefit or may substitute for it as a surrogate.

A clinical outcome directly reflects how a patient feels, functions, or survives. FDA’s clinical outcome assessment types include patient-reported, observer-reported, clinician-reported, and performance outcome measures.[2]

A biomarker is a defined characteristic objectively measured as an indicator of normal biological processes, pathologic processes, or responses to an exposure or intervention. Molecular, histologic, radiographic, and physiologic characteristics can be biomarkers.[1] The joint FDA–NIH BEST resource maintains this terminology and the distinctions among biomarkers, clinical outcomes, and surrogate endpoints.[3]

A biomarker is not automatically a surrogate endpoint. It may serve diagnostic, prognostic, predictive, monitoring, safety, participant-selection, or response purposes without being accepted as a substitute for clinical benefit.[1]

Measurement type and endpoint role are separate layers. A clinical outcome assessment or biomarker generates measurements from which a trial endpoint may be defined. A COA-derived endpoint can directly assess clinical benefit; a biomarker-derived endpoint may be used for diagnosis, prognosis, enrichment, safety, response, or, with the required evidence and intended role, as a surrogate endpoint. “COA,” “biomarker,” “endpoint,” and “surrogate” therefore should not be treated as mutually exclusive labels.[1][2][3]

What makes an endpoint a surrogate

A surrogate endpoint does not directly measure the clinical benefit of primary interest; it is used as a substitute because an effect on it is expected to predict that benefit. The evidence supporting that expectation determines whether the endpoint is candidate, reasonably likely, or validated.[1]

For U.S. regulatory decision-making, FDA–NIH BEST distinguishes three categories based on the level of clinical validation:[1][3]

These are context-specific regulatory categories, not permanent labels attached to a measurement. Status may differ by disease, population, intervention or mechanism, endpoint definition, magnitude and duration of change, and intended claim.[4][6]

Individual-level association is not trial-level treatment-effect surrogacy

An individual-level or prognostic association asks whether people with different biomarker values tend to have different clinical outcomes. That association can be strong even when an intervention that changes the biomarker fails to improve, or worsens, the clinical outcome.

Trial-level surrogacy asks a different question: across appropriate randomized comparisons, do the estimated treatment effects on the surrogate predict the corresponding treatment effects on the direct clinical outcome, including their direction and magnitude? FDA–NIH BEST states that correlation between a biomarker and clinical outcome across individuals is necessary but not sufficient for treating the biomarker as a trial-level surrogate.[6]

Mechanistic plausibility and prognostic evidence are supportive, but neither shows by itself that treatment-induced changes predict clinical benefit. Validation generally draws on a clear mechanistic rationale and clinical data, often from multiple randomized trials spanning relevant interventions or mechanisms. The number and diversity of trials, the range and duration of surrogate effects, direct-outcome estimates, and uncertainty in the treatment-effect relationship all matter. A single favorable study ordinarily cannot establish broad validation outside its studied context.[6]

This distinction prevents a common leap:

1. The biomarker is associated with the disease.
2. The intervention changes the biomarker.
3. Therefore, the intervention improves how people feel, function, or survive.

Step three needs its own treatment-effect evidence. A biological response can document that the measured process changed without establishing clinical benefit.

Why trials use surrogate endpoints

Direct clinical outcomes may take years to observe, require very large studies, or be impractical or unethical in a particular setting. A credible surrogate can make a trial more efficient and allow an earlier assessment of an intervention’s effect.[1]

The advantage comes with uncertainty. An earlier or easier measurement may not capture:

• off-target effects that change the clinical outcome;
• harms that offset the expected benefit;
• the duration needed for the surrogate change to matter;
• whether the magnitude of change predicts a meaningful outcome; or
• differences among interventions that act on the same marker through different pathways.

The shorter route is informative only to the extent that the surrogate relationship is supported. These limitations are part of why FDA treats endpoint acceptability and validation as context specific.[4][6]

Traditional and accelerated approval are context specific

FDA’s public surrogate-endpoint table includes endpoints used to support traditional approval, accelerated approval, or endpoints FDA anticipates may be appropriate in a development program.[4] The agency states that acceptability is case by case and context dependent; an endpoint suitable in one program should not be assumed suitable in another.

For traditional approval, a surrogate known to predict clinical benefit may support approval; BEST calls a surrogate with strong evidence for a specific benefit in a defined context “validated.”[4][6] For accelerated approval, FDA may approve a drug for a serious condition based on an adequate and well-controlled trial showing an effect on a surrogate or intermediate clinical endpoint reasonably likely to predict clinical benefit, subject to the pathway’s requirements.[5][7]

These authorities describe marketing approval of a specific drug product under its application, not a general approval attached to an ingredient name.[4][5][7] Under 21 CFR 314.510, approval based on a reasonably likely surrogate is subject to further adequate and well-controlled study where uncertainty remains, and the applicant must carry out required studies with due diligence. FDA states that approval may be withdrawn or the labeled indication changed if confirmatory trials fail to verify sufficient clinical benefit.[5][7]

That product-and-application scope does not establish the identity, quality, safety, effectiveness, or comparability of a separately sourced research material sharing an ingredient name. See the separate FDA-approved drug versus research-material guide.

How to read a surrogate-endpoint trial

1. Identify the direct clinical outcome of interest

Write down what matters beyond the surrogate: a symptom, functional ability, event, survival outcome, or other patient-centered benefit. If the paper never names the outcome being predicted, the interpretation is incomplete.

2. Identify the measurement and then its endpoint role

First identify what generated the measurement: a patient-, observer-, clinician-, or performance-based clinical outcome assessment, a survival event, a biomarker, or another defined measure. Then identify the precise endpoint derived from it and ask whether that endpoint directly assesses clinical benefit or is being used as a substitute expected to predict benefit.[1][2][3]

3. Find the exact context of use

Record the disease, population, intervention class or mechanism, treatment setting, endpoint definition, measurement method, and time frame. Evidence does not automatically transfer across these boundaries.

4. Determine the validation level and evidence basis

Identify whether the endpoint is a candidate, reasonably likely, or validated surrogate for the exact context. Separate individual-level prognostic association from trial-level prediction of treatment effects. Record the number and design of supporting trials, intervention classes or mechanisms represented, direct clinical outcomes evaluated, magnitude and duration of surrogate changes, treatment-effect relationship, and uncertainty. Also ask whether the new trial’s surrogate effect lies within the range for which prediction has support. A prognostic citation or one biomarker-response study is not equivalent to surrogate validation.[1][3][6]

5. Read the surrogate effect and its uncertainty

Report the numerical treatment effect on the surrogate, confidence interval, denominators, time point, missing-data handling, prespecified status, and multiplicity treatment. Statistical significance on the surrogate does not quantify a clinical-benefit effect. Do not translate a one-unit surrogate change into a particular symptom, event, or survival benefit unless a validated prediction relationship supports that translation in the same context, and report the uncertainty in that prediction.[1][3][6]

6. Find direct-outcome, harm, and follow-up evidence

Ask whether direct clinical outcomes and harms were measured in the same trial, another trial, or a required confirmatory study. Report their effect estimates, confidence intervals, follow-up duration, completeness, and prespecified or exploratory status. An imprecise or underpowered direct-outcome result does not prove no clinical benefit, but a favorable surrogate result also cannot override observed clinical harm or a credible discordant direct-outcome result. Note whether confirmatory evidence is complete, ongoing, delayed, unavailable, or failed to verify the anticipated benefit.[5][6]

For related interpretation steps, use the research-study reading framework and the confidence intervals, p values, and multiplicity guide.

The FDA table is a starting point, not a universal endorsement

FDA updates its surrogate-endpoint table and warns that scientific understanding can change.[4] An endpoint previously used may no longer be accepted in the same way, while a listed endpoint may be relevant only under specified conditions.

The table also includes endpoints that FDA anticipates could be appropriate even if they have not yet supported an approved application. Presence in the table therefore should not be rewritten as “FDA validated this marker for all uses” or “a product using this endpoint is approved.”

The reliable statement is narrower: identify the endpoint, the table category, the development context, the specific application if one exists, and the evidence-review date.

Important limitations

• A biomarker response is not automatically a surrogate endpoint or a direct clinical benefit.
• Prognostic association alone does not validate a surrogate for treatment effects.
• Strong individual-level or prognostic association does not establish trial-level prediction of treatment effects.
• Mechanistic plausibility alone does not validate a surrogate, and surrogate validation does not require assuming that the marker fully mediates every treatment effect.
• A statistically significant surrogate effect does not provide a one-to-one estimate of clinical benefit.
• Validation evidence should report the range of interventions, effect magnitudes, follow-up, and uncertainty over which treatment-effect prediction has support.
• COA, biomarker, endpoint, and surrogate describe different measurement or use layers and are not four mutually exclusive categories.
• Surrogate validity is context specific and may differ by disease, population, intervention, and intended claim.
• A favorable surrogate result can coexist with uncertain clinical benefit or unmeasured harms.
• FDA’s surrogate table is updated and is not a substitute for application-specific review.
• Accelerated approval is approval of a named product and application, not of every material sharing an ingredient name.
• No endpoint result establishes the identity or quality of a separately sourced material.

Frequently asked questions

Is every laboratory measurement a surrogate endpoint?

No. A laboratory measurement may qualify as a biomarker when it is measured as an indicator of a biological process or response. It is a surrogate endpoint only when used as a substitute for a direct clinical outcome because it is expected to predict clinical benefit. Whether it is candidate, reasonably likely, or validated depends on the evidence and context.

Is a surrogate endpoint the same as a clinical outcome?

No. A direct clinical outcome reflects how people feel, function, or survive. A surrogate is used as a substitute because it is expected to predict that outcome; a candidate or reasonably likely surrogate may ultimately fail to do so.

Does correlation with a clinical outcome validate a surrogate?

Not by itself. Individual-level association or prognostic value does not establish that treatment effects on the surrogate predict treatment effects on the clinical outcome. Validation requires evidence for that treatment-effect relationship in the intended context.[6]

What does “reasonably likely to predict” mean?

It means the endpoint has strong mechanistic and/or epidemiologic rationale but insufficient clinical data to qualify as validated. Such an endpoint may support accelerated approval in a qualifying program, with further study required to verify and describe clinical benefit.[1][7]

Is accelerated approval a form of FDA approval?

Yes. It is marketing approval of the specific drug product evaluated under its application; it does not approve an ingredient in the abstract. Approval based on a reasonably likely surrogate carries follow-up obligations and residual uncertainty that should be described.[4][5][7]

Does an entry in FDA’s surrogate-endpoint table mean the endpoint works for every trial?

No. FDA states that acceptability is case by case and context dependent.

What should a study summary report about a surrogate endpoint?

Name the endpoint, the direct clinical outcome it is intended to predict, its validation level and context, the numerical effect and uncertainty, and any direct-outcome or confirmatory evidence.

Primary and official sources

1. FDA: Surrogate Endpoint Resources for Drug and Biologic Development
2. FDA: Clinical Outcome Assessment Frequently Asked Questions
3. FDA–NIH: BEST Biomarkers, EndpointS, and Other Tools Resource
4. FDA: Table of Surrogate Endpoints Used for Drug Approval or Licensure
5. FDA: Accelerated Approval
6. FDA–NIH BEST: Validated Surrogate Endpoint
7. Electronic Code of Federal Regulations: 21 CFR 314.510

Review and corrections

Sources checked: September 1, 2026. Questions or corrections can be sent through the Serum Research contact page.

Editorial responsibility: Serum Research prepared and maintains this educational page. No independent scientific, statistical, regulatory, or clinical review is claimed. See how Serum Research reviews evidence or return to the Research Peptide and Compound Library.