Composite & Co-Primary Endpoints

A trial may need more than one outcome to represent a complex condition. It can require success on several separate endpoints, allow success on any one of several endpoints, or combine events into one endpoint. Those choices create different success rules and different interpretations.

The label matters, but the prespecified rule matters more. A favorable composite result does not establish a favorable effect on every component. Success on one endpoint does not satisfy a co-primary design that requires all co-primary endpoints to succeed.[1][2]

Start by identifying the endpoint architecture

FDA’s October 2022 final guidance and accompanying explanation distinguish different success rules.[1][2][3] Importantly, multiple primary endpoints is an umbrella description: co-primary endpoints are the all-required case, not an entirely separate category. A trial with several primary endpoints might instead allow success on any one, or on a specified subset. Read the rule rather than assuming the label settles it.[2][6]

Structure	Unit analyzed	Typical success rule	Main interpretation risk
Co-primary endpoints	Two or more separate endpoints	All specified endpoints must meet their criteria	One failed co-primary endpoint means the full co-primary objective was not met
Multiple primary endpoints with an any-one success rule	Two or more separate endpoints	Any one may be sufficient under a prespecified multiplicity plan	Repeated opportunities can inflate false-positive risk
Composite endpoint	Several event types combined into one endpoint	One test of the prespecified combined outcome	Less important or more frequent components can drive the result
Multi-component endpoint	Several measurements combined within each participant	One score or responder rule generated from all components	Weighting or scoring can hide discordant components

These structures are not interchangeable, and their names are not always used consistently across sources. ICH E9 uses “composite variable” broadly enough to include combined rating scales; this guide uses FDA’s 2022 distinction between event composites and within-participant multi-component outcomes.[2][6] “The study had multiple outcomes” is too vague to reveal which rule determined success. FDA guidance provides nonbinding recommendations, not a decision that a particular trial or product qualifies for approval.[1][2]

Composite endpoints combine events into one outcome

A composite endpoint treats any of several prespecified event types as contributing to one outcome. A common time-to-first-event composite records the time until a participant experiences the first qualifying component. Another composite may count total qualifying events. The exact definition, adjudication rules, time horizon, and handling of repeat events should be stated.

When the trial performs one prespecified statistical test on a single composite, FDA explains that the composite itself does not create a multiple-endpoint multiplicity problem.[2] That does not make interpretation simple. Components may differ greatly in clinical importance, frequency, susceptibility to bias, and expected treatment effect.

Suppose a composite includes an irreversible event and a more frequent discretionary event. A favorable result could be driven mostly by the frequent component even if the irreversible component shows little information, no apparent benefit, or a numerically unfavorable pattern. The correct statistical headline concerns the composite as defined, not each event type. A serious adverse direction in an important component can also undermine an overall clinical-benefit interpretation, even when the composite test is statistically significant.[2] The surrogate and clinical-outcome guide explains why the outcome’s clinical meaning matters.

Component results explain but do not automatically confirm

FDA recommends examining and reporting the components of a composite.[2] This helps readers understand which events were common and which appear to contribute to the overall estimate.

Component question	Why it matters
How clinically important is the component?	A change in a less consequential event may not carry the meaning of a change in death or irreversible morbidity
How frequent was it?	Frequent components can dominate the number of composite events
How objectively was it measured?	Components requiring discretionary decisions can be more vulnerable to ascertainment bias
Was the direction consistent?	Opposing component estimates can complicate the composite’s clinical meaning
How precise was the estimate?	Rare components may have wide intervals and little information
Was component testing prespecified?	Descriptive component estimates do not become confirmatory merely because the composite succeeded

A statistically positive composite does not transfer its significance to each component. For analyses that merely explain the demonstrated composite effect, FDA does not recommend a multiplicity adjustment solely for that descriptive purpose. Separate confirmatory component-effect claims require prospective hypotheses and appropriate error control; component tests may also be underpowered because the trial was sized for the composite.[2] Report estimates and uncertainty without turning descriptive p values into protected claims.

Time-to-first-event composites do not count total event burden

A time-to-first-event composite counts the first qualifying event, not every later event. A breakdown of those first events can therefore omit a later, more serious component experienced by the same participant. This does not mean follow-up or data collection necessarily stopped. FDA also recommends component analyses that include events occurring after another component type.[2]

Check whether:

components compete with or preclude one another;
a frequent early component prevents later components from entering the primary analysis;
recurrent events were analyzed separately;
follow-up after the first event continued; and
the hazard ratio or other effect measure matches the composite’s time-to-event definition.

A first-event breakdown, the number of participants ever experiencing each component, and a recurrent-event analysis answer different questions. Participants can experience more than one component, so component-specific participant counts should not simply be added and called unique composite participants. The composite result is not automatically a statement about the probability of ever experiencing each component.[2]

Co-primary endpoints require success on all specified features

Co-primary endpoints are used when demonstrating an effect on every named feature is considered necessary to establish the intended benefit. In this all-required form of multiple-primary design, every co-primary endpoint criterion must be satisfied.[2]

The all-required criterion does not, by itself, require splitting the significance level to protect against several alternative paths to success. Other comparisons in the trial may still require multiplicity control. Requiring all endpoints can reduce joint power: with two independent endpoints each having 80% power under the assumed effects, the probability of success on both is 64% (0.8 × 0.8). Correlation or unequal powers changes the calculation. FDA explicitly rejects relaxing each endpoint’s significance threshold merely to compensate for this power loss.[2][6]

This leads to a simple reporting rule:

If all co-primary criteria are met, describe success on the co-primary endpoint family.
If one succeeds and one fails, state that the full co-primary objective was not met.
Do not promote the favorable endpoint to “the primary result” after seeing the data.

A favorable p value on one co-primary endpoint cannot compensate for failure on another in an all-required design. A genuinely different prospective success rule must be described as such. Failure to meet the objective is not proof of zero effect; the null-result guide explains how estimates and intervals constrain that interpretation.

Any-one multiple-primary designs provide several paths

Some trials designate several primary endpoints and allow an effect on any one to support the study objective. That creates multiple opportunities to declare success. FDA recommends a prospective plan controlling the overall false-positive probability for the relevant hypothesis family, using ordering, allocated significance levels, gatekeeping, or another justified procedure.[2] ICH E9 likewise places primary-variable specification and error control within statistical planning.[6] The confidence-interval and p-value guide provides related interpretation boundaries.

The phrase “one of three primary endpoints was significant” is uninterpretable without the testing strategy. Ask:

Were all primary endpoints prespecified?
Was success on any one truly sufficient?
What overall Type I error rate was controlled?
Was the reported p value adjusted or compared with an allocated threshold?
Did an earlier test in a fixed sequence fail, preventing formal testing of later endpoints?

In a pure fixed-sequence procedure, protected testing stops at the first failed test. A later small p value cannot bypass that gate. It may be reported as nominal or supportive, but does not inherit confirmatory status under that procedure.[2] This does not retroactively change an endpoint’s original primary or secondary designation; its planned role and the status of the inference are separate.

Multi-component endpoints create one participant-level value

A multi-component endpoint combines observations within each participant to produce one overall score or responder status according to a specified rule. This differs from the event-composite usage above.[2]

“All required” can mean two different things. Two separate comparisons showing group-level improvement on two endpoints are co-primary tests. One responder endpoint requiring each individual to meet two criteria is a within-participant multi-component outcome. Success on the former does not mean every individual improved on both; the latter depends on the joint response in each person.[2]

Interpretation depends on the construction rule. Determine:

which components enter the score;
whether they are summed, weighted, ranked, or required jointly;
how missing components are handled;
whether one component can dominate the total;
whether higher values always have the same meaning; and
whether the combined change is clinically interpretable.

A favorable overall score does not establish improvement on every input. Some individual subscales may not be independently clinically interpretable, even when the combined scale is. Clinically meaningful, separately confirmatory component claims need prospective specification and multiplicity control.[2][6] A scoring rule also does not resolve missing-data assumptions.

Endpoints must be completely defined

CONSORT 2025 recommends defining every primary and secondary outcome by the specific measurement variable, participant-level metric, method of aggregation, and time point, and reporting numerical group comparisons with estimates and precision.[4][5] A composite needs clarity about components and the combination rule. A multiple-primary family needs its actual success rule, all, any one, or a specified subset, and the corresponding error-control justification. CONSORT’s recommendation to report components does not itself make their tests confirmatory.[2][4]

This hypothetical reporting example is more informative than “the primary endpoint was positive”:

The prespecified time-to-first-event composite met its criterion. The estimate was driven mainly by the most frequent component; individual component analyses were descriptive and imprecise.

That statement preserves what succeeded without turning the overall finding into unsupported component claims.

A reading checklist

Name every primary endpoint and reproduce its complete definition.
Identify separate endpoints versus a combined outcome, and the exact all-required, any-one, or subset rule.
State the exact prespecified success rule.
Find the sample-size assumptions for each required endpoint.
Identify the multiplicity plan and test sequence.
For a composite, tabulate every component by group and note first versus recurrent events.
Compare component importance, frequency, ascertainment, direction, and precision.
Separate descriptive component analyses from protected confirmatory tests.
Check the protocol, analysis plan, registry, and report for changed definitions.
Limit the statistical conclusion to the endpoint family that met its rule, then assess clinical meaning and harms separately.

If the original protocol or statistical analysis plan is unavailable, distinguish what the publication says was prespecified from what a dated plan independently establishes. A current registry label or later document does not prove the original testing sequence. ICH E9 expects principal analyses in the protocol, documented amendments and a plan finalized before unblinding; CONSORT asks authors to identify and explain deviations.[5][6] Report missing versions and unresolved timing rather than certifying a hierarchy that was not obtained.

Important limitations

A statistically favorable composite does not establish favorable effects on all components.
A composite can be driven by a frequent or less important component.
Individual component estimates may be underpowered and should not be called equivalent when uncertain.
Success on one co-primary endpoint does not satisfy a design requiring success on all.
Multiple primary endpoints require a prospective multiplicity strategy when any endpoint can provide success.
A multi-component score can hide opposing changes in its inputs.
Prespecification reduces interpretive flexibility but does not ensure clinical relevance or unbiased measurement.
No endpoint result authenticates or establishes comparability with a separately sourced catalog material.

Frequently asked questions

What is a composite endpoint?

It is one endpoint formed from multiple prespecified event types. The trial tests the combined outcome according to a defined rule, such as time to the first qualifying event.

Does a positive composite mean every component improved?

No. The overall result may be driven by one or a few components. Component estimates can differ in direction and precision.

What are co-primary endpoints?

They are separate primary endpoints for which the protocol requires success on all named endpoints to meet the study objective.

Are co-primary endpoints the same as multiple primary endpoints?

Co-primary endpoints are the all-required case within FDA’s multiple-primary framework. Other multiple-primary designs allow success on any one or a specified subset. The explicit success rule, not the broad label alone, determines the relevant multiplicity problem.

Does one test of a composite solve every multiplicity issue?

It avoids multiple-endpoint multiplicity for the single composite test. Separate confirmatory claims about components, time points, populations, or secondary endpoints can create additional multiplicity.

Why report components if they are not separately powered?

They show what drove the composite and whether clinically important components point in consistent or uncertain directions. Their descriptive role should be stated.

Can a favorable endpoint result validate a similarly named catalog material?

No. It applies to the defined intervention and trial conditions. It does not establish identity, quality, or comparability of another material.

Primary and official sources

Review and corrections

Sources checked September 3, 2026 (America/Edmonton and UTC). Serum Research is responsible for this editorial guide. No independent clinical, statistical, scientific or regulatory review is claimed. This is not medical advice, a regulatory determination, or evidence of a material’s identity, quality or equivalence.

See how we review evidence, browse the research library and study records, or contact us with a correction and its source.

Are you over 21?

Composite Endpoints, Co-Primary Endpoints, and Components