Primary vs Secondary Endpoints: Audit a Trial Readout

A trial announcement highlights an improved secondary endpoint after the primary analysis misses. Does that result establish the study's intended benefit, or does it generate a question for another study? Start with the protocol and statistical analysis plan (SAP), then compare the release with those documents.

This worksheet helps you record what is supported before changing a pipeline valuation assumption. It does not turn a positive headline into an approval prediction.

1. Identify the success rule before reading the headline

A study can have one primary endpoint or several. With co-primary endpoints, the intended conclusion requires success on each specified endpoint. Other designs permit success on at least one of several primary endpoints, with appropriate multiplicity control. Secondary endpoints can support additional effects when their testing is protected; exploratory results generate hypotheses. A label such as “key secondary” alone does not establish that protection. FDA Multiple Endpoints guidance, sections III.A and III.C

Create one row per reported outcome:

Field to capture	Evidence to locate
Endpoint and assessment time	Dated protocol and amendments
Primary, secondary or exploratory role	Protocol and SAP
Success rule and testing order	SAP's multiplicity procedure
Population and treatment comparison	Analysis definition and results
Effect estimate, uncertainty and reported p-value	Original results table

Keep the document version and page beside each entry. If the relevant SAP is unavailable, record the testing rule as UNKNOWN. A registry record is a useful starting point; use our ClinicalTrials.gov guide to distinguish its dated fields from additional evidence you still need.

2. Follow the actual testing procedure

In a fixed-sequence procedure, a non-significant test stops formal testing of the subsequent hypotheses. Later small p-values do not restart that sequence. Other multiplicity procedures allocate or recycle significance levels differently, so do not assume every trial uses the same gate. Statistical testing also does not establish clinical relevance by itself. FDA Multiple Endpoints guidance, section II.A and appendix 5–8

Hypothetical example, not a real trial: a SAP tests endpoint A first, then B, each at a two-sided 0.05 significance level only if the preceding test succeeds. Reported p-values are 0.08 for A and 0.01 for B. A does not pass; B's value alone cannot support a confirmatory claim under that fixed sequence. It may motivate further investigation. Do not apply this conclusion to a different testing plan without reading it.

3. Check what treatment effect was estimated

An endpoint name is not the whole clinical question. Identify the population, treatments, outcome and summary measure, plus how events such as treatment discontinuation or rescue medication enter the question. Those choices define the estimand. Sensitivity analyses examine whether conclusions remain robust to assumptions for that same estimand; an analysis targeting a different question needs separate interpretation. FDA ICH E9(R1), sections III.C and V.B–C

Ask whether the press release and results table use the same population and time point. Record the analyzed denominator and missing assessments instead of substituting planned enrollment. Keep an effect estimate and its uncertainty attached to the comparison. For blinding and comparator questions, use the trial-design companion guide.

4. Write an evidence conclusion you can revisit

Use three short statements in your research notes:

1. Established: which prespecified test succeeded under the documented procedure, and what effect was reported?
2. Not established: which claim depends on an unavailable SAP, an unprotected analysis or a different population?
3. Next evidence: which protocol, results table or subsequent study would resolve the uncertainty?

Keep safety findings in a separate part of the worksheet. Missing a primary test is not proof of zero effect, and passing one is not an approval guarantee. Do not assign an approval probability from endpoint wording alone.

To apply the trial-design framework in BioSniper, choose a company in the research entry on this page. The existing task asks about randomization, comparator, blinding, the primary endpoint and missing-data handling, using original trial evidence where available and marking unknowns. Keep your endpoint worksheet alongside that research; the article is background, not evidence of a company's current results. You review the company and research task before running it.

Primary sources checked October 9, 2026. Examples marked hypothetical are teaching assumptions, not company facts.