Rules/misleading presentation/US-MIS-023

Do not imply statistical rigor (e.g., via p-values) that the analysis lacks

AI-verifiedwarningFDA guidance · non-bindingRaised as a question — a person decides

Rule

Presenting p-values for a separate analysis of an individual component of a composite endpoint that does not support the efficacy conclusion being represented or suggested — e.g., because the analysis was not adequately powered for that endpoint and/or type 1 error (false positive rate) was not controlled — implies a statistical rigor that does not exist and would be false or misleading.

Check

Check whether p-values are shown for a separate analysis of an individual component of a composite endpoint where the analysis was not adequately powered to determine treatment effect on that component and/or type 1 error was not controlled (e.g., no multiplicity adjustment). A violation looks like a p-value attached to such an analysis, or a direct claim of demonstrated efficacy on the individual component. A.6 endorses presenting such results descriptively without p-values and without claiming a demonstrated effect, together with disclosure of the material limitations, as an acceptable alternative. This concern is stated for analyses of individual components of a composite endpoint; extending it to post-hoc subgroup analyses generally is an inference beyond this text, not something A.6 itself addresses.

Escalate when

Always — requires biostatistical review of whether the analysis was adequately powered/pre-specified.

Where this comes from

Q.6/A.6 · page 16. The highlighted words are quoted exactly; a script confirms they appear in the source before the rule can be published.

…ample, not adequately powered to determine treatment effect on the individual components of the composite endpoint and/or type 1 error (false positive rate) was Contains Nonbinding Recommendations not controlled in these analyses, these analyses would generally not support conclusions about a treatment effect of the product on the individual components of the composite endpoint. In such a case, representing or suggesting that the data support such efficacy conclusions, either directly (e.g., by claiming the product has demonstrated efficacy on the individual components) or indirectly (e.g., by presenting p-values, which would imply a statistically rigorous conclusion where one does not exist), would be false or misleading. If the firm wishes to present this information in a CFL promotional communication, it must do so in a truthful and non-misleading way. For example, the firm could present the results from these analyses of the individual components of the composite endpoint descriptively without p-values and without claiming that the results on the individual components are demonstrated additional effects of the drug. The firm sho…

Open the original at page 16 on fda.gov ↗Read our full copy →

Document: FDA guidance: Medical Product Communications that Are Consistent With the FDA-Required Labeling – Questions and Answers (final)
Version: as retrieved — the publisher does not version this document; the fingerprint below identifies the exact file
Retrieved: 2026-09-21 from www.fda.gov
Fingerprint: 1a18b7f242bfbbf897f198df…

FDA guidance describes the agency's current thinking. It is not binding law, and this rule is a recommendation, not an obligation.

Applies when

Naturepromotional

Basisconsistent-with-labeling

How this rule was checked

Citation proven by script and passed an independent verifier running on a different model. Not yet confirmed by a human.

Drafted
claude-code:sonnet · 2026-09-21
Independent verification — revise
claude-code:opus · 2026-09-21
Reviewer's reasons
- The quote is verbatim from A.6 and it is a statement of FDA's position on what would be false or misleading, not merely an illustration of what a rule does not require — so it can support a rule. But it sits inside a worked example ('For example, certain analyses of pivotal trial data ...' / 'In such a case ...') and is conditioned on a specific predicate: the pivotal trial was not adequately powered on the individual components of a composite endpoint and/or type 1 error was not controlled, so the analyses 'would generally not support conclusions about a treatment effect'. The rule's trigger ('not adequately powered or pre-specified') drops the type 1 error / multiplicity element and drops the requirement that the presentation actually represents or suggests an efficacy conclusion, which is the operative condition in the source.
- The Check broadens the rule beyond the source in two ways the source does not license. First, it extends from p-values to 'confidence intervals, or similar rigor markers'; A.6 names only p-values, and elsewhere in the same answer FDA expressly endorses presenting component results 'descriptively without p-values' — it does not say confidence intervals are themselves misleading. Second, it extends from separate analyses of components of a composite endpoint to 'post-hoc or non-pre-specified subgroup' analyses generally. Both are the drafter's extrapolation, not this source's text.
- 'judgment' is correct: although the presence of a p-value is a literal match, whether the analysis was adequately powered or pre-specified — and whether the presentation implies an efficacy conclusion — is a matter of biostatistical judgment.
- 'guidance' is correct for authority: the sentence is FDA's characterization in a nonbinding guidance, and the rule does not claim statutory or regulatory force.
- Severity 'warning' is defensible; the CFL scope limit present on the sibling rule US-SUB-008 (basis: consistent-with-labeling) is missing here.
Revised by the drafter
claude-code:sonnet · 2026-09-21
Independent verification — pass
claude-code:opus · 2026-09-21
Reviewer's reasons
- The quote is a self-contained clause (gerund subject 'representing or suggesting that the data support such efficacy conclusions' plus predicate 'would be false or misleading') and does not run into unrelated text.
- The quote's back-reference ('such efficacy conclusions') is cured in the rule statement, which supplies the source's own conditions — separate analyses of individual components of a composite endpoint where the trial was not adequately powered for those components and/or type 1 error was not controlled. That is exactly the scenario A.6 describes, so the rule does not over-generalize the passage.
- Not overstated: FDA itself says this presentation 'would be false or misleading,' and the rule says the same; the check also correctly records A.6's endorsed alternative (descriptive presentation without p-values, with limitations disclosed) and explicitly declines to extend the rule to post-hoc subgroup analyses generally.
- Atomic: a single prohibition — representing or suggesting an efficacy conclusion the analysis does not support, whether by claim or by p-value. The direct and indirect forms named in the check are two faces of the same requirement, not separable obligations.
- Determinism 'judgment' is correct: whether an analysis was adequately powered/pre-specified and whether the presentation implies unwarranted rigor requires biostatistical assessment. Authority 'guidance' matches the source; the locator and applies_when are right, and 'warning' is within a defensible range.

Nearby rules

US-MIS-022 US-MIS-024 US-MIS-025 US-MIS-056 US-SUB-008