Rules/misleading presentation/US-MIS-018

Retrospective statistical mining to find unsupported findings may mislead

AI-verifiedwarningBinding lawRaised as a question — a person decides

Rule

An ad may be misleading if it uses retrospective statistical mining to discover and cite findings not soundly supported by the original study.

Check

Determine whether cited findings resulted from pre-specified analysis or retrospective/post-hoc statistical mining not soundly supported by the study.

Escalate when

Distinguishing sound retrospective analysis from unsupported data-mining requires statistical expertise.

Where this comes from

21 CFR 202.1(e)(7)(iii). The highlighted words are quoted exactly; a script confirms they appear in the source before the rule can be published.

…t if it: (i) Contains favorable information or conclusions from a study that is inadequate in design, scope, or conduct to furnish significant support for such information or conclusions. (ii) Uses the concept of “statistical significance” to support a claim that has not been demonstrated to have clinical significance or validity, or fails to reveal the range of variations around the quoted average results. (iii) Uses statistical analyses and techniques on a retrospective basis to discover and cite findings not soundly supported by the study, or to suggest scientific validity and rigor for data from studies the design or protocol of which are not amenable to formal statistical evaluations. (iv) Uses tables or graphs to distort or misrepresent the relationships, trends, differences, or changes among the variables or products studied; for example, by failing to label abscissa and ordinate so that the graph creates a misleading impression. (v) Uses repor…

Open the original at these words on ecfr.gov ↗Read our full copy →

Document: 21 CFR 202.1
Version: 2026-09-16 — the link above is pinned to this date, so it shows the text as it stood then
Retrieved: 2026-09-21 from www.ecfr.gov
Fingerprint: a542640a872b3f2702d549e8…

Applies when

Naturepromotional

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-sonnet-5 · 2026-09-21
Independent verification — pass
claude-opus-5 · 2026-09-21
Reviewer's reasons
- The quote is a verbatim clause of 21 CFR 202.1(e)(7)(iii), which sits under the operative heading '(7) Advertisements that may be false, lacking in fair balance, or otherwise misleading. An advertisement may be false ... if it:'. This is an operative regulatory provision, not an example, definition, or preamble, so the rule statement follows directly from it.
- Modality is faithful: the source says an advertisement 'may be' misleading and the rule says 'may be misleading', with severity 'warning' rather than a hard prohibition. No strengthening of 'may' into 'is prohibited'.
- Scope is not broadened. Subsection (iii) contains two independent clauses (retrospective mining for unsupported findings; and using statistics to suggest validity for studies whose design is not amenable to formal statistical evaluation). The rule carves out only the first clause, so it is a narrower slice rather than a bundle — atomic is satisfied. The check correctly retains the qualifier 'not soundly supported by the study', so it does not condemn all post-hoc analysis.
- determinism 'judgment' is right: whether a retrospectively derived finding is 'soundly supported by the study' turns on statistical and scientific assessment, not presence/absence of a string. The escalation note (statistical expertise) is consistent with this.
- authority 'binding' is correct — 21 CFR 202.1 is a codified regulation, and this paragraph is operative (unlike (e)(6)(ii) and (vii), which the note records as stayed).
- applies_when 'promotional' is justified: § 202.1 governs prescription-drug advertisements, and § 202.1(l) confirms the reach over advertising and promotional labeling pieces. No invented limitation and no obvious missing limit.
- Locator 21 CFR 202.1(e)(7)(iii) matches the quoted text exactly; category 'misleading-presentation' is a defensible fit for a provision about misleading use of statistical findings.

Nearby rules

US-MIS-019