When a sponsor states in Section 14 that its drug outperforms a comparator, or presents a secondary endpoint result, FDA reads that language as a labeling claim. Like an indication, it has to rest on substantial evidence. Regulatory and clinical teams therefore need to understand the evidentiary bar early: trial design, comparator dosing, and the multiplicity strategy are all fixed long before labeling negotiations begin, and a decision made at protocol stage can later rule out the label language the commercial team expects.
The analysis below covers the regulatory basis for comparative and superiority statements under 21 CFR 201.56 and 201.57 and the FDA guidance that applies it. It then looks at how reviewers have handled active-comparator data and secondary endpoint results in specific applications, including when those results were kept in the label, moved to descriptive presentation, stripped of p-values, or removed.
Comparative claims and secondary endpoints in the Clinical Studies section: what FDA requires
FDA treats a comparative or superiority statement in prescribing information as a claim. It must be supported by substantial evidence from adequate and well-controlled trials designed to test that comparison. Secondary endpoints face a related but separate standard. A secondary result can support labeling language when it was prespecified and tested inside a multiplicity-controlled strategy, and, for patient-reported outcomes, measured with a fit-for-purpose instrument. If not, reviewers have consistently pushed it out of the label, into descriptive presentation, or stripped of its p-values.
This article covers the regulatory basis, the guidance framework, and how FDA reviewers have applied both standards in specific applications.
Key takeaways
- An explicit superiority or similar-effectiveness claim requires substantial evidence (21 CFR 201.57(c)(2)(iii) for comparative statements in Indications and Usage, 201.57(c)(7) for adverse-reaction comparisons, and FDA's Clinical Studies guidance for Section 14). For superiority, that means adequate and well-controlled trials designed to establish superiority, with the comparator given at an appropriate dose and regimen (generally its highest recommended dose) in an appropriate population 66.
- Non-inferiority alone does not establish that two drugs are equally or nearly equally effective 67. A similar-effectiveness claim needs a margin smaller than the one that merely establishes effectiveness 66.
- Active-comparator results generally should not appear in Section 14 when placebo establishes effectiveness, unless they support an explicit comparative claim 63.
- Whether an endpoint is labeled "primary" or "secondary" is not decisive. What matters is a well-documented, statistically and clinically meaningful effect on a prospectively defined endpoint 208209.
- A secondary endpoint supports a definitive claim only after the primary endpoint family succeeds and only within a prespecified Type I error control strategy 27282930.
The regulatory foundation
The labeling regulations set the baseline. Under 21 CFR 201.56, labeling must be based on human data whenever possible. It may not carry implied claims or suggestions of use when evidence of safety is inadequate or substantial evidence of effectiveness is lacking 293.
21 CFR 201.57(c)(15) defines the Clinical Studies section. It should discuss the studies that help clinicians understand how to use the drug safely and effectively, including design, population, endpoints, and results. It must not become an encyclopedic listing of the development program 266267. Any study discussed in connection with an indication must be adequate and well-controlled (nonbiological drugs) or constitute or contribute to substantial evidence (biological products). The section must not imply indications, uses, or dosing regimens that are absent from Indications and Usage or Dosage and Administration 268.
Section 201.57(c)(15) contains no separate rule on drug-to-drug comparisons 266267. The comparative standard is stated elsewhere in 201.57:
- Statements in Indications and Usage that compare a drug's safety or effectiveness with other agents for the same indication must be supported by substantial evidence from adequate and well-controlled studies 147.
- Claims comparing adverse reaction rates between drugs must rest on adequate and well-controlled studies for nonbiological drugs, or on substantial evidence for biological products 148.
In practice, the implied-claims prohibition in 201.56 and the Section 14 requirement that studies be adequate and well-controlled are the main tools reviewers use to keep unsupported comparisons out of the label.
What FDA guidance requires for comparative claims
Clinical Studies section guidance
FDA's guidance on the content and format of the Clinical Studies section separates two things that are easy to confuse: reporting active-control data and making a comparative claim.
- Placebo-established effectiveness. If effectiveness rests on placebo-controlled data, active-comparator results generally should be left out unless they adequately support an explicit superiority or similar-effectiveness claim. For example, when an active arm in a three-arm trial served only to confirm assay sensitivity, its identity and results should be omitted if they do not support a comparative claim and are not otherwise important to clinicians 63.
- Active-control-dependent effectiveness. If effectiveness can be shown only against an active control (in a superiority or non-inferiority design) and knowing the comparator matters to clinicians, the comparator's identity and data should be included. If no comparative claim was established, the labeling should say so and disclose limitations, such as a suboptimal or unapproved comparator regimen 6465.
- Exclusion criteria. The guidance says Section 14 usually should exclude active-control studies that imply comparative claims unsupported by substantial evidence, studies that imply unapproved uses or regimens, and studies that are not adequate and well-controlled. If such a study is included under an exception, its limitations and the reason for including it should be stated 238215.
For an explicit claim, the evidentiary bar is specific 66:
| Claim type | Evidence the guidance identifies |
|---|---|
| Superiority | Adequate and well-controlled trials designed to establish that one treatment is superior to the other |
| Similar effectiveness | Adequate and well-controlled trials designed to show non-inferiority and that any difference is not clinically significant. The margin must be tighter than an effectiveness margin; a 50% preservation-of-effect margin would ordinarily be too large |
| Comparator conduct (both) | Comparator at an appropriate dose and regimen, generally its highest recommended dose, in an appropriate patient population |
The guidance also calls for a fair comparison that does not inappropriately favor either treatment 158.
Non-inferiority and superiority from the same trial
FDA's non-inferiority guidance allows a trial planned for non-inferiority to show superiority to the active control. That finding can be interpreted without a multiplicity adjustment across the non-inferiority and superiority tests, and it is persuasive evidence of effectiveness. When multiple endpoints or doses are tested, the testing order needs careful planning and multiplicity adjustment may be needed 142143. The guidance is explicit that non-inferiority methods establish that a drug is effective, not that it is as effective as the comparator 67.
Disease-specific and promotional guidance
- Rheumatoid arthritis. FDA's RA clinical development guidance calls for two adequate and well-controlled trials showing superiority to support a claim of superiority over a specific comparator 159.
- Consistent-with-labeling communications. FDA's Q&A on communications consistent with FDA-required labeling states that a numerically better result against an active comparator does not support a superiority claim when the study was not designed to test superiority between the drugs. Such a study should not be presented as evidence of superiority or non-inferiority 160.
How reviewers have applied the comparative-claim standard
Drugs@FDA review documents show the same reasons for rejecting comparative claims coming up again and again: a single trial, an unfair comparator dose or setting, reliance on a non-inferiority analysis, and differences between the trial and the marketed product. The examples below record reviewer conclusions. The review excerpts do not always show the final approved Section 14 wording 6871737482.
| Product vs comparator | Reviewer reasoning | Recorded outcome |
|---|---|---|
| Alimta (pemetrexed) vs docetaxel | The single trial did not show superior overall survival and missed its prespecified fixed-margin non-inferiority test. The alternative non-inferiority analysis depended on an uncertain estimate of docetaxel's effect from one small study, and crossover made interpretation harder | Statistical reviewer found the evidence insufficient for the proposed comparative efficacy claim 86 |
| Zyprexa (olanzapine) vs haloperidol | Statistically significant differences were clinically small, and "numerically superior" differences were neither statistically nor clinically significant. The haloperidol dose range and patients' prior unfavorable experience with haloperidol made the comparison unfair. Non-randomized extensions could not support long-term comparison | Reviewer recommended not using the study to support an efficacy claim and removing language implying superiority 87 |
| Lipitor (atorvastatin) vs other statins | Well-conducted trials supported greater LDL-C lowering with atorvastatin 10 mg against the specific comparator doses tested. Some confidence limits approached clinically insignificant differences, and higher comparator doses might match the effect | Reviewer supported a dose-specific superiority presentation, with confidence limits and a statement that superiority over higher comparator doses was not established 88 |
| Bydureon (exenatide ER) vs Byetta | Study 105 showed non-inferiority and superiority on HbA1c, but the to-be-marketed batches produced lower exenatide exposure than the trial batches. Extension data suggested the marketed formulation might be less effective | Reviewer concluded that using Study 105 to imply superiority could overstate the marketed product's benefit and would not support the claim in the package insert 89 |
| Cibinqo (abrocitinib) vs dupilumab | The proposed trial planned separate comparisons with placebo, not a formal head-to-head test | In advice on the proposed design (before the Phase 3 program was finalized), FDA advised that a comparative claim would need prespecified, formally tested endpoints with appropriate timepoints, and generally two Phase 3 trials for superiority. Descriptive comparisons alone generally would not support comparative labeling 90 |
| Effient (prasugrel) vs clopidogrel | FDA stressed testing clopidogrel where its efficacy was established (within 24 hours of symptom onset, at its most efficacious regimen). Beating clopidogrel in a setting where its efficacy was not established would be an unfair comparison | FDA rejected that setting as a basis for the claim. It indicated that particularly strong evidence from one trial, or positive results from two trials, might support superiority 9192 |
Several principles follow from these reviews:
- The claim must be bounded by what was tested. The Lipitor review allowed superiority only at the comparator doses studied and required qualifying language about higher doses 88.
- Comparator fairness is a threshold question. Under-dosed or inappropriately selected comparators (Zyprexa 87) and clinical settings where the comparator's efficacy was not established (Effient 9192) were enough to defeat the claim.
- Replication is the default expectation. Reviewers and guidance repeatedly point to two adequate and well-controlled superiority trials, with single-trial support reserved for particularly strong results 909192159.
- The trial product must be the marketed product. The Bydureon review shows that a comparative result can be set aside if exposure differs between clinical and commercial batches 89.
- Statistical significance is not the same as clinical meaningfulness. Small but significant differences did not support superiority language for Zyprexa 87.
Secondary endpoints: the guidance framework
Multiple Endpoints guidance
FDA's Multiple Endpoints guidance separates secondary endpoints that describe the primary finding from those meant to demonstrate an additional clinically important effect. A positive secondary result can support a conclusion about an additional effect only after a treatment effect has been shown in the primary endpoint family, with overall Type I error controlled across the primary and secondary families 272829. A definitive claim of an additional effect should come from an analysis inside the prespecified multiple-testing strategy. Descriptive analyses should not be presented in labeling as if they establish a rigorous conclusion 30.
Sponsors are expected to prespecify the endpoints, the analyses, and a testing procedure that controls overall Type I error. Changes made after unblinding can undermine interpretability 3132. The guidance describes hierarchical (fixed-sequence) testing and gatekeeping across ordered endpoint families, where conclusions in later families depend on results in earlier ones 3334. In a fixed-sequence procedure, once an earlier test fails, the remaining endpoints are not statistically significant under that procedure. A nominal secondary p-value therefore does not become a labeling claim 293536.
Clinical Studies guidance on endpoints and data presentation
- The primary or secondary designation does not decide the question. The test is whether the drug has a well-documented, statistically and clinically meaningful effect on a prospectively defined endpoint. Section 14 may also discuss endpoints that were expected to be affected but were not 208209.
- Section 14 should summarize the evidence of effectiveness, critical design features, and important limitations, not just favorable findings 237.
- Individual components of a composite endpoint should generally be presented, but discussion of a single component should usually be descriptive, without statistical analysis, unless that component was assessed under a prospectively defined hypothesis and analysis plan 244.
- Confidence intervals and p-values give complementary information and should usually both be presented. A p-value generally should not be used alone. Results should show both the study-drug and comparator groups, because the between-group difference conveys the treatment effect 241242.
- Descriptive subgroup analyses may be considered for labeling without p-values, but they should not be presented as establishing effects with more certainty than the trial supports 30.
Patient-reported outcome endpoints
Under FDA's PRO guidance, a PRO endpoint can support a labeling claim with the same amount and kind of evidence as any other claim. The trial must be appropriately designed and use a well-defined, reliable instrument whose measurement properties match the claim 321. To support a claim, a PRO should be primary, co-primary, or a secondary endpoint in a prespecified hierarchy. Endpoints outside the primary and key-secondary hierarchy are exploratory, and favorable PRO results should not be selected post hoc 322323. In oncology, exploratory findings are descriptive and generally unlikely to support a comparative-benefit claim 324. Content validity is foundational: reliability or other validity evidence cannot make up for inadequate content validity 326327328329.
How reviewers have handled secondary-endpoint claims
Review documents show a fairly consistent set of outcomes. Secondary endpoints that pass the prespecified hierarchy are labeled. Those below a failed step are dropped or shown descriptively. Analyses without multiplicity control lose their p-values or are left out entirely.
| Product | Secondary endpoint issue | Reviewer position or outcome |
|---|---|---|
| Invega (paliperidone) | Personal and Social Performance Scale (PSP) in a Dunnett-based hierarchy controlling overall Type I error | Reviewer supported a PSP claim only for doses positive on both the primary PANSS endpoint and PSP 186113114. Reviewer separately rejected the sponsor's request to keep certain exploratory findings as labeling claims 113 |
| Perjeta (pertuzumab) | Overall survival, the first secondary in the hierarchy, was not significant | The next secondary endpoint was included descriptively, not as a successful hierarchical test 187 |
| Spiriva (tiotropium), UPLIFT | Trial failed at the first layer of its prespecified closed-testing hierarchy | Primary statistical reviewer opposed additional claims beyond the long-term lung-function effect because of Type I error inflation. A secondary reviewer disagreed and argued that consistent evidence across studies could be considered 115 |
| Fabior (tazarotene) | Secondary acne measures analyzed without multiplicity adjustment | Reviewer said other secondary endpoints should not be in labeling. Percent reduction in lesion counts was noted as typically included as supportive of absolute reduction 185 |
| Zytiga (abiraterone) | Three secondary endpoints, including PSA response, under a Hochberg procedure | Despite Type I error control, concerns about endpoint validity and clinical interpretation led the reviewer to recommend leaving the secondary endpoints out of labeling 188 |
| Cymbalta (duloxetine), fibromyalgia | Patient global improvement and FIQ Total, with many unadjusted dose and endpoint comparisons | Reviewer found insufficient evidence and recommended either multiplicity control or descriptive presentation without p-values 116117286 |
| Liptruzet (ezetimibe/atorvastatin) | Secondary lipid endpoints under a false-discovery-rate procedure | Reviewer concluded the procedure did not control Type I error across the comparisons in the proposed label. The excerpt does not show the final Section 14 disposition 189190 |
| Otezla (apremilast) | Endpoints below a nonsignificant MASES result in the hierarchy, including ACR 50 | Reviewer noted they had only nominal significance despite appearing in the proposed label 191 |
| Wakix (pitolisant) | Cataplexy endpoint in HARMONY 1 | No prospective secondary-endpoint multiplicity plan, and the significant finding depended on how missing data were handled 192 |
| Calquence (acalabrutinib) | Investigator-assessed ORR, ORR+PRL, and OS in ELEVATE-TN; sensitivity analyses in ASCEND | FDA disagreed with presenting p-values because the analyses were exploratory, had no alpha allocation, or followed a failed prerequisite test (OS also had immature data). The excerpts concern review tables and do not confirm Section 14 placement 287288289290291 |
| Nocdurna (desmopressin) | Proposed Nocturia-Quality of Life and Sleep Quality Diary PRO endpoints | Study Endpoints review found the measures lacked content validity and could not adequately support labeling claims 118 |
Patterns in reviewer practice
- Hierarchy position decides the claim. When an earlier step fails, later endpoints lose inferential status. Reviewers have then presented them descriptively (Perjeta 187), opposed claims (Spiriva 115), or flagged them as nominal (Otezla 191, Calquence 290).
- Claims can be granted dose by dose. Invega shows FDA tying a secondary claim to the specific doses that succeeded on both the primary and secondary tests 186.
- Multiplicity control is necessary but not sufficient. Reviewers recommended leaving Zytiga's secondary endpoints out despite a Hochberg procedure because of concerns about endpoint validity and clinical interpretation 188. Nocdurna's PRO endpoints failed on content validity 118.
- Descriptive presentation without p-values is the usual fallback. Reviewers have offered it in place of outright deletion when the data have clinical value but no inferential support 116117286.
- Statistical reviewers do not always agree. The Spiriva review records a disagreement over whether the totality of evidence can override a single trial's testing hierarchy 115. This leaves some room for argument in programs with consistent replication.
Practical implications for sponsors
- Design for the claim you want. If a superiority claim against a named competitor is a commercial goal, prespecify a formal head-to-head test with an appropriately dosed comparator in a setting where its efficacy is established. Plan for replication 669091159.
- Do not count on non-inferiority to support a "comparable" claim. A similar-effectiveness claim needs a margin tighter than the effectiveness margin plus evidence that any difference is not clinically significant 6667.
- Put labeling-relevant secondaries in the alpha-controlled hierarchy. Anything outside it should be expected to appear descriptively, if at all 2730322323.
- Order the hierarchy with care. A failure early in the sequence removes inferential support for every endpoint below it, as Perjeta, Spiriva, and Calquence show 187115290.
- Validate PRO instruments before pivotal trials. Content validity is a gating requirement for PRO-based claims 326327118.
- Expect qualifiers. Even supported comparative claims may be limited to the doses studied and accompanied by confidence limits and language describing what was not shown 88241242.
Most review excerpts cited here capture reviewer recommendations made during labeling negotiation. They do not always show the final approved Section 14 text, so the approved labeling for each product is the reference for the final outcome.