Tirzepatide SURMOUNT-OSA Phase 3 Trials: NCT05412004

SURMOUNT-OSA was one master protocol containing two separate 52-week phase 3 trials in adults with obesity and moderate-to-severe obstructive sleep apnea. Trial 1 enrolled people not receiving positive-airway-pressure therapy at baseline; Trial 2 enrolled people receiving it. Tirzepatide produced larger average reductions in apnea–hypopnea index than placebo in both trials, alongside weight and other secondary changes. The trials should not be pooled into one effect estimate, and they do not show the same result for people without obesity, with diabetes, or outside the studied protocol. This page uses the corrected article, posted registry results, published erratum, and current product-specific FDA sources.[1][2][3][4][5][6][7]

Study at a glance

The protocol’s amounts, escalation, route, and schedule identify the tested intervention. They are not dosing, preparation, administration, self-use, or treatment guidance.[1]

Why the two trials must stay separate

Trial 1 asked about participants who were unable or unwilling to receive PAP. Trial 2 asked about participants already receiving PAP for at least three months who planned to continue it. For Trial 2 polysomnography and patient-reported assessments, PAP was suspended for seven days so measurements would not simply reflect the machine’s immediate effect. These designs address related but different clinical contexts.[1][2]

The randomized counts were 114 active and 120 placebo in Trial 1, and 120 active and 115 placebo in Trial 2. Reporting only the combined 469 obscures both allocation and PAP context. Likewise, combining outcomes would manufacture a single estimate that the protocol did not designate as the trial-specific primary result.[1][2]

Participants had severe OSA on average: baseline mean apnea–hypopnea index (AHI) was 51.5 events per hour in Trial 1 and 49.5 in Trial 2. Mean BMI was 39.1 and 38.7. The trials excluded diabetes and required obesity, so they do not directly answer how tirzepatide affects OSA in people without obesity or in the broader diabetes population.[1]

Participant flow and analysis populations

Of 469 randomized participants, 467 received at least one study exposure: 114, 120, 119, and 114 across the Trial 1 active, Trial 1 placebo, Trial 2 active, and Trial 2 placebo groups. Completion counts were 101, 86, 113, and 89, or 389 total. The paper reports 91.5% of tirzepatide recipients and 74.4% of placebo recipients completing the trials, while 87.6% and 71.9% adhered to the assigned regimen.[1][2]

The posted registry primary-outcome table analyzed 100, 86, 110, and 87 participants. Its modified intention-to-treat description requires randomization, at least one exposure, and evaluable outcome data. The paper’s intention-to-treat language also begins with people receiving at least one exposure. Therefore 469 randomized, 467 exposed for safety, 389 completing, and the endpoint-specific analyzed counts answer different questions. Differential completion also makes the prespecified handling of missing outcomes important.[1][2]

Primary endpoint and estimand

The primary endpoint was change in AHI at week 52 in each trial. AHI counts apnea and hypopnea events per hour of sleep; a reduction is not itself a direct measurement of cardiovascular events, daytime function, PAP adherence, hospitalization, or survival.[1]

The main treatment-regimen estimand asked about assignment effects regardless of intervention discontinuation. The analysis used prespecified assumptions for missing data: values considered missing at random were imputed from the assigned group, while a not-at-random approach used placebo-based imputation. A separate efficacy estimand assumed planned adherence. These are model-based answers to different questions; neither is a simple completer-only average.[1]

Trial-specific AHI results

In Trial 1, mean AHI change was −25.3 events/hour with tirzepatide (95% CI −29.3 to −21.2) and −5.3 with placebo (−9.4 to −1.1). The difference was −20.0 events/hour (95% CI −25.8 to −14.2; p<.001).[1]

In Trial 2, mean change was −29.3 events/hour (95% CI −33.2 to −25.4) and −5.5 (−9.9 to −1.2). The difference was −23.8 events/hour (95% CI −29.6 to −17.9; p<.001). Percentage-AHI differences were −47.7 points (95% CI −65.8 to −29.6) in Trial 1 and −56.2 (−73.7 to −38.7) in Trial 2.[1]

At least 50% AHI reduction occurred in 70/114 active versus 23/120 placebo participants in Trial 1, relative risk 3.3 (95% CI 2.1 to 5.1). In Trial 2 it occurred in 86/120 versus 27/115, relative risk 3.1 (2.1 to 4.5). The denominators shown here are randomized-group denominators used in the paper’s estimand presentation, not the smaller registry outcome-table counts.[1][2]

Multiplicity-controlled secondary outcomes

The prespecified hierarchy included percentage AHI change; at least 50% AHI reduction; a composite of AHI below 5 or AHI 5–14 with Epworth Sleepiness Scale no more than 10; body weight; sleep-apnea-specific hypoxic burden; high-sensitivity C-reactive protein; pooled PROMIS sleep-related impairment and sleep-disturbance scores; and systolic blood pressure.[1]

The composite occurred in 48/114 versus 19/120 in Trial 1, relative risk 2.9 (95% CI 1.8 to 4.8), and 60/120 versus 16/115 in Trial 2, relative risk 3.3 (2.0 to 5.4). Weight differences were −16.1 percentage points (95% CI −18.0 to −14.2) and −17.3 (−19.3 to −15.3). Hypoxic-burden differences were −70.1% (−90.9 to −49.3) and −61.3% (−84.7 to −37.9).[1]

Week-48 systolic-blood-pressure differences were −7.6 mm Hg (95% CI −10.5 to −4.8) and −3.7 (−6.8 to −0.7). Pooled PROMIS differences favored tirzepatide by −3.9 points (−5.7 to −2.2) for sleep-related impairment and −3.1 (−4.5 to −1.5) for sleep disturbance. The paper notes that clinically important change thresholds for those instruments in OSA were not established. Biomarker, physiologic, weight, and questionnaire outcomes should not be treated as interchangeable.[1]

Adverse events and discontinuation

At least one adverse event occurred in 91/114 (79.8%), 92/120 (76.7%), 99/119 (83.2%), and 83/114 (72.8%) exposed participants. Serious adverse events affected 9/114, 7/120, 7/119, and 12/114. Adverse events led to discontinuation in 5/114, 2/120, 4/119, and 8/114. No deaths were reported.[1]

Gastrointestinal events were most frequent and mostly mild to moderate. Diarrhea affected 30/114, 15/120, 26/119, and 10/114; nausea 29/114, 12/120, 26/119, and 6/114; vomiting 20/114, 5/120, 11/119, and 1/114. Two adjudicated acute-pancreatitis events occurred in tirzepatide recipients in Trial 2. The trials were not large or long enough to exclude uncommon or delayed harms.[1]

Correction, sponsor role, and FDA scope

The journal published an erratum on October 17, 2024, DOI 10.1056/NEJMx240005, and the accessible PMC article is marked corrected. An erratum corrects a publication; it is not a retraction. This page uses the corrected article and preserves the linked correction in its source chain.[1][3]

Eli Lilly designed the protocol. Sponsor employees performed statistical work, and the sponsor participated in data collection, management, analysis, interpretation, manuscript preparation, and the decision to publish. Employee-authors and other relationships are disclosed.[1]

On December 20, 2024, FDA approved the finished product Zepbound (tirzepatide) for moderate-to-severe OSA in adults with obesity, in combination with a reduced-calorie diet and increased physical activity. The most recent publicly available application-specific label, revised February 2026, carries that indication. Drugs@FDA also lists an August 26, 2026 efficacy-labeling supplement whose label was not available on the site at review.[4][5][7] This action applies to that finished product and population; it does not approve raw tirzepatide, every formulation, a blend, compounded material, or a third-party product.

Statistical significance, effect, uncertainty, and importance

• Statistical significance: both trial-specific primary AHI differences and the controlled secondary hierarchy favored tirzepatide under the prespecified analyses.
• Effect size: AHI differences were −20.0 and −23.8 events/hour, with separate confidence intervals.
• Uncertainty: missing-data assumptions, differential completion, 52-week follow-up, and selected obesity/OSA populations limit transfer.
• Clinical importance: sleep-disordered breathing and several intermediate outcomes improved; long-term cardiovascular outcomes, survival, and PAP adherence were not established.

Important limitations

Follow-up was 52 weeks. The trials selected adults with obesity and moderate-to-severe OSA, excluded diabetes, and were not cardiovascular-outcome studies. Trial 2 did not test improved PAP adherence, and its PAP suspension procedure matters when interpreting assessments. Attrition differed between assignments. The study was not powered to establish effects within every symptom subgroup, and clinically meaningful PROMIS thresholds in OSA were not established.[1][2]

What these trials do not prove

SURMOUNT-OSA does not prove a durable reduction in heart attack, stroke, hospitalization, or death; eliminate the need for individualized OSA evaluation; or establish benefit for people without obesity or with diabetes. The two trials cannot be collapsed into one universal effect. They also cannot authenticate another tirzepatide formulation, supplier, material, or lot.[1][2][4][5]

Frequently asked questions

Was SURMOUNT-OSA one trial or two?

It was one master protocol containing two separate randomized trials. Their populations, allocations, and outcome estimates must remain distinct.[1][2]

How did the PAP populations differ?

Trial 1 enrolled people unable or unwilling to use PAP. Trial 2 enrolled established PAP users who intended to continue it, with PAP suspended before specified assessments.[1]

What was the primary endpoint?

The primary endpoint was trial-specific change from baseline in AHI at week 52, not body weight, blood pressure, a questionnaire score, or a cardiovascular event.[1]

Why do 469, 467, 389, and smaller analyzed counts appear?

They represent randomized, exposed, completed, and endpoint-analyzed populations. Substituting one for another would misstate participant flow.[1][2]

Did the trials show that PAP was no longer needed?

No. Trial 2 was not a PAP-withdrawal or PAP-adherence trial, and treatment decisions cannot be inferred from its temporary assessment procedure.[1]

Was the paper retracted because it has an erratum?

No. A correction was published, and the PMC article is marked corrected. The source chain includes both the corrected article and erratum.[1][3]

What exactly did FDA approve?

FDA approved finished-product Zepbound for moderate-to-severe OSA in adults with obesity, with the labeled lifestyle context. That does not approve every tirzepatide material.[4][5][7]

Primary and official sources

1. Corrected primary full text, PMCID PMC11598664
2. ClinicalTrials.gov details and results: NCT05412004
3. Published erratum, PMID 39413392
4. FDA approval announcement
5. Most recent publicly available FDA Zepbound label
6. PubMed: PMID 38912654
7. Drugs@FDA overview: NDA 217806