The obvious way to test a weight drug in sleep apnea is to enroll people who are not being treated and watch the event count fall. SURMOUNT-OSA did that, and then did something harder alongside it: a second, simultaneous trial in people who were already using positive airway pressure and stayed on it. That decision is what makes the program worth reading as architecture rather than as a headline, because it is the difference between asking whether a drug works and asking where it fits. The event-count result itself is set out in the sleep apnea article; what follows is the machinery around it.
Two trials, run at the same time
Participants with moderate-to-severe obstructive sleep apnea and obesity were assigned to trial 1 if they were not receiving positive airway pressure at baseline and to trial 2 if they were, then randomized 1:1 within their trial to the maximum tolerated dose of tirzepatide, 10 mg or 15 mg, or to placebo for 52 weeks. The trials ran from 21 June 2022 through 29 March 2024.[1]
A total of 469 participants were randomized: 234 in trial 1 (tirzepatide 114, placebo 120) and 235 in trial 2 (tirzepatide 120, placebo 115). Dosing began at 2.5 mg weekly and rose by 2.5 mg every four weeks until the maximum tolerated dose was reached at week 20. Both arms received regular lifestyle counseling with a 500 kilocalorie per day deficit, and the apnea-hypopnea index was measured by laboratory polysomnography at screening, at week 20 and at week 52.[1]
Two sleep studies after baseline, scored centrally, is a heavier measurement burden than most drug trials carry, and it is why the assignment to trial 1 or trial 2 could not be randomized: nobody can be assigned to already own a machine. The two trials are therefore parallel cohorts rather than arms of one experiment, and a difference between them is an observation, not a comparison.
A cohort that looks nothing like the weight-loss trials
Women were 77 of 234 participants (32.9%) in trial 1 and 65 of 235 (27.7%) in trial 2. Mean age was 47.9 years (SD 11.5) and 51.7 years (11.0) respectively, and mean glycated hemoglobin sat near 5.65%, inside the non-diabetic range.[1]
A cohort roughly two-thirds men is the inverse of the obesity trials that produced the familiar percentages, where the tirzepatide pivotal trial and its semaglutide counterparts ran predominantly female.[2] That is not an accident of recruitment; it is the epidemiology of the condition. It does mean that any generalization that travels from a weight-loss cohort into this one is crossing a real difference in who was measured.
The endpoint that describes a patient rather than a signal
Beyond the change in the event count, the trials carried a multiplicity-controlled endpoint built to describe disease control: an apnea-hypopnea index below 5, or between 5 and 14 with an Epworth Sleepiness Scale score of 10 or less, at week 52. It was reached by 48 participants (42.2%) on tirzepatide in trial 1 against 19 (15.9%) on placebo, a relative risk of 2.9 (95% CI, 1.8 to 4.8), and by 60 (50.2%) against 16 (14.3%) in trial 2, a relative risk of 3.3 (95% CI, 2.0 to 5.4).[1]
Both ratios are large. Both absolute figures leave roughly half the treated group short of the definition. A reduction of at least 50% in events was more common — 70 (61.2%) against 23 (19.0%) in trial 1 and 86 (72.4%) against 27 (23.3%) in trial 2 — which is the gap between improving a condition and controlling it.[1]
In relative terms the index fell 50.7% in trial 1 and 58.7% in trial 2, against 3.0% and 2.5% on placebo.[1] Weight fell 17.7% and 19.6% against 1.6% and 2.3%, and the sleep apnea-specific hypoxic burden fell 95.2 and 103.0 %·min/hr against 25.1 and 41.7.[1]
A timing decision worth noticing
Systolic blood pressure was a key secondary endpoint, and it was measured at week 48 rather than week 52 in both trials. The reason is stated in the paper: suspending positive airway pressure therapy in trial 2 for the end-of-trial sleep study would otherwise have confounded the assessment.[1]
That is a small design note carrying a large implication. Withdrawing a machine for a single night was expected to move blood pressure enough to contaminate a 52-week endpoint, which is a reminder of how much work the machine is doing for the people in trial 2 who kept using it. Systolic pressure fell 9.5 mmHg on tirzepatide against 1.8 in trial 1 (difference −7.6; 95% CI, −10.5 to −4.8) and 7.6 against 3.9 in trial 2 (difference −3.7; 95% CI, −6.8 to −0.7).[1]
The dropout ran backwards
Overall 82.9% of participants completed the trials, but the two arms did not complete them equally: 91.5% in the tirzepatide groups against 74.4% in the placebo groups. Adherence to the assigned regimen followed the same split, at 87.6% against 71.9%.[1]
The safety table points the same way in one trial and not the other. In trial 1, adverse events ended treatment in 5 tirzepatide participants (4.4%) against 2 on placebo (1.7%) — the expected direction. In trial 2 the figures were 4 (3.4%) against 8 (7.0%), and serious adverse events were reported by 7 (5.9%) on tirzepatide against 12 (10.5%) on placebo.[1]
Gastrointestinal events were clearly attributable to the drug — diarrhea 26.3% and 21.8% against 12.5% and 8.8%, nausea 25.4% and 21.8% against 10.0% and 5.3%[1] — and the same profile appears in the side-effect article. What did not follow is the leaving. In a year-long trial with two overnight sleep studies and an untreated breathing disorder, the arm without the drug was the arm that left, which is a reason to read any placebo-controlled apnea trial's completion column before its efficacy column.
Serious adverse events across both trials affected 35 participants (7.5%), with similar percentages in the tirzepatide and placebo groups. There were two adjudication-confirmed cases of acute pancreatitis, both in the trial 2 tirzepatide group, no cases of medullary thyroid cancer, and five cases of severe or serious depressive disorder or suicidal ideation or behavior across both trials, two of them on tirzepatide.[1]
Feeling better and testing better came apart
Pooled across both trials, the PROMIS Sleep-related Impairment T score fell 7.5 points on tirzepatide against 3.6 on placebo, an estimated treatment difference of −3.9 (95% CI, −5.7 to −2.2), and the Sleep Disturbance T score fell 5.7 against 2.7, a difference of −3.1 (95% CI, −4.5 to −1.5), both P < 0.001.[1]
A 2026 post hoc analysis then grouped participants by how they described themselves at baseline. Among those who reported being sleepy, the least-squares mean difference from placebo on sleep-related impairment was −7.1; among those who did not, it was −1.6. Poor sleepers improved 5.8 points against placebo on sleep disturbance where good sleepers improved 1.2. Fatigued participants gained 8.5 points of SF-36v2 Vitality against placebo where non-fatigued participants gained 1.0.[3]
The objective measurements did not behave that way. The same analysis reports that changes in the apnea-hypopnea index, hypoxic burden, weight and remission were similar across every baseline symptom subgroup.[3] The physiology moved the same amount for everyone; only the people who felt bad to begin with felt the difference. A separate subgroup analysis found the index falling across every baseline stratum of age, sex, severity, body-mass index and neck circumference, with 68% to 79% of tirzepatide participants improving their severity category while 64% to 70% of placebo participants saw no clinically relevant change.[4]
For a buyer without daytime symptoms, that dissociation is the most decision-relevant number on the page, and it is the one least likely to appear beside a product.
What these trials do not establish
They ran for 52 weeks and stopped nobody's treatment, so they say nothing about what the index does when the drug does. The only randomized withdrawal evidence for this molecule concerns weight in people without apnea, and it is in the SURMOUNT-4 article.
Trial 2 added tirzepatide to positive airway pressure. No arm removed the machine, so nothing here supports replacing it, and the design note about week 48 is direct evidence that the trialists expected its removal to matter. Whether better sleep on the drug is downstream of a lower event count, of lost weight, or of something else again is not separable in these data, and the broader relationship is covered in the sleep article. Apnea is diagnosed by polysomnography, which no subscription service performs.
What was in the syringe
Every figure belongs to branded, FDA-approved tirzepatide at 10 or 15 mg, escalated under supervision, in participants whose breathing was measured in a laboratory three times. Most sellers on the tirzepatide board dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed — the distinction drawn in the compounded-versus-brand article. An indication earned by a reviewed product does not transfer to a vial that shares its molecule name, and how each number here was checked is described in the methodology.