Tirzepatide for the Treatment of Obstructive Sleep Apnea and Obesity
- Design
- Two phase 3, double-blind, randomized, controlled trials under one master protocol; parallel-arm, placebo-controlled, 52-week treatment duration; trial 1 (gpi1, not on pap) and trial 2 (gpi2, on pap). · 469 participants · Intermediate outcome
- Match to healthy normal-weight adults aged 55–75
- Different populationSevere obesity (mean BMI 39) with OSA; AHI improvement is expected from weight loss of this magnitude.
- Could weight loss explain it?
- LikelyWeight loss ~18-20% (full text); AHI, hs-CRP and BP improvements are consistent with weight-loss effects; not separated.
- Study tier
- Study tier 2Two placebo-controlled RCTs with a physiological primary endpoint; no clinical event outcomes.
- Assessment
- Version 3 · manual · Sep 13, 2026
Study facts come from the paper. Population match, weight-loss explanation and study tier are our judgments, made against the reference group of healthy normal-weight adults aged 55–75.
In adults with severe obesity and sleep apnoea, tirzepatide roughly halved the apnoea-hypopnoea index over a year, with reductions in hs-CRP and blood pressure. These changes track large weight loss and do not show a weight-independent effect.
01Findings
What the study reported
- Drugs
- Tirzepatide
- Dose
- 10 or 15 mg once weekly
- Route
- subcutaneous
- Treatment duration
- 52 weeks
- Comparator
- placebo
- Primary outcome
- Change in apnea-hypopnea index at week 52 (two trials)
- Effect
- AHI difference -20.0 (trial 1) and -23.8 (trial 2) events/hour; hs-CRP and systolic BP also reduced
- 95% confidence interval
- -25.8 to -14.2; -29.6 to -17.9
- P value
- <0.001
- Follow-up
- 52 weeks treatment; hsCRP assessed baseline to Week 52.
- Adverse events
- GI events most common, mostly mild to moderate.
- Limitations
- Auto-classified from abstract only; effect estimates, adverse events and limitations not extracted. Requires manual review.
Who was studied
- Condition
- Moderate-to-severe obstructive sleep apnea and obesity
- Baseline ahi trial1
- 51.5 events per hour
- Baseline ahi trial2
- 49.5 events per hour
- Mean bmi trial1
- 39.1
- Mean bmi trial2
- 38.7
- Diabetes status
- Excluded — participants with type 1 or type 2 diabetes were not eligible (not simply unreported).
Study quality details
- Study design
- Randomized controlled trial
- Sample size
- not extracted
- Randomization
- yes
- Blinding
- double-blind
- Comparator
- placebo
- Follow up duration
- 52 weeks
- Outcome type
- physiological surrogate (AHI)
- Replication
- not assessed (auto)
- Consistency with other evidence
- not assessed (auto)
- Population applicability
- INDIRECT
- Statistical precision
- CI reported: 95% confidence interval [CI], -29
- Funding conflicts
- partial
- Peer review status
- yes
- Rob2
- O1: {"D1":{"judgment":"Low","rationale":"Large industry-sponsored, multicenter phase 3 registration trial with centralized randomization/dispensing via an interactive web-response system (IWRS); per collection rule this supports allocation-concealment PY absent contradicting evidence, and nothing in the sources contradicts it. Applies identically to trial 1 and trial 2 (same master protocol/IWRS)."},"D2":{"judgment":"Some concerns","rationale":"Blinding is claimed (double-blind, participant+investigator masked) and GLP-1 GI effects can unblind participants, but per the collection rule D2 here is judged on discontinuation imbalance and ITT handling rather than unblinding per se. Discontinuation ('NOT COMPLETED') was markedly imbalanced by arm in both trials — trial 1: 34/120 (28.3%) placebo vs 13/114 (11.4%) tirzepatide; trial 2: 26/115 ("},"D3":{"judgment":"High","rationale":"The posted hsCRP denominators are well below the number randomized and are differential by arm in both trials. Trial 1: 95/114 (16.7% missing) tirzepatide vs 84/120 (30.0% missing) placebo — a 13.3-point difference; trial 2: 102/120 (15.0% missing) vs 82/115 (28.7% missing) — a 13.7-point difference (14.3% vs 28.1% against the dosed denominators). Both placebo arms exceed the guide's 20% threshold"},"D4":{"judgment":"Low","rationale":"hsCRP is an objective, central-laboratory blood assay ('Lilly-designated laboratory'), consistent with the collection rule that central-lab hsCRP measurement bias is Low absent an assay change or between-arm difference in measurement; no such change or differential measurement is reported in any source. Applies identically to both trials (same lab/assay under the master protocol)."},"D5":{"judgment":"Some concerns","rationale":"hsCRP is explicitly prespecified as a key secondary objective 'controlled for type I error' in both the protocol and SAP, tested via the closed graphical multiple-testing procedure (Bretz et al.) that strongly controls family-wise Type I error across the primary and key secondary objectives, and only one prespecified transformation/analysis (ANCOVA on log-hsCRP, reported as geometric LS means) is "},"overall":{"judgment":"High"},"result":"Change from baseline in high-sensitivity C-reactive protein (hsCRP) concentration at Week 52, tirzepatide (MTD 10/15 mg) vs placebo, key multiplicity-controlled secondary endpoint under master protocol I8F-MC-GPIF; reported separately for trial 1/GPI1 (not on PAP) and trial 2/GPI2 (on PAP) in the registry's posted results module (results_crp_measures), ANCOVA on log-hsCRP, mITT/treatment-regimen estimand population. No numeric results for this outcome appear in the abstract.","passes":[{"pass":"A","model":"claude-sonnet"},{"pass":"B","model":"claude-opus"}],"guide_version":"rob2-guide v1 + v1.1 calibration rulings (2026-09-13)","label":"Human-reviewed","resolution":"agreed domains accepted; disagreements decided by the owner 2026-09-13 (IN-009)"}
Methodological notes
Calibration two-pass assessment 2026-09-13 (drafts in data/assessments/38912654/); sources: abstract, registry, public protocol/SAP where available; no paper full text.
01bRisk of bias
How much the result can be relied on
Risk of bias is judged per result, not per paper: the same study can report one marker at low risk and another at high. Two reviewers assess each result independently against a written guide, and every domain judgment below carries its reasoning and the sentences it rests on.
These words are not quality scores. On the RoB 2 scale, High risk of bias is the worst rating a result can receive and Low risk of bias is the best — the opposite of how the same words read on some other scales. Levels are always written out in full here for that reason.
Assessed, not settled (1)
Recorded rather than omitted. Leaving these out would make the appraisal look cleaner than it is, and a disagreement between two careful readers is itself worth knowing.
O1
Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it
This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.
| Where they differ | The two readings |
|---|---|
| D3 · Missing outcome data | One reviewer: Some concerns about risk of bias The other: High risk of bias |
| Domain | Judgment and reasoning |
|---|---|
| D1 · Randomisation process Low risk of bias | Large industry-sponsored, multicenter phase 3 registration trial with centralized randomization/dispensing via an interactive web-response system (IWRS); per collection rule this supports allocation-concealment PY absent contradicting evidence, and nothing in the sources contradicts it. Applies identically to trial 1 and trial 2 (same master protocol/IWRS).Register visit with IWRS Provided centrally by the sponsor and dispensed via IWRS |
| D2 · Deviations from intended interventions Some concerns about risk of bias | Blinding is claimed (double-blind, participant+investigator masked) and GLP-1 GI effects can unblind participants, but per the collection rule D2 here is judged on discontinuation imbalance and ITT handling rather than unblinding per se. Discontinuation ('NOT COMPLETED') was markedly imbalanced by arm in both trials — trial 1: 34/120 (28.3%) placebo vs 13/114 (11.4%) tirzepatide; trial 2: 26/115 (22.6%) placebo vs 7/120 (5.8%) tirzepatide — mostly driven by 'Withdrawal by Subject' being far higher on placebo (19 vs 4 in trial 1; 14 vs 4 in trial 2), the reverse of the classic GI-AE-driven tirzepatide dropout pattern. Analyses are pre-specified to mITT/treatment-regimen or efficacy estimands with defined intercurrent-event handling, which mitigates but does not eliminate concern from this large differential dropout. NI on whether GI AE rates differed markedly by arm (no arm-level AE table in the available sources) — abstract only states GI AEs were the most frequent with tirzepatide, without a placebo comparator rate."type": "NOT COMPLETED" ... "groupId": "FG001", "numSubjects": "34" ... "groupId": "FG000", "numSubjects": "13" "type": "Withdrawal by Subject" ... "groupId": "FG001", "numSubjects": "19" ... "groupId": "FG000", "numSubjects": "4" |
| D3 · Missing outcome data Some concerns about risk of bias | hsCRP outcome data availability was well below all-randomized levels and differed materially by arm in both trials, so per the collection rule this is PN to 'data available for nearly all.' Trial 1: hsCRP evaluable n=95/114 (83.3%) tirzepatide vs 84/120 (70.0%) placebo (13.3-point gap; 23.5% missing overall). Trial 2: n=102/120 (85.0%) tirzepatide vs 82/115 (71.3%) placebo (13.7-point gap; 21.7% missing overall) — both trials exceed the 20%-overall / 5-point-differential thresholds. Per the collection rule, missingness that could depend on the true value defaults to PY unless addressed; here the reported hsCRP result comes from the 'treatment regimen' estimand, whose SAP-specified handling of missing data uses reason-based multiple imputation (including MNAR-oriented 'retrieved dropout' imputation for non-COVID/non-inadvertent discontinuations) rather than simple case exclusion, which partially addresses dependence but is itself model-based and unverifiable from the sources available. Also notable, and cutting against the usual GI-AE-driven mechanism: missingness is higher on placebo, not tirzepatide, in both trials. Applies to both trials with the same magnitude/direction of imbalance."groupId": "OG000", "value": "95" ... "groupId": "OG001", "value": "84" ... "groupId": "OG002", "value": "102" ... "groupId": "OG003", "value": "82" "groupId": "FG000", "numSubjects": "114" ... "groupId": "FG001", "numSubjects": "120" ... "groupId": "FG002", "numSubjects": "120" ... "groupId": "FG003", "numSubjects": "115" |
| D4 · Measurement of the outcome Low risk of bias | hsCRP is an objective, central-laboratory blood assay ('Lilly-designated laboratory'), consistent with the collection rule that central-lab hsCRP measurement bias is Low absent an assay change or between-arm difference in measurement; no such change or differential measurement is reported in any source. Applies identically to both trials (same lab/assay under the master protocol).hsCRP Results will not be provided to the investigative sites. Assayed by Lilly-designated laboratory. |
| D5 · Selection of the reported result Some concerns about risk of bias | hsCRP is explicitly prespecified as a key secondary objective 'controlled for type I error' in both the protocol and SAP, tested via the closed graphical multiple-testing procedure (Bretz et al.) that strongly controls family-wise Type I error across the primary and key secondary objectives, and only one prespecified transformation/analysis (ANCOVA on log-hsCRP, reported as geometric LS means) is posted per trial — so there is no evidence of cherry-picking among multiple CRP transformations. This supports Low on prespecification per se. However, Some concerns is assigned because (a) the exact position of hsCRP within the graphical alpha-propagation tree, and thus the nominal alpha it was actually tested at, is not verifiable from the available text (figure not reproduced in the text extraction), and (b) the registry's posted per-trial hsCRP analyses show an internal inconsistency between the printed p-value and the printed 95% CI/SE for the same comparison (trial 1: CI -1.21 to -0.22 excludes 0, yet pValue printed as 0.752; trial 2: CI -1.57 to -0.51 excludes 0, yet pValue printed as 0.350) — this cannot be resolved from the sources available and undermines confidence that the exact reported statistic matches the prespecified analysis without transcription/selection error. Applies to both trials (same pattern in both).Key Secondary Objectives (controlled for type I error) change from baseline to Week 52 in C-reactive protein (high-sensitivity C-reactive protein) |
02Funding
Funding and conflicts
- Funding
- Eli Lilly and Company
- Industry funded
- 1
- Manufacturer
- Eli Lilly and Company (tirzepatide/LY3298176 sponsor and drug supplier)
- Sponsor role
- Sponsor (Eli Lilly) is the responsible party; registry does not further break out sponsor's role in design/analysis/writing beyond this.
- Author conflicts
- Authors report Eli Lilly relationships.
- Independent replication
- no
- Notes
- Authors declare relationships with the drug's manufacturer: Eli Lilly
Funding is shown on every study and never used to score it.
03Claims
Claims this study bears on
- GLP-1 receptor agonist treatment reduces systemic inflammatory markers beyond what can be explained by treatment-associated weight loss.Mixed
SURMOUNT-OSA: hs-CRP fell with tirzepatide alongside large weight loss; not separated from weight.
04Source
The source, as retrieved
Abstract
[BACKGROUND] Obstructive sleep apnea is characterized by disordered breathing during sleep and is associated with major cardiovascular complications; excess adiposity is an etiologic risk factor. Tirzepatide may be a potential treatment. [METHODS] We conducted two phase 3, double-blind, randomized, controlled trials involving adults with moderate-to-severe obstructive sleep apnea and obesity. Participants who were not receiving treatment with positive airway pressure (PAP) at baseline were enrolled in trial 1, and those who were receiving PAP therapy at baseline were enrolled in trial 2. The participants were assigned in a 1:1 ratio to receive either the maximum tolerated dose of tirzepatide (10 mg or 15 mg) or placebo for 52 weeks. The primary end point was the change in the apnea-hypopnea index (AHI, the number of apneas and hypopneas during an hour of sleep) from baseline. Key multiplicity-controlled secondary end points included the percent change in AHI and body weight and changes in hypoxic burden, patient-reported sleep impairment and disturbance, high-sensitivity C-reactive protein (hsCRP) concentration, and systolic blood pressure. [RESULTS] At baseline, the mean AHI was 51.5 events per hour in trial 1 and 49.5 events per hour in trial 2, and the mean body-mass index (BMI, the weight in kilograms divided by the square of the height in meters) was 39.1 and 38.7, respectively. In trial 1, the mean change in AHI at week 52 was -25.3 events per hour (95% confidence interval [CI], -29.3 to -21.2) with tirzepatide and -5.3 events per hour (95% CI, -9.4 to -1.1) with placebo, for an estimated treatment difference of -20.0 events per hour (95% CI, -25.8 to -14.2) (P<0.001). In trial 2, the mean change in AHI at week 52 was -29.3 events per hour (95% CI, -33.2 to -25.4) with tirzepatide and -5.5 events per hour (95% CI, -9.9 to -1.2) with placebo, for an estimated treatment difference of -23.8 events per hour (95% CI, -29.6 to -17.9) (P<0.001). Significant improvements in the measurements for all prespecified key secondary end points were observed with tirzepatide as compared with placebo. The most frequently reported adverse events with tirzepatide were gastrointestinal in nature and mostly mild to moderate in severity. [CONCLUSIONS] Among persons with moderate-to-severe obstructive sleep apnea and obesity, tirzepatide reduced the AHI, body weight, hypoxic burden, hsCRP concentration, and systolic blood pressure and improved sleep-related patient-reported outcomes. (Funded by Eli Lilly; SURMOUNT-OSA ClinicalTrials.gov number, NCT05412004.).