Sleep apnea is associated in observational studies with elevated cardiovascular (CV) risk, and positive airway pressure (PAP — usually CPAP) is the established treatment for its symptoms. The decision this page serves is narrower and easily conflated with symptom relief: does treating the apnea with PAP reduce hard CV events or death, or only the daytime symptoms and the disease-severity marker? The pooled RCT answer is a clean split — symptoms and quality of life improve; CV events and mortality do not.
[EXTRACTED — the effect estimates below are the source's pooled figures]
The pooled RCT result — no CV or mortality benefit
Yu pooled 10 RCTs (9 CPAP, 1 adaptive servo-ventilation) of adults with obstructive or central sleep apnea, N = 7266 (mean age 60.9 y, 80.5% men, mean BMI 30.0), with 356 major adverse cardiovascular events (MACE) and 613 deaths recorded; PAP was compared with no treatment or sham, random-effects model. (Yu et al., 2017)
| Outcome | Pooled RR (95% CI) | Reading |
|---|---|---|
| MACE (CV death + nonfatal ACS + nonfatal stroke) | 0.77 (0.53-1.13), P=.19 | NS; CI spans meaningful benefit to modest harm |
| MACE + hospitalization for unstable angina | 0.92 (0.71-1.20) | NS |
| Acute coronary syndrome | 1.00 (0.65-1.55) | NS |
| Stroke | 0.90 (0.66-1.21) | NS |
| Heart failure | 1.03 (0.92-1.16) | NS |
| Cardiovascular death | 1.15 (0.88-1.50) | NS; point estimate favours harm |
| All-cause death | 1.13 (0.99-1.29), P=.08 | NS; trends toward harm |
The absolute risk differences were all near zero (MACE RD -0.01 [95% CI -0.03 to 0.01]; all-cause death RD 0.00 [-0.01 to 0.01]) — a large relative CI on a small event count, not a precise null. (Yu et al., 2017)
The authors’ bottom line:
«The use of PAP, compared with no treatment or sham, was not associated with reduced risks of cardiovascular outcomes or death for patients with sleep apnea. Although there are other benefits of treatment with PAP for sleep apnea, these findings do not support treatment with PAP with a goal of prevention of these outcomes.»
What PAP does buy — symptoms and quality of life, not the CV surrogates
The benefit is real but sits on a different outcome axis. Pooled intermediate outcomes:
- Daytime sleepiness (Epworth) improved: -1.92 points (95% CI -2.79 to -1.06), though highly heterogeneous (I2 = 91%).
- Disease-specific QoL (SAQLI) improved 0.51 (0.31 to 0.70); depression (HADS) -0.70 (-1.09 to -0.31); anxiety (HADS) -0.40 (-0.67 to -0.13); SF-36 mental component +1.73 (0.01 to 3.46).
- The cardiovascular intermediate markers did NOT move: no PAP effect on systolic BP (-0.20 mm Hg [-2.29 to 1.89]), diastolic BP, BMI, any lipid, fasting glucose, HbA1c, or EQ-5D.
This matters for the surrogate reasoning: prior reports that PAP «modest decreases in blood pressure» underwrote the expectation of a CV benefit, and PAP entered guidance partly on that basis — the 2014 AHA/ASA stroke guidelines advise PAP «be con- sidered for patients with acute ischemic stroke or transient ischemic attack.» (Yu et al., 2017)
But in these pooled RCTs the BP surrogate itself was null, so the mechanistic rationale did not transmit. The authors make the ladder explicit: the joint expectation that PAP would help hard outcomes rested on «the apparent beneficial effectsofPAPonintermediatebiomarkers», and
«The absence of any significant association of PAP with intermediate markers of vascular risk in the trials included in this overview may explain the null associations of PAP with hard vascular outcomes.»
The adherence caveat — no-effect for average use, insufficient for high adherence
The single most important interpretive hazard: RCT CPAP adherence is low (trial means ranged 1.4 to 6.6 h/night; the largest trial, SAVE/McEvoy, averaged 3.3 h), so the ITT null is an average-adherence null, not a high-adherence null. Three lines bear on whether adherence rescues a benefit, and they cut against a strong claim in either direction:
- A post hoc subgroup of the 4 trials achieving >=4 h/night showed MACE RR 0.58 (95% CI 0.34 to
0.99) — a CI that just excludes unity. But the authors immediately discount it:
«the absence of any comparable associa- tion for other outcomes, the post hoc nature of the subgroup analyses, and the multiple comparisons made also make chance a plausible explanation.»
- The meta-regression (a within-trial, randomization-preserving approach) found no association of adherence, follow-up length, or baseline apnea-hypopnea index with the per-trial RR (all P > .13) — «although statistical power to detect modest effects was low.»
- Cross-trial subgroups of <4 vs >=4 h/night showed no heterogeneity.
Evidence-state placement. For the realistic decision — prescribing PAP at achievable adherence to prevent CV events — this is a no-demonstrated-benefit result, reinforced by the null on the CV intermediate markers and the null meta-regression; a large CV benefit is unlikely. But it is not a tight no-effect: the MACE CI (0.53-1.13) does not exclude a clinically meaningful benefit, and whether high, sustained adherence would help remains genuinely insufficient-evidence, not refuted. The authors concede a well-tolerated future device «may be worth evaluat- ing in another large trial since it remains possible that strong adherence may produce benefits.» The honest reading holds no-benefit-at-average-adherence and insufficient-at-high-adherence together, rather than collapsing to a proven «PAP does not work.» (Yu et al., 2017)
Decision-change and the open gap
- Do not treat sleep apnea for the purpose of CV-event or mortality prevention. PAP is warranted for symptom relief and quality of life — which are patient-important outcomes in their own right — but the CV-prevention expectation is not supported by the pooled RCTs. The authors’ own advice: it is «reasonable to recom- mend PAP therapy for the improvement of symptoms in pa- tients with OSA but not for protection against vascular disease or death», and in the meantime to «emphasize the importance of proven therapies, such as blood pressure lowering, lipid lowering, and antiplatelet therapy» — i.e. treat the CV risk factors directly. (Yu et al., 2017)
- Two possibilities the trials could not resolve (transportability limits): (i) participants were «mostly those without excessive sleepiness», so a benefit restricted to severely symptomatic patients is untested and arguably untestable (withholding PAP from them is not randomizable); (ii) the observational apnea->CV association «may represent disease processes that cannot be ameliorated by PAP delivered at the average intensity achieved in these clinical trials.» (Yu et al., 2017)
- The upstream-lever gap — now partly cashed. If splinting the airway nightly does not reduce CV risk, the question becomes whether reducing apnea severity at its source — chiefly weight loss / adiposity, the dominant driver of obstructive sleep apnea — moves the apnea burden or the downstream markers differently. That is a distinct exposure Yu cannot speak to; the SURMOUNT-OSA tirzepatide RCTs now do -> the upstream vs downstream section below.
The upstream lever — tirzepatide reduces OSA severity and moves the CV surrogates CPAP left null
The SURMOUNT-OSA phase-3 RCTs (Malhotra 2024; two 52-week double-blind trials, N=469, in moderate-to-severe OSA WITH obesity, baseline mean AHI ~50; manufacturer-funded) test the opposite kind of lever from CPAP: not a device that splints the airway each night, but weekly tirzepatide (a dual GIP/GLP-1 agonist) driving ~16-17% weight loss — attacking the adiposity that causes the apnea. Trial 1 enrolled non-PAP users; trial 2 enrolled established PAP users (assessed after a 7-day PAP washout, so the AHI reflects true off-device breathing).
On the disease-severity surrogate it is a large effect. The AHI treatment difference vs placebo was -20.0 events/hr (95% CI -25.8 to -14.2) in trial 1 and -23.8 (-29.6 to -17.9) in trial 2 (both P<0.001) — a ~48-56% reduction, with up to 50.2% of tirzepatide participants reaching «fewer than 5 AHI events per hour or 5 to 14 AHI events per hour and an ESS of 10 or less, which is relevant because these thresholds for disease severity represent a level at which PAP therapy may not be recommended». Patient-reported sleep symptoms also improved (PROMIS Sleep-related Impairment -3.9 [-5.7 to -2.2]; Sleep Disturbance -3.1 [-4.5 to -1.5], pooled). (Malhotra et al., 2024)
The design contrast (why this is not the same exposure as CPAP). CPAP normalizes AHI only while worn and only at the achieved adherence (Yu’s trials averaged 1.4-6.6 h/night) — it treats the symptom mechanically and reverses the moment the device comes off. Tirzepatide reduces the AHI measured off-device by shrinking the causal driver (adiposity), so the severity reduction is not adherence-gated to a nightly device. The two therefore differ on exactly the support factor Yu identified as decisive (delivered dose / adherence). (Malhotra et al., 2024; inferred from Yu et al., 2017)
The load-bearing cross-source claim — the CV surrogate CPAP left null, the upstream lever moves. The one intermediate CV marker Yu’s pooled CPAP RCTs measured was systolic blood pressure, and it was null. Tirzepatide moved it. Same quantity, opposite result:
| Parameter | CPAP (Yu 2017, pooled RCTs) | Tirzepatide (Malhotra 2024) | Same quantity? |
|---|---|---|---|
| Intervention | airway-splinting device (downstream) | weight-loss drug (upstream) | No — different exposure |
| Population | OSA, mean BMI 30, «mostly without excessive sleepiness» | OSA + obesity, mean BMI ~39 | No — Malhotra heavier, all obese |
| Systolic BP, treatment diff vs control | -0.20 mm Hg (95% CI -2.29 to 1.89) — null | -7.6 (-10.5 to -4.8) T1 / -3.7 (-6.8 to -0.7) T2 | YES — same quantity, opposite verdict |
| hsCRP (inflammation) | not moved in pooled markers | -0.7 (-1.2 to -0.2) T1 / -1.0 (-1.6 to -0.5) T2 | partial — Yu reports no CRP effect |
| Hypoxic burden | not reported | -70.1 T1 / -61.3 T2 %min/hr | No — Yu did not measure it |
| Hard CV events / mortality | measured, null (MACE RR 0.77, 0.53-1.13; all-cause 1.13) | not assessed — 52 wk, no powered endpoint, no deaths | NO — Malhotra cannot speak to it |
The emergent reading — and the discipline that stops it becoming a claim (type-A + G-gap). Yu’s own mechanistic ladder was that «the absence of any significant association of PAP with intermediate markers of vascular risk … may explain the null associations of PAP with hard vascular outcomes». Malhotra restores those intermediate markers — BP, hsCRP, hypoxic burden all move — which raises the hypothesis that the upstream route could transmit to hard outcomes where the downstream route did not. But that hypothesis is explicitly untested, and the same surrogate-skepticism that discounted CPAP’s hoped-for benefit binds tirzepatide’s markers too. AHI, BP, hsCRP and hypoxic burden are all surrogates; moving them is not moving events. The authors say so themselves — «the design and shorter duration of the current trials does not support the assessment of long-term cardiovascular outcomes» — and point to the ongoing SURMOUNT-Morbidity and Mortality in Obesity trial (NCT05556512) as the test. Malhotra tellingly cites the CPAP null as its own rationale: «PAP has not been shown to affect cardiovascular complications and death in obstructive sleep apnea; therefore, there is a need for additional treatment options». So the honest composite: the upstream lever wins on OSA severity and on the CV surrogates, but the hard-CV/mortality question is unproven for BOTH interventions — open for CPAP (insufficient at high adherence) and not-yet-tested for tirzepatide (SURMOUNT-MMO pending). A surrogate advantage over CPAP is not a demonstrated outcome advantage. (Malhotra et al., 2024)
Decision-change. For a patient with moderate-to-severe OSA and obesity who cannot or will not adhere to PAP, tirzepatide is now an evidenced option that reduces apnea severity at its source and improves symptoms and CV risk markers — a genuine addition to the menu, and for many the AHI falls below the level at which PAP is even indicated. What it does NOT yet license is this will prevent your heart attack or death: that transmission is unproven, exactly as it is for CPAP. Weigh it on the outcomes demonstrated (severity, symptoms, weight, BP) plus its harm arm (GI adverse events; two adjudicated pancreatitis cases), not on an inferred hard-outcome benefit.
Synthesis — a worked surrogate-vs-outcome split
This is a clean instance of the Surrogate Outcomes hazard on a treatment rather than a biomarker: PAP corrects the symptom (sleepiness) and the disease it is prescribed for, yet the outcome people ultimately care about — CV events and survival — does not follow, and even the intermediate CV surrogate (blood pressure) was null in the pooled RCTs. Symptom benefit and outcome benefit are separate claims; here the first held and the second did not. It also exercises the guidance-null: a body (AHA/ASA) advises considering PAP for ischemic stroke/TIA — a cardiovascular indication the subsequent pooled RCT evidence does not support -> Which Objective Moved This Recommendation.