Long-term effects on clinical event, mental health, and related outcomes of CPAP for obstructive sleep apnea: a systematic review

Abstract

Study Objectives:

We performed a systematic review of long-term health outcomes of continuous positive airway pressure (CPAP) use in adults with obstructive sleep apnea.

Methods:

We updated prior systematic reviews with searches in multiple databases through January 3, 2023. We included randomized controlled trials (RCTs) and adjusted nonrandomized comparative studies that reported prespecified long-term (mostly > 1 year) health outcomes. We assessed risk of bias, conducted meta-analyses, and evaluated strength of evidence.

Results:

We found 38 eligible studies (16 trials, 22 observational). All conclusions were of low strength of evidence given study and data limitations. RCTs found no evidence of effect of CPAP on mortality (summary effect size [ES] 0.89; 95% confidence interval [CI] 0.66, 1.21); inclusion of adjusted nonrandomized comparative studies yields an association with reduced risk of death (ES 0.57; 95% CI 0.44, 0.73). RCTs found no evidence of effects of CPAP for cardiovascular death (ES 0.99; 95% CI 0.64, 1.53), stroke (ES 0.99; 95% CI 0.73, 1.35), myocardial infarction (ES 1.05; 95% CI 0.78, 1.41), incident atrial fibrillation (ES 0.89; 95% CI 0.48, 1.63), or composite cardiovascular outcomes (all statistically nonsignificant). RCTs found no evidence of effects for incident diabetes (ES 1.02; 95% CI 0.69, 1.51) or accidents (all nonsignificant) and no clinically significant effects on depressive symptoms, anxiety symptoms, or cognitive function.

Conclusions:

Whether CPAP use for obstructive sleep apnea affects long-term health outcomes remains largely unanswered. RCTs and nonrandomized comparative studies are inconsistent regarding the effect of CPAP on mortality. Current studies are underpowered, with relatively short duration follow-up and methodological limitations.

Citation:

Balk EM, Adam GP, Cao W, Bhuma MR, D’Ambrosio C, Trikalinos TA. Long-term effects on clinical event, mental health, and related outcomes of CPAP for obstructive sleep apnea: a systematic review. J Clin Sleep Med. 2024;20(6):895–909.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Although continuous positive airway pressure devices are commonly prescribed for adults with obstructive sleep apnea, the evidence does not support their effectiveness to reduce long-term health outcomes. Comparative observational studies suggest an association between continuous positive airway pressure use and long-term mortality but, with low strength of evidence. Randomized trials fail to support an effect of continuous positive airway pressure on death, numerous cardiovascular events, incident diabetes, or accidents, with no clinically important effect on cognitive function or mental health symptoms.

Study Impact: Published trials have been underpowered and of too short a duration to demonstrate effects and have other methodological limitations. We believe there remains clinical equipoise regarding the long-term effects of continuous positive airway pressure and that future well-conducted trials and rigorous analyses of observational studies are warranted.

INTRODUCTION

Obstructive sleep apnea (OSA) is a common disorder characterized by periods of airflow cessation (apnea) or reduced airflow (hypopnea) during sleep. The predominant first-line therapy for OSA is prescription of a continuous positive airway pressure (CPAP) device, or variants such as auto-adjusting or bilevel devices, for use during sleep. It is well known that CPAP is highly effective at improving sleep measures, including apnea-hypopnea index (AHI), oxygen desaturation index (ODI), and other surrogate and intermediate measures of OSA severity, as well as measures of sleepiness.¹ Nevertheless, questions remain about the effectiveness of CPAP to reduce or improve long-term health outcomes (eg, cardiovascular [CV] events, stroke, and mortality). To address this question, SAVE (Sleep Apnea cardioVascular Endpoints), a large randomized trial of long-term CPAP use, was designed to test the effectiveness of CPAP to reduce CV events in people with CV disease (CVD).^2–4 Despite improvements in AHI, SAVE found no improvement of CV and other clinical outcomes. Because CPAP effectively lowers AHI in clinical practice and across studies, the SAVE trial raises questions about whether, as has commonly been assumed, change in AHI is a valid intermediate or surrogate outcome for patient-centered health outcomes and whether long-term CPAP use prevents morbidities and mortality.

We conducted a Technology Assessment, nominated by the Centers for Medicare and Medicaid Services to the Agency for Healthcare Research and Quality (AHRQ), addressing the comparative effectiveness of CPAP specifically on long-term health outcomes and other related research questions.⁵ Here we present our systematic review (SR) of the relative effectiveness of CPAP vs no CPAP treatment on prespecified long-term health outcomes. The full Technology Assessment also addresses comparisons of CPAP with alternative treatments—for which there was limited long-term evidence—and the validity of changes in measures of sleep-disordered breathing (eg, AHI) as proxies for clinical outcomes—for which evidence is lacking.⁵ Our review does not evaluate sleep measures, such as AHI, or symptoms of OSA, such as sleepiness or snoring, for which CPAP is well known to be effective.¹

METHODS

We conducted this SR based on the AHRQ Methods Guide for Effectiveness and Comparative Effectiveness Reviews⁶ in accordance with the Preferred Items for Reporting in Systematic Reviews and Meta-Analyses (PRISMA)⁷ and A Measurement Tool to Assess Systematic Reviews (AMSTAR 2).⁸ The protocol was posted on June 16, 2020,⁹ and registered in PROSPERO on July 17, 2020 (CRD42020192725).

We included published, peer-reviewed randomized controlled trials (RCTs) and nonrandomized comparative studies (NRCSs) that used analytic methods to minimize confounding bias. Here we include comparisons of CPAP vs no CPAP. The full report also includes comparisons of different devices and CPAP vs other nonsurgical treatments.⁵ Study participants were adults diagnosed with OSA (per each study’s criteria), excluding other types of sleep apnea (including related to prior stroke or severe heart failure). Health outcomes of interest included all-cause death, CV death, CV and cerebrovascular events or incident diagnoses and composite CV outcomes, and incident diagnoses (or resolution) of hypertension, arrhythmias, diabetes, or prediabetes. For these outcomes we required a minimum 1-year follow-up. We also evaluated mental health conditions, cognitive function, quality of life, sexual function, missed workdays, and accidents, for which we required a minimum 6-month follow-up.

For literature published through 2010, we rescreened all studies that had been included in existing SRs on OSA diagnosis and treatment conducted for AHRQ.^1,10–12 For more recent articles, de novo literature searches were conducted in PubMed, Embase, Cochrane databases, CINAHL, ClinicalTrials.gov, and Epistemonikos for primary studies, existing SRs, and published guidelines from January 1, 2010, through January 3, 2023. Note that the full report is current only to March 22, 2021.⁵ We further collected studies suggested by technical experts and peer reviewers. The search strategies were peer-reviewed by an independent medical librarian and are provided in the supplemental material.

Identified citations were independently double-screened using Abstrackr by a team of seven researchers (the update since March 2021 was screened by 2 researchers); conflicts were resolved by group discussion. All potentially relevant studies were rescreened in full text in duplicate. We extracted study data into the Systematic Review Data Repository. Study- and outcome-level risk of bias (RoB) assessment was conducted during data extraction within the Systematic Review Data Repository. For RCTs, we used the Cochrane Risk of Bias 2.0 tool¹³; for NRCSs, we included assessments of specific elements from the Risk Of Bias In Nonrandomized Studies – of Interventions (ROBINS-I) tool,¹⁴ particularly related to selection bias (comparability of groups). Each study was rated as being at high, moderate, or low RoB. Because we downgraded RCTs for issues related to randomization and blinding, we determined that NRCSs could be at best at moderate RoB. We further downgraded NRCSs that did not use propensity score (or equivalent) analyses to control for inherent differences among participants who were prescribed or used CPAP or not. Inadequate propensity score analyses were also considered to increase the RoB.¹⁵

Where feasible and appropriate, we conducted random effects model pairwise meta-analyses using the restricted maximum likelihood approach. We separately meta-analyzed RCTs and NRCSs, in addition to combined meta-analyses. Where feasible, we also separately meta-analyzed intention-to-treat (ITT) and per-protocol (or adherer) analyses across study designs.

For each outcome, we assessed the strength of evidence using standard AHRQ methodology.^6,16 We considered the number of studies, the study limitations (ie, RoB), the directness of the evidence to research question, the consistency of study results, the precision of any estimates of effect, and other limitations (particularly sparseness of evidence). For each outcome with sufficient (ie, not insufficient) evidence, we summarized the overall conclusion based on a combination of the direction of effect (or association), the statistical significance, and the magnitude of the effect.

RESULTS

The literature searches (for all questions addressed by the full report) yielded 22,189 citations; in addition, we found and screened the references from 87 existing SRs (see Figure S1 ^{(1.6MB, pdf)} in the supplemental material). Ultimately 38 eligible studies compared CPAP prescription, use, or adherence with no (or rarely, sham) CPAP prescription, use, or adherence. The 16 RCTs included a total of 7,664 participants.^2,17–45 The 22 NRCSs with multivariable adjustments for outcomes of interest included a total of 268,033 participants,^46–68 among which 254,724 were included in 3 relatively large studies.^46,54,57 Among the NRCSs, 5 used propensity score matching.^{46,51,52,58,63} The remaining 17 NRCSs used traditional logistic regression multivariable analyses.

The studies are summarized in Table 1 and described in detail in Table S1 ^{(1.6MB, pdf)} , Table S2 ^{(1.6MB, pdf)} , Table S3 ^{(1.6MB, pdf)} , Table S4 ^{(1.6MB, pdf)} , Table S5 ^{(1.6MB, pdf)} , and Table S6 ^{(1.6MB, pdf)} , with RoB assessment in Table S7 ^{(1.6MB, pdf)} and Table S8 ^{(1.6MB, pdf)} . Study eligibility criteria were variable across studies. Among the 16 RCTs, 5 included adults with OSA regardless of other major comorbidities,^{18,22,31,34,35} 6 were restricted primarily to participants with CVD,^{2,17,26,28,29,38} 3 excluded patients with CVD,^24,30,45 and 2 were restricted to patients with type 2 diabetes.^27,44 One RCT was restricted to older adults (≥ 65 years)³⁵ and 8 RCTs excluded older adults (≥ 65–75 years).^{2,17,24,26,28,31,34,45} Five of the 17 NRCSs restricted their participant samples based on comorbidities, including percutaneous coronary interventions,⁶⁸ hypertension,⁶¹ or type 2 diabetes⁶⁶; 2 excluded patients with either CVD⁵³ or diabetes.⁴⁸ Five NRCSs were restricted to older adults (≥ 60–80 years).^{52–54,59,64}

Table 1.

Summary of design and reported outcomes for CPAP vs no CPAP studies.

Study	Design	RoB	Analysis	Total N	Duration (mo)	Population	Age Restriction	AHI Eligibility	ESS	Adherence	All-Cause Mortality	CV Mortality	Stroke/TIA	AMI	Angina	Revascularization	CHF	Atrial Fibrillation	Composite CV	Accident	Hypertension	Diabetes	Mental Health	Cognitive	QoL/Function	Sexual Function
A317	RCT	Moderatea	ITT	83	12	AFib	≤75	≥15	≤15	100% (12 mo)	.	.	.	.	.	.	.	X	.	.	.	.	.	.	X	.
Aarab 201718,19	RCT	Moderatea	ITT	37	6	All	None	≥5	≥10	83% noc (6 mo)	.	.	.	.	.	.	.	.	.	.	.	.	X	.	.	X
APPLES20–23	RCT (sham)	Low	ITT	1,098	6	All	None	>10	None	42% (6 mo)	.	.	.	.	.	.	.	.	.	.	.	.	.	X	X	.
Barbé 201224	RCT	Highb	ITT	723	48	No CVD	≤70	≥20	≤10	64% (48 mo)	X	X	X	X	.	.	X		X	.	X	.	.	.	.	.
BestAIR25,26	RCT (sham)	Moderatec	ITT	169	12	CVD	45–75	≥10	None	43% (6 mo)	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.
DIANA27	RCT	Moderatea	ITT	132	12	DM	≤80	≥10	≤18	71% (12 mo)	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.
Huang 201528	RCT	Moderatea	As-treated	83	36	CVD	45–75	≥15	None	86% (36 mo)	.	.	.	.	.	.	.	.	X	.	X	X	.	.	.	.
ISAACC29	RCT	Moderatea	ITT	1,255	40	CVD (ACS)	None	≥15	≤10	38% (40 mo)	X	X	X	X	X	X	X	X	P	.	.	X	.	.	X	.
Monasterio 200130	RCT	Moderatea	ITT	125	6	No CVD	None	10–30	Not severe	4.8 h/noc	.	.	.	.	.	.	.	.	.	.	.	.	.	X	X	.
MOSAIC31–33	RCT	Highd	ITT	188	60	All	45–75	>7.5 (ODI)	<10	38% (6 mo)	X	.	.	.	.	.	.	.	X	.	.	.	.	.	X	.
Pelletier-Fleury 200434	RCT	Highd	As-treated	171	6	All	≤70	NR	None	4.8-5.5 h/noc	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.	.
PREDICT35	RCT	Moderatea	ITT	231	12	All	≥65	>7.5 (ODI)	≥9	35% (12 mo)	.	.	X	X	X	.	.	X	X	X	.	.	X	X	X	.
RICCADSA36–39	RCT	Highe	ITT	244	57	CVD (revasc)	None	≥15	<10	60% (12 mo)	X	X	X	X	.	X	.	.	P	.	.	.	X	.	X	.
SAVE2,40–43	RCT	Moderatea	ITT	2687	44	CVD	45–75	≥30 (ODI ≥ 12)	<16	42% (44 mo)	X	X	X	X	X	X	X	X	P	X	.	X	X	.	X	.
Shaw 201644	RCT	Highf	ITT	302	6	DM	none	≥15 (ODI)	None	61% (6 mo)	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.
Wu 2016 (Yangzhou)45	RCT	Highg	ITT	136	6	No CVD or RF	30–65	≥15	None	NR	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.	.
ALASKA46	NRCS (ps)	Moderateh	Adh vs nonuse	176,014	36	All	None	NR	None	100% (pp)	X	.	.	.	.	.	X	.	.	.	X	.	.	.	.	.
Bjornsdottir 201547	NRCS	Highi	Adh vs nonuse	562	24	All	None	≥15	None	100% (pp)	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.
Botros 200948	NRCS	Highg	Adh vs nonusej	266	32	No DM	None	≥20	None	NR	.	.	.	.	.	.	.	.	.	.	.	X		.	.	.
Budweiser 201349	NRCS	Highg	Adh vs nonuse	83	36	All	None	NR	None	NR	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X
Campos-Rodriguez 200550	NRCS	Highk	Adh vs nonadh	749	48	All	None	≥10	None	variable	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Chang 202051	NRCS (ps)	Highg	ITT-likel	1,918	59	All	None	NR	None	NR	.	.	X											.	.	.
CPAGE-MODE52	NRCS (ps)	Highm	ITT-likel	236	47	All	≥70	15–30	None	NR	.	.	X	.	.	.	.	X	X	.	.	.	.	.	.	.
Crawford-Achour 201553	NRCS	Highg	ITT-likel	126	120	No CVD	≥65	>30	None	NR	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.	.
Dunietz 202154	NRCS	Highn	ITT-likel	53,321	36	All	≥65	NR	None	74% (36 mo)	.	.	.	.	.	.	.	.	.	.	.	.	.	X	.	.
Gervès-Pinquié 202255	NRCS	Highk	Adh vs nonadh	2,783	79	All	none	≥5	None	100% (pp)	X	.	.	.	.	.	.	.	X	.	.	.	.	.	.	.
Jara 201856	NRCS	Highg	Adh vs nonuse	182	12	All	None	≥5	None	100% (pp)	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.	X
Jennum 201557	NRCS	Highg	ITT-likel	25,389	36	All	None	NR	None	NR	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Lisan 201958	NRCS (ps)	Moderateh	ITT-likel	255	133	All	≥40	NR	None	NR	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
López-Padilla 201659	NRCS	Highk	Adh vs nonusej	155	53	All	≥80	≥20	None	100% (pp)	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Myllylä 201960	NRCS	Highg	Use vs nonuseo	2060	104	All	None	≥15	None	100% (pp)	.	.	.	.	.	.	.	.	X	.	.	.	.	.	.	.
Navarro-Soriano 202161	NRCS	Highk	Adh vs nonusej	163	58	HTN	None	≥15	None	100% (pp)	.	.	.	.	.	.	X	X	X	.	X	.	.	.	.	.
ONSLEEP62,63	NRCS (ps)	Moderate / Highp	ITT-likel	1274	79	All	None	>5	None	NR	X	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Ou 201564	NRCS	HighK	ITT-likel	124	60	All	≥60	≥20	None	67% (60 mo)	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Schipper 201765	NRCS	Highk	Adh vs nonadh	283	71	All	None	≥5	None	100% (pp)	.	.	.	.	.	.	.	.	X	.	.	.	.	.	.	.
Sheth 202166	NRCS	Highk	Adh vs nonadh	578	30	DM	None	NR	None	100% (pp)	.	.	X	.	.	.	.	.	X	.	.	.	.	.	.	.
Silveira 202267	NRCS	Highk	Adh vs nonuse	1,217	60	All	None	≥5	None	71% (60 mo)	X	.	.	.	.	.	.	.	.	.	.	.	.	.	.	.
Wu 2015 (Beijing)68	NRCS	Highk	Adh vs nonusej	295	60	CVD (revasc)	None	≥15	None	100% (pp)	.	.	.	.	.	X	.	.	X	.	.	.	.	.	.	.

Dots within cells indicate outcome not reported. ^aInadequate blinding. ^bFailure to account for baseline differences in AHI (and time with oxygen saturation < 90%) between groups and inadequate blinding. ^cDifferent follow-up times between groups, but double-blinded. ^dLack of blinding and high dropout rate. ^eLack of blinding and high crossover from no CPAP to CPAP. ^fUnblinded, poor allocation concealment, different withdrawal rates between groups. ^gInadequate reporting of methods and/or participants. ^hNonrandomized (unblinded) with propensity score analysis. ⁱParticipants were categorized based on their decision to use or forego CPAP (thus participant selection bias based on participant characteristics observed after the start of intervention). ^jAnd nonadherent users. ^kInadequate adjustment for differences between groups. ^lAs-prescribed (ie, prescribed vs not prescribed). ^mNonrandomized (unblinded) with propensity score analysis, but intervention poorly defined. ⁿPoorly defined interventions and results. ^oLong-term CPAP users vs those who discontinued CPAP. ^pAll-cause death moderate RoB: nonrandomized with propensity score matching; CV death high RoB: unadjusted for baseline imbalances. ACS = acute coronary syndrome, Adh = adherent, AHI = apnea-hypopnea index eligibility criteria, AMI = acute myocardial infarction, CHF = congestive heart failure, CV = cardiovascular, CVD = cardiovascular disease, DM = diabetes mellitus, EQ-5D = EuroQoL-5D scale, ESS = Epworth Sleepiness Scale eligibility criteria, FOSQ = Functional Outcomes of Sleep Questionnaire, h = hour(s), HTN = hypertension, ITT = intention-to-treat, MMSE = Mini-Mental State Exam, mo = months, noc = night(s), nonadh = nonadherent, NR = not reported, NRCS = nonrandomized comparative study, ODI = oxygen desaturation index, P = outcome for which study was powered, pp = per protocol, ps = adjusted with propensity score, QoL = quality of life, QWB = Quality of Well-Being, RCT = randomized controlled trial, RCT (sham) = RCT using sham device comparator, revasc = coronary revascularization, RF = cardiovascular risk factors, TIA = transient ischemic attack, X = outcome reported.

Most of the 16 RCTs included patients with at least “moderate” OSA (eg, AHI ≥ 15 events/h); 4 explicitly included patients with “milder” OSA (AHI ≥ 5 or 10 events/h),^18,22,26,27 1 excluded “severe” OSA (AHI > 30 events/h),³⁰ and 1 restricted to patients with “severe” OSA (ODI ≥ 12).² Among the 15 NRCSs that reported AHI eligibility criteria, 7 included patients with at least “moderate” OSA (AHI ≥ 15 or 20 events/h),^{47,48,59–61,64,68} 6 included patients with “milder” OSA,^{50,55,56,63,65,67} 1 excluded both milder and “severe” OSA (included AHI 15–30 events/h),⁵² and 1 included only patients with “severe” OSA (AHI ≥ 30 events/h).⁵³

Among all studies, 22 reported ITT (or as-prescribed) analyses, comparing participants who were prescribed CPAP with those not prescribed CPAP (or given a sham CPAP device); these included all but 2 of the 16 RCTs^28,34 and 8 of the 22 NRCSs.^{51–54,57,58,63,64} Six NRCSs compared adherent and nonadherent CPAP users (nonadherence was generally defined as < 4 hours/night for > 70% of nights).^{48,50,55,60,65,66} Eight NRCSs compared adherent users with nonusers (either due to not being prescribed CPAP, refusing or terminating CPAP, or nonadherence).^{45–47,49,56,59,61,67}

Adherence with CPAP use varied across studies. We were unable to discern any obvious patterns across studies regarding low vs high lack of adherence, including duration of the studies. Among studies that reported adherence and did not require adherence for inclusion, the median adherence level was 61% (at 6 months), ranging from 35 to 100% (at 6–60 months). The adherence levels in the 3 NRCSs that reported data were mostly higher (67–74%, at 36–60 months) than in the RCTs (where half the RCTs had adherence levels of 35–43%).

Table 2 and Table 3 summarize the evidence from RCTs and adjusted NRCSs regarding the effect of CPAP (vs no CPAP) on CV-related and other health outcomes, respectively. Descriptions of the findings for outcomes with insufficient evidence (transient ischemic attack, angina, revascularization, congestive heart failure, hypertension, functional status, sexual function, and missed workdays), mostly due to sparse data, are reported in the supplemental material.

Table 2.

Evidence profile for CPAP vs no CPAP: mortality and cardiovascular outcomes.

Outcome (Table/Figure)	Study Design	No. of Studies (Patients)	Risk of Bias	Consistency	Precision	Directness	Other	Overall Strength of Evidence	Conclusion Statements Summary ES (95% CI)
Death, all-cause (Table S9 (1.6MB, pdf) / Figure 1)	RCT (ITT)	5 (5,097)	Moderate	Consistent	Imprecise	Indirect	None	Low	No evidence of effect 0.89 (0.66, 1.21)
	adj NRCS (ITT)	4 (28,453)	Moderate	Consistent	Precise	Direct	None	Not evaluated	Association with reduced death 0.55 (0.42, 0.74)
	adj NRCS (adherent users)	5 (180,918)	High	Inconsistent	Precise	Indirect	None	Not evaluated	Association with reduced death 0.44 (0.26, 0.72)
	Overall	14 (213,057)	Moderate	Inconsistent	Precise	Direct	None	Low	CPAP may reduce death 0.57 (0.44, 0.73)
Death, cardiovascular (Table S10 (1.6MB, pdf) / Figure S2 (1.6MB, pdf) )	RCT	4 (4,909)	High	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.99 (0.64, 1.53)
	adj NRCS	1 (2,558)	High	N/A	Precise	Direct	Single study	Not evaluated	No conclusion
	Overall	5 (7,457)	High	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.79 (0.50, 1.26)
Stroke (Table S11 (1.6MB, pdf) / Figure S3 (1.6MB, pdf) )	RCT	5 (5,140)	Moderate	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.99 (0.73, 1.35)
	adj NRCS	3 (2,732)	High	Consistent	Imprecise	Direct	None	Not evaluated	No evidence of association 0.84 (0.54, 1.29)
	Overall	8 (7,872)	Moderate	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.94 (0.73, 1.20)
Transient ischemic attack (Table S11 (1.6MB, pdf) / Figure S4 (1.6MB, pdf) )	RCT	4 (4,891)	Moderate	Consistent	Highly imprecise	Direct	None	Insufficient	No conclusion
Myocardial infarction (Table S12 (1.6MB, pdf) / Figure S5 (1.6MB, pdf) )	RCT	5 (5,140)	Moderate	Consistent	Imprecise	Direct	None	Low	No evidence of effect 1.05 (0.78, 1.41)
Anginaa (Table S13 (1.6MB, pdf) )	RCT	3 (4,172)	Moderate	Consistent	Imprecise	Direct	Sparse a	Insufficient	No conclusion
Revascularization, coronary arteryb (Table S14 (1.6MB, pdf) )	RCT	2 (1,499)	Moderate	Consistent	Precise	Direct	Sparsec	Insufficient	No conclusion
	adj NRCS	1 (295)	Moderate	N/A	Precise	Direct	Single study	Insufficient	No conclusion
	Overall	2 (2,982)	High	Inconsistent d	Imprecise	Direct	None	Insufficient	No conclusion
Congestive heart failuree (Table S16 (1.6MB, pdf) )	RCT	3 (4,665)	Moderate	Consistent	Highly imprecise	Direct	Sparsee	Insufficient	No conclusion
	adj NRCS	2 (176,177)	High	Consistent	Highly imprecise	Direct	Sparsef	Not evaluated	No conclusion
	Overall	4 (4,828)	Moderate	Consistent	Highly imprecise	Direct	Sparse f	Insufficient	No conclusion
Atrial fibrillation (Table S16 (1.6MB, pdf) / Figure S6 (1.6MB, pdf) )	RCT	3 (4,173)	Moderate	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.89 (0.48, 1.63)
	adj NRCS	2 (399)	High	Consistent	Highly imprecise	Direct	Single study	Not evaluated	No conclusion
	Overall	5 (4,572)	Moderate	Consistent	Imprecise	Direct	None	Low	No evidence of effect 0.92 (0.63, 1.34)
Composite cardiovascular outcomes (Tables S17 (1.6MB, pdf) and S18 (1.6MB, pdf) )	RCT	7 (5,401)	High	Consistent	Imprecise	Indirectf	Clinical heterogeneity	Low	No evidence of effect ES range 0.42–1.10 (all NS)g
	adj NRCS	7 (6,356)	High	Inconsistent	Precise	Indirectf	Clinical heterogeneity	Not evaluated	CPAP associated with lower risk (in patients not restricted by CV history), but overall no evidence of effect in patients with CV risk ∼0.37h and 1.05
	Overall	12 (11,757)	High	Inconsistent	Imprecise	Indirectf	Clinical heterogeneity	Low	No evidence of effect

Evaluations of RCT evidence base are in bold font. Evaluations of all studies (RCTs and adjusted NRCSs together) are in italic font. For each outcome, cited tables and figures indicate where the study-level data can be viewed. ^aOne study reported on risk of incident angina and 2 studies reported on risk of hospitalization for unstable angina, providing an imprecise estimate. ^bAn additional RCT (SAVE) also evaluated risk of any major artery revascularization; as a unique study, it provided insufficient evidence. ^cEach study evaluated a similar, but distinct, outcome (revascularization in patients with acute coronary syndrome, repeat coronary revascularization in patients with prior revascularization). ^dInconsistent findings regarding repeat revascularization. ^eTwo studies reported on risk of incident congestive heart failure and 2 studies reported on risk of hospitalization for congestive heart failure. ^fFindings fully recapitulate single study from very large database. adj NRCS = adjusted nonrandomized comparative studies, CI = confidence interval, CPAP = continuous positive airway pressure (device), CV = cardiovascular, ES = (summary) effect size, N/A = not applicable, NS = not statistically significant, RCT = randomized controlled trials.

Table 3.

Evidence profile for CPAP vs no CPAP: other outcomes.

Outcome	Study Design	No. of Studies (Patients)	Risk of Bias	Consistency	Precision	Directness	Other	Overall Strength of Evidence	Conclusion Statements Summary ES (95% CI)
Hypertension (Table S19 (1.6MB, pdf) )	RCT	2 (423)	High/moderate	N/A	Precise	Direct	Sparse	Insufficient	No conclusion
	adj NRCS	2 (176,177)	High/moderate	N/A	Precise	Direct	Sparse	Not evaluated	No conclusion
	Overall	3 (176,600)	High	Inconsistent a	Precise	Direct	Sparse a	Insufficient	No conclusion
Diabetes (Table S20 (1.6MB, pdf) / Figure S7 (1.6MB, pdf) )	RCT	3 (3,639)	Moderate	Consistent	Imprecise	Direct	Clinical heterogeneity	Low	No evidence of an effect 1.02 (0.69, 1.51)
	adj NRCS	1 (266)	High	N/A	Precise	Direct	Sparse	Not evaluated	No conclusion
	Overall	3 (2,999)	Moderate	Consistent	Precise	Direct	Clinical heterogeneity	Low	No evidence of an effect 0.88 (0.57, 1.35)
Depression (Table S21 (1.6MB, pdf) / Figure S8 (1.6MB, pdf) )	RCT	4 (2,824)	Moderate	Consistent	Precise	Indirect	Clinical heterogeneity	Low	No clinically significant improvement in depression symptom score
Anxiety (Table S21 (1.6MB, pdf) / Figure S9 (1.6MB, pdf) )	RCT	4 (2,824)	Moderate	Consistent	Precise	Indirect	Clinical heterogeneity	Low	No clinically significant improvements in anxiety symptom scores
Cognitive function (Tables S22 (1.6MB, pdf) and 23 (1.6MB, pdf) / Figures S10 (1.6MB, pdf) and S11 (1.6MB, pdf) )	RCT	4 (628)b	Moderate	Inconsistente	Imprecise	Indirect	Clinical heterogeneity	Low	No evidence of a clinically significant effect on MMSE or TMT-A or Bc
	adj NRCS	1 (53,321) 1 (126)	High Moderate	N/A N/A	Precise Precise	Direct Indirect	Sparse Sparse	Not evaluated	No conclusions
	Overall	6 (54,075)	Moderate	Inconsistent d	Imprecise	Indirect	Clinical heterogeneity	Low	Same as RCT conclusions
QoL & functional status (Table S24 (1.6MB, pdf) / Figures S12 (1.6MB, pdf) and S13 (1.6MB, pdf) )	RCT	10 (6,460)e	Moderate–High	PCS: Inconsistent Others: Consistent	Precise	Indirect	None	Low	PCS, MCS, EuroQol-5D, and SAQLI: Small, not clinically significant improvementsf
	Adj NRCS	1 (214)	High	N/A	Precise	Indirect	Sparse	Not evaluated	No conclusion
	Overall	11 (6,874) c	Moderate-High	Consistent	Precise	Indirect	None	Low	Same as RCT conclusions
Sexual function (Supplemental material)	RCT	1 (43)	Moderate	N/A	Precise	Indirect	Sparse	Insufficient	No conclusion
	adj NRCS	2 (265)	High	Inconsistent	Precise	Indirect	None	Not evaluated	No conclusion
	Overall	3 (308)	High	Inconsistent	Precise	Indirect	None	Insufficient	No conclusion
Accidents (Table S25 (1.6MB, pdf) )	RCT	2 (2,917)	Moderate	Consistent	Imprecise	Direct	Clinical heterogeneity	Low	No evidence of an effect on driving accidents
Missed workdays (Supplemental material)	RCT	1 (2,687)	Moderate	N/A	Precise	Direct	Sparse	Insufficient	No conclusion

Evaluations of RCT evidence base are in bold font. Evaluations of all studies (RCTs and adjusted NRCSs together) are in italic font. For each outcome, cited tables and figures indicate where the study-level data can be viewed. ^aTwo studies (1 RCT, 1 NRCS) reported on incident hypertension but had extremely different rates of incident hypertension (∼50% vs ∼0.35%). One study each reported on resolution of hypertension and hypertensive crises. ^bA fifth RCT (n = 1,098) evaluated unique measures of cognitive function and thus provided only insufficient evidence. ^cInsufficient evidence for other measures of cognitive function, which were evaluated by only 1 study each. ^dAlthough all effects on Mini-Mental Status Examination (MMSE) and Trail Making Test B (TMT-B) were small and statistically nonsignificant, the net effect on MMSE favored CPAP in direction whereas the net effect on TMT-B favored no CPAP. ^eAcross all quality-of-life and functional status measures. ^fInsufficient evidence regarding functional status as measured by Functional Outcomes Sleep Questionnaire (FOSQ) in 1 RCT. adj NRCS = adjusted nonrandomized comparative studies, CI = confidence interval, ES = (summary) effect size, MCS = Short Form (SF) 36 Mental Component Summary, N/A = not applicable, PCS = Short Form (SF) 36 Physical Component Summary, QoL = quality of life, RCT = randomized controlled trials, SAQLI = Sleep Apnea Quality of Life Index.

All-cause mortality

Fourteen studies (5 RCTs and 9 NRCSs) reported on all-cause mortality across 2–11 years (see Table S9 ^{(1.6MB, pdf)} ).^{2,24,29,31,38,46,50,55,57–59,63,64,67} The five RCTs evaluated variable populations of patients from 2 to about 5 years of follow-up, including otherwise healthy, nonelderly adults with at least “moderate” OSA (AHI ≥ 20 events/h or ODI ≥ 7.5) in 2 RCTs,^24,31 known CV or cerebrovascular disease with “severe” OSA (ODI ≥ 12) in 1 RCT,² and recent CV events (acute coronary syndrome or coronary revascularization) with at least “moderate” OSA (AHI ≥ 15 events/h) in 2 RCTs.^29,38 Adherence varied across studies at about 40% in 3 RCTs^2,29,31 and about 60% in 2^24,38 (Table 1). All RCTs reported ITT analyses. Two trials were determined to be at moderate RoB^2,29; the other trials were determined to be at high RoB due to high dropout rate,³¹ high crossover rate,³⁸ and failure to account for baseline differences in sleep study measures.³⁸

Across RCTs, effect sizes (ESs) ranged from 0.33–2.6, all statistically nonsignificant. The summary odds ratio across the RCTs was nonsignificant at 0.89 (95% CI 0.66–1.21) (Figure 1). The ESs did not clearly correlate with adherence rates across the ITT analyses. SAVE and ISAACC (Impact of Sleep Apnea syndrome in the evolution of Acute Coronary syndrome) also reported propensity-score-matched analyses of adherent CPAP users, with a still-nonsignificant, but stronger, hazard ratio (HR) in SAVE (0.60; 95% CI 0.32–1.10) but a nearly identical HR in ISAACC (0.80; 95% CI 0.52–1.23).

Figure 1. Meta-analysis of CPAP vs no CPAP: all-cause mortality.

*CPAP use 1–6 hours per night vs < 1 hour per night. † Reported adjusted hazard ratio inverted to align with CPAP vs no CPAP. ‡ SAVE and ISAACC reported both ITT and CPAP adherent (users) analyses. The all ITT and the “as-prescribed” summary estimates include the pertinent analysis from each study. The overall analysis includes the ITT analyses. adj = adjusted, CI = confidence interval, CPAP = continuous positive airway pressure, F/up = follow-up, HR = adjusted hazard ratio, I² = measure of statistical heterogeneity ranging from 0% (none) to 100%, IDR = incidence density ratio, ISAACC = Impact of Sleep Apnea syndrome in the evolution of Acute Coronary syndrome, ITT = intention-to-treat, MOSAIC = Multicentre Obstructive Sleep Apnea Interventional Cardiovascular trial, NR = not reported, NRCS = nonrandomized comparative study, ONSLEEP = Okinawa Nakamura Sleep, OR = odds ratio, RCT = randomized controlled trial, RICCADSA = Randomized Intervention with Continuous Positive Airway Pressure in CAD and OSA trial, SAVE = Sleep Apnea cardioVascular Endpoints trial.

Four of the 9 NRCSs evaluated ITT-like analyses of people provided CPAP devices,^57,58,63,64 2 of which used propensity score analyses and were determined to be at moderate RoB.^58,63 The other 5 NRCSs compared adherent users with either nonadherent users^50,55 or nonusers^46,59,67; 1 of the latter used a propensity score analysis and was determined to be at moderate RoB.⁴⁶ The remaining NRCSs were at high RoB based on use of simple regression analyses, but also because of poor reporting and inadequate adjustment of potential confounders. Across studies, all-cause mortality was reported at a wide range of follow-up times from 3 to 11 years, with mostly longer-term follow-up than the RCTs (Table 1). The NRCSs mostly included a broad range of participants, regardless of AHI or ODI threshold or comorbidities; however, 2 were restricted to older adults (≥ 60 years or ≥ 80 years) with AHI ≥ 20 events/h.^59,64 A higher percentage of individuals in most of the NRCSs died than in the RCTs, particularly in the studies of older adults and with longest-term follow-up.⁵⁸

The 4 adjusted NRCSs that reported ITT-like analyses all found statistically significant associations between CPAP prescription and reduced risk of all-cause mortality.^57,58,63,64 Adjusted HRs ranged from 0.08 (in a small, poorly analyzed study) to 0.67 (in a large database analysis). The summary adjusted HR across the NRCSs was statistically significant at 0.55 (95% CI 0.42–0.74) (Figure 1).

The 5 as-treated (users vs nonusers) analyses also reported statistically significant associations between (adherent) CPAP use and lower risk of all-cause mortality (Figure 1).^50,55,59,67 The findings were generally consistent with the ITT-analysis NRCSs, but with wider confidence intervals.

The summary estimate across all studies (RCTs and NRCSs) was 0.57 (95% CI 0.44–0.73) (Figure 1). The NRCSs provided 75% of the weight of the meta-analysis, but the ESs of the RCTs and NRCSs were not significantly different from each other (P = .076). The ITT and as-treated analyses (across RCTs and NRCSs) also yielded generally similar summary associations (P = .27).

Cardiovascular mortality

Five studies (4 RCTs and 1 NRCS) reported incidence of CV mortality after about 4–7 years (see Table S10 ^{(1.6MB, pdf)} ).^{2,24,29,38,63} Two RCTs were at moderate RoB^2,29 and 2 were at high RoB.^24,38 Across trials, ESs of ITT analyses ranged from 0.41–3.08 (see Figure S2 ^{(1.6MB, pdf)} ), all statistically nonsignificant and mostly highly imprecise. The summary ES across the RCTs was imprecise at 0.99 (95% CI 0.64–1.53). A single, high RoB NRCS reported an HR of 0.54 (95% CI 0.28–1.03) over a mean of 6.5 years of follow-up.

Stroke

Five RCTs reported incidence of stroke after 1 or about 4 years (see Table S11 ^{(1.6MB, pdf)} ).^{2,24,29,35,38} Three adjusted NRCSs also reported stroke.^51,52,66 Three RCTs were at moderate RoB^2,29,35 and 2 RCTs and the 3 NRCSs were at high RoB.^{24,38,51,52,66} Across the 5 RCTs, ITT ESs ranged from 0.49–1.59, all statistically nonsignificant (see Figure S3 ^{(1.6MB, pdf)} ). The summary ES was 0.99 (95% CI 0.73–1.35). In a propensity-score-matched analysis of adherent CPAP users, SAVE reported a stronger ES that was just statistically significant, but the study did not report whether this analysis was significantly different from its primary ITT analysis. The 3 adjusted NRCSs each reported imprecise effect estimates of stroke (summary ES 0.84; 95% CI 0.54–1.29).

Acute myocardial infarction

Five RCTs reported incidence of acute myocardial infarction after 1 or about 4 years (see Table S12 ^{(1.6MB, pdf)} ).^{2,24,29,35,38} Three RCTs were at moderate RoB^2,29,35 and 2 at high RoB.^24,38 There was a wide range of ES estimates from 0.25–7.38 (see Figure S5 ^{(1.6MB, pdf)} ), which were all imprecise and statistically nonsignificant. The summary ES across the RCTs was 1.05 (95% CI 0.78–1.41).

Atrial fibrillation

Three RCTs and 2 NRCSs reported incident atrial fibrillation between 1 and 4.8 years (see Table S16 ^{(1.6MB, pdf)} ).^{2,29,35,52,61} Study patients either had CVD or CV risk factors or were of older age. All studies were at moderate RoB. The RCTs all found imprecise estimates of ES, ranging from 0.48–1.47. Meta-analysis was also imprecise effect (ES 0.89, 95% CI 0.48–1.63) (see Figure S6 ^{(1.6MB, pdf)} ). The NRCSs (1 ITT and 1 as-treated) had similar results.^52,61

Composite cardiovascular outcomes

Fourteen studies (7 RCTs and 7 NRCSs) reported various composite CV outcomes.^{2,24,28,29,31,32,35,38,52,55,60,61,65,66,68} The 7 RCTs evaluated 11 unique (but overlapping) composite CV outcomes across 1 to about 5 years of follow-up (see Table S17 ^{(1.6MB, pdf)} ). Participant eligibility criteria varied across the 7 RCTs including studies restricted to patients with known CVD and studies of generally healthy patients (other than OSA). Four RCTs were at moderate RoB^2,28,29,35 and 3 at high RoB.^24,31,32,38

Summarization across the RCTs is hampered by the variable definitions of composite outcomes (ie, which CV and related events were included) and the variability in the underlying risk of CV events and death, based on prior history of CVD and age. Meta-analysis was determined to not be appropriate. With the exception of 1 small adherer analysis RCT,²⁸ the estimates of the ITT effect of CPAP on composite CV events ranged from an odds ratio of 0.42 to an HR of 1.10. Three ITT RCTs also reported sensitivity analyses among CPAP adherers,²⁸ which together with the 1 small adherer analysis RCT²⁸ all found stronger ESs in the CPAP adherer analyses than in the ITT analyses, but the differences between analyses varied across studies. Two RCTs found no significant difference between ITT and adherers analyses.^2,24 One trial reported a large shift in HR between the ITT analysis (HR 0.62, 95% CI 0.34–1.13) and the adherent vs nonadherent and nonusers analysis (HR 0.29, 95% CI 0.10–0.86).³⁸ Across all studies, a rough comparison between ITT and adherent/as-treated analyses among the RCTs found no significant difference in effect between analyses (P = .53).

Seven NRCSs evaluated 8 distinct (but overlapping) composite CV outcomes (also distinct from the RCTs) at between 2.5 and 9 years of follow-up (see Table S18 ^{(1.6MB, pdf)} ).^{52,55,60,61,65,66,68} One NRCS conducted an ITT-like analysis⁵²; the rest evaluated adherer analyses. All were at high RoB. The NRCSs found adjusted HRs that mostly favored CPAP use, ranging from 0.37–1.05. The 3 NRCSs conducted in patients with preexisting CVD⁶⁸ or CV risk factors^61,66 were not statistically significant, whereas the 4 NRCSs that did not restrict eligibility based on prior CVD history mostly were.^60,65

Diabetes mellitus

Three RCTs and 1 NRCS evaluated incident type 2 diabetes (see Table S20 ^{(1.6MB, pdf)} ).^2,28,29,48 No study reported on reversion to normoglycemia or incidence of other hyperglycemia diagnoses (eg, impaired glucose tolerance). Across the RCTs (all moderate RoB), CPAP did not affect the risk of incident diabetes at about 3–4 years (summary ES 1.02, 95% CI 0.69–1.51) (see Figure S7 ^{(1.6MB, pdf)} ). In contrast, in an adherers analysis, the high-RoB NRCS found a significantly lower association between regular CPAP use and risk of incident diabetes at about 3 years of follow-up.⁴⁸ With addition of the adjusted NRCS to the meta-analysis, the summary ES remained nonsignificant and imprecise (ES 0.88, 95% CI 0.57–1.35).

Mental health

Four RCTs reported on the effect of CPAP vs no CPAP on both depression and anxiety symptoms.^{18,35–37,43} No study addressed diagnoses (or resolution) of any mental health conditions. Three trials were at moderate RoB^18,35,43 and 1 at high RoB.^36,37 The trials evaluated different depression and anxiety symptom scales (see Table S21 ^{(1.6MB, pdf)} ). Overall, depression and anxiety scores improved over time in all studies among those using CPAP or no CPAP. Only the SAVE trial found statistically significant relative improvements in depression and anxiety scores (after 4 years) among those receiving CPAP compared with controls. Meta-analyses of the standardized mean differences across trials yielded a statistically significant summary standardized mean difference of −0.18 (95% CI −0.26 to −0.11) for depression scores (see Figure S8 ^{(1.6MB, pdf)} ) and −0.10 (95% CI −0.17 to −0.03) for anxiety scores (see Figure S9 ^{(1.6MB, pdf)} ). Both of these differences may be interpreted as small, likely not clinically significant.

Cognitive (executive) function

Seven studies (5 RCTs^{22,30,34,35,45} and 2 NRCSs^53,54) reported on cognitive executive function. One RCT was at low RoB,²² 2 RCTs were at moderate RoB,^30,35 and 2 RCTs and both NRCSs were at high RoB.^34,45,53 The studies reported numerous specific measures of cognitive function, most of which were reported by a single study each. One NRCS evaluated risk of incident dementia in Medicare enrollees ≥ 65 years of age.⁵⁴ In both ITT and adherent users analyses, statistically significant fewer CPAP users developed dementia (ITT analysis: adjusted odds ratio 0.78, 95% CI 0.69–0.89) (see Table S22 ^{(1.6MB, pdf)} ).

Three studies evaluated the Mini-Mental State Examination^35,45,53 and 3 studies the Trail Making Test B.^30,35,53 Mini-Mental State Examination scores improved in all study groups (with or without CPAP) in the RCTs, on average by a small, nonclinically significant degree (< 0.4 points) at 6 or 12 months (see Table S23 ^{(1.6MB, pdf)} ). Differences between interventions were nonsignificant and small (≤ 0.1 points). The NRCS was imprecise (see Figure S10 ^{(1.6MB, pdf)} ). Similarly, Trail Making Test B scores improved in all study groups in the RCTs, but Trail Making Test B scores worsened after 10 years in the NRCS. There were no differences between groups (see Figure S11 ^{(1.6MB, pdf)} ).

Quality of life

Ten RCTs^{2,17,20,21,26,27,29,31,35,39,44} and 1 NRCS⁴⁷ reported on quality of life based on a variety of measures (see Table S24 ^{(1.6MB, pdf)} ). The NRCS evaluated adherent vs nonadherent users.⁴⁷ One RCT (that used a sham CPAP control) was at low RoB,^20,21 6 RCTs were at moderate RoB,^{2,17,26,27,29,35} and 3 RCTs and the NRCS were at high RoB for a variety of reasons.^31,39,44,47

The studies found no clinically significant difference between groups on the Short Form-36, assuming a minimum clinical important difference of about 4 to 7 points.^69,70 Meta-analysis of the six RCTs that reported Short Form-36 (or Short Form-12) Physical Component Scores (after 6 months–3.7 years) yielded a summary net difference of 1.5 points (95% CI −0.1 to 3.2), favoring CPAP (see Figure S12 ^{(1.6MB, pdf)} ).^{2,17,26,27,31,39} The NRCS found a similar net difference of 1.6 points (95% CI −0.1 to 3.3) after 2 years (using Short Form-12).⁴⁷ Meta-analysis of the 7 RCTs that reported mean Short Form-36 Mental Component Scores found a summary net difference of 1.2 points (95% CI 0.7–1.7), favoring CPAP (see Figure S13 ^{(1.6MB, pdf)} ).^{2,17,26,27,31,39,44} The NRCS found no significant difference (−0.2; 95% CI −1.9 to 1.5).⁴⁷ Studies also did not find clinically significant differences (and mostly not statistically significant) on the EuroQol-5D, Quality of Well-Being, and Sleep Apnea Quality of Life Index scales.^{2,20,21,27,29,31}

Accidents

Two moderate-RoB RCTs reported on various accident outcomes. Because the likelihood of accidents is probably associated with degree of sleepiness, it is worth noting that the SAVE trial excluded patients with extreme daytime sleepiness (Epworth Sleepiness Scale > 15), and although PREDICT included only patients with excessive daytime sleepiness (Epworth Sleepiness Scale ≥ 9), professional drivers and people reporting sleepiness while driving were excluded. For numerous specific outcomes, the PREDICT trial³⁵ at 1-year follow-up and the SAVE trial² at 3.7-year follow-up found no statistically significant differences between groups (see Table S25 ^{(1.6MB, pdf)} ).

DISCUSSION

The primary clinical question addressed by this SR—the effectiveness of CPAP devices on long-term clinically important outcomes in patients with OSA—remains largely unanswered. Much of the evidence base is sparse and many of the studies provided largely imprecise estimates of comparative effectiveness, resulting in conclusions of generally low strength of evidence. Thus, with low strength of evidence, RCTs do not demonstrate that CPAP affects all-cause mortality or various CV outcomes including composite CV events, diabetes diagnoses, accidents, or clinically significant effects on various mental health and related outcomes. However, these conclusions do not imply that CPAP has been proven to be ineffective to reduce adverse health outcomes. It is unclear whether the failure to find an effect of CPAP treatment on long-term health outcomes is related to a lack of power, insufficient follow-up duration, or due to an actual lack of effect of CPAP.

Notably, in contrast with the RCTs, the NRCSs found a statistically significant association between CPAP use (or prescription) and decreased mortality. The NRCSs differed from the RCTs in several important ways, in addition to study design, including longer duration of follow-up (3–11 years vs 2 to about 5 years) and higher event rates (despite the RCTs mostly including only those at high risk for CVD). However, the NRCSs were mostly at high RoB related to conducting simple, often incomplete, regression analyses, which substantially reduces the certainty that the findings of the NRCSs are valid. A combined analysis of the RCTs and NRCSs provides (low-strength) evidence that CPAP reduces all-cause mortality. This conclusion, though, is based largely on the high-RoB NRCS evidence base, despite their inconsistency in magnitude and statistical significance with the RCTs.

For most categorical outcomes, ESs appear to be somewhat stronger in the analyses of CPAP-adherent user than in the ITT analyses (of all whom were prescribed CPAP), but almost universally the differences between analyses were not statistically significant (both within and across studies), and the differences were generally small. This may suggest that CPAP is ineffective, because its “effect” is equivalent among users and nonusers, but the comparisons between ITT and adherent-user analyses suffer from a lack of power to detect a difference. Further complicating adherent-user analyses is that patients who choose to be adherent are intrinsically different from those who are nonadherent (or stop using CPAP). Prior analyses have found that patients with worse baseline AHI or Epworth Sleepiness Scale scores are more likely to be adherent¹; thus, to some degree comparisons of users with nonusers may be comparisons of more vs less severe or symptomatic OSA.

The American Academy of Sleep Medicine recently commissioned a broad SR of CPAP vs no CPAP⁷¹ to support its clinical practice guideline.⁷² The review included a broader range of studies and outcomes than our review, including lower-quality evidence such as unadjusted observational studies, short-term outcomes, and various intermediate and symptom measures. Recapitulating prior reviews,^1,73 they found that CPAP reduces AHI and improves sleepiness symptoms. In addition, the review found that CPAP improves (mostly short-term) sleep-related, but not overall, quality of life; reduces (mostly short-term) blood pressure (in patients with hypertension); does not reduce (mostly short-term) blood glucose (in patients with or without type 2 diabetes); and does not improve neurocognitive function or improve (short-term) anxiety or depression scores by clinically significant degrees. They reached no conclusions regarding all-cause mortality or composite cardiovascular events. They ascribed differences in effects between randomized and observational studies to average differences in CPAP adherence, but the impact of unadjusted confounders was not considered.

For most health outcomes, an insufficient number of studies and an insufficient number of study participants have been evaluated to allow for precise estimates of effects. Few RCTs were powered for specific health outcomes. The emphasis on composite CV outcomes in most eligible studies hampers a clear understanding of the effect of CPAP on specific health outcomes of interest to patients and clinicians. Most RCTs were determined to be at moderate risk of bias, where the primary concern was a lack of blinding of participants or their clinicians. Acknowledging the difficulties of maintaining sham CPAP, the lack of blinding raises the possibility of differences in how patients acted or were treated based on their CPAP use. Although we restricted inclusion of NRCSs to those with multivariable analyses, there remain concerns that patients selected to be treated with CPAP (or self-selected to use or comply with CPAP) are inherently different from patients not treated with (or not adherent with) CPAP to such a degree that even well-adjusted analyses are biased toward CPAP users. Only 5 NRCSs used propensity score analyses (that adjust for the factors that predict for choice to use CPAP). Nevertheless, for most outcomes, ESs from NRCSs tended to be consistent with those from RCTs, although, particularly for all-cause mortality and, to a lesser degree, composite CV outcomes, NRCSs generally had somewhat stronger ESs that were more likely to be statistically significant than RCTs.

OSA is a heterogenous disease with a clinical diagnosis based on variably defined and applied criteria. Although data are still emerging on phenotypes and endotypes of OSA and their effect on outcomes, the variability we found in AHI among studies and lack of long-term benefit of CPAP may indicate that patients’ phenotypes or endotypes may contribute to the decision to treat, type of treatment, and, possibly, long-term benefits of treatment. Although we could not discern consistent differences across studies based on their eligibility criteria, studies used highly variable inclusion criteria (eg, based on AHI thresholds) and exclusion criteria (eg, based on comorbidities). In addition, even where OSA definitions, such as AHI threshold, were consistent, as we describe in a separate article,⁷⁴ the criteria used to define sleep apnea indices (apnea, hypopnea, and oxygen desaturation) were highly variable across studies and studies did not appear to apply standard sleep study criteria consistently. In addition, while AHI or ODI are used to diagnose OSA and to assess its severity, these measures do not include the severity of arousals, autonomic imbalance, sleep-stage fragmentation, or other consequences of sleep apnea. The frequency of apnea (as measured by AHI) may be a less important predictor of cardiovascular mortality than the severity of hypoxemia as measured by the hypoxic burden.⁷⁵ No studies of long-term outcomes with CPAP treatment reported on other measures of OSA, including cumulative exposure to reduced air intake during a night of sleep, duration of respiratory events (beyond 10 seconds), the total duration of respiratory events weighted by the oxygen desaturation that accompanies them,⁷⁶ hypoxic burden,⁷⁵ or the morphology of the oxygen desaturation events,⁷⁷ among others. Patients who have similar AHI can differ greatly in these other metrics,^76,77 and AHI is generally poorly correlated with symptoms and signs of OSA.⁷⁸ However, it is an open question whether definitions of OSA (and management of CPAP) based on alternative metrics would be more strongly correlated with long-term outcomes.

Several types of new studies and new analyses of existing studies are needed before stronger evidence is available to allow better clinical and policy decision-making. More in-depth future research suggestions are included in the full report.⁵ Ideally, large long-term RCTs are needed to assess the effectiveness of CPAP with respect to long-term health outcomes. Future RCTs should be adequately powered to find relatively small differences in effect (eg, odds ratio = 0.89, the summary estimate for death among RCTs), preferably for specific important health outcomes (such as all-cause or cardiovascular death), rather than nonstandardized, composite clinical event outcomes that are difficult to interpret. Studies should be continued for a sufficient duration to allow for any real differences to become apparent (probably more than the current maximum duration of about 4–5 years).

Given the large effect of CPAP on sleep and breathing measures (eg, AHI) and on symptoms including sleepiness and snoring, clinicians, patients, and institutional review boards are increasingly skeptical about the concept of randomizing people with “severe” OSA to long-term CPAP or no (or sham) CPAP, despite the low strength of evidence that CPAP affects long-term health outcomes (mostly with no effect). However, our finding supports the conclusion of a recent panel discussion at the 2021 American Academy of Sleep Medicine meeting that highlighted that there is clinical equipoise regarding CPAP use for CV and other outcomes, even for patients with excessive daytime sleepiness (which, it was noted, may improve with simple advice and training, regardless of OSA treatment).⁷⁹ Pertinent suggestions were offered to make future RCTs more efficient and effective, including use of an adaptive study design, in which study eligibility criteria and analysis plans adapt to information accrued during the trial and interim analyses.^80,81 Future RCTs are, thus, feasible, necessary, and appropriate to determine the effect of CPAP on long-term health outcomes. In addition, as a complement to RCT evidence, careful observational studies can inform the evidence base and may enable the RCTs to be maximally informative. Priority should be given to evaluating observational data using state-of-the-science causally explicit analyses, such as propensity score analysis methods to adequately adjust for confounders and minimize bias.^82,83

In conclusion, although prior reviews have substantiated benefits of CPAP on symptoms such as sleepiness, along with intermediate (eg, blood pressure) and short-term sleep-related quality of life,^1,73 RCTs have not found significant effects of CPAP on long-term health outcomes for adults with OSA, including all-cause mortality, various CV events, or incident diabetes or clinically significant effects on mental health measures (all with low strength of evidence). The effect of CPAP on other long-term health outcomes from RCTs is unclear due to insufficient evidence related to sparse studies and/or highly imprecise estimates. Inclusion of evidence from adjusted NRCSs mostly does not alter these conclusions, except that when considering NRCS evidence there is low strength of evidence that CPAP reduces the risk of all-cause death with longer-term follow-up. Given the low strength of evidence across outcomes, future needed studies may substantially alter these conclusions.

DISCLOSURE STATEMENT

The authors report no conflicts of interest. Carolyn D’Ambrosio is an advisor for Hicuity Health Inc., a triage leader for Dynamed Inc., and holds a patent for a circadian programming device (patent no. 8,979,913). This project was funded under contract no. 290-2015-00002-I/75Q80119F32017 from the Agency for Healthcare Research and Quality (AHRQ), US Department of Health and Human Services (HHS). The authors of this manuscript are responsible for its content. Statements in this manuscript do not necessarily represent the official views of or imply endorsement by AHRQ or HHS. Representatives from AHRQ served as the Contracting Officer’s Technical Representatives and provided technical assistance during the conduct of the full evidence report and provided comments on protocol development and draft versions of the full evidence report. AHRQ did not directly participate in the literature search, determination of study eligibility criteria, data analysis or interpretation, or preparation, review, or approval of the manuscript for publication. The Centers for Medicare and Medicaid Services (CMS) nominated the systematic review to AHRQ, which selected the topic for systematic review by an Evidence-based Practice Center. Representatives from CMS provided extensive input during protocol development and the conduct of the full evidence report and provided comments on draft versions of the full evidence report that formed the basis for this manuscript. CMS did not directly participate in the literature search, determination of study eligibility criteria, data analysis or interpretation, or preparation, review, or approval of the manuscript for publication.

ABBREVIATIONS

AHI

apnea-hypopnea index

AHRQ

Agency for Healthcare Research and Quality

CPAP

continuous positive airway pressure

CV

cardiovascular

CVD

cardiovascular disease

ES

effect size

HR

hazard ratio

ITT

intention-to-treat

NRCS

nonrandomized comparative study

ODI

oxygen desaturation index

OSA

obstructive sleep apnea

RCT

randomized controlled trial

RoB

risk of bias

SAVE

Sleep Apnea cardioVascular Endpoints

SR

systematic review