Sleep Disorder Phenotypes and the Risk of Incident Heart Failure: A meta-analysis with Mendelian randomization validation

Summary

Objectives:

To quantify how specific sleep-disorder phenotypes are related to incident heart failure (HF) using a meta-analysis of cohort data and to probe causal relevance via 2-sample Mendelian randomization (MR).

Methods:

Fourteen cohorts comprised of participants without HF at baseline were pooled (49,501 exposed; 337,317 controls). A random effects meta-analysis produced pooled hazard ratios (HRs) with 95% confidence intervals (CIs). A bidirectional two-sample MR was used to evaluate the causal effects of sleep phenotypes on HF.

Results:

Overall, sleep disorders were linked to a higher incidence of HF (pooled HR 1.46; 95% CI 1.30–1.63; p<0.001). In terms of phenotype, obstructive sleep apnea (OSA) was significantly associated (pooled HR 1.39; 95% CI 1.18–1.64; p < 0.001), whereas insomnia was not independently associated with HF (pooled HR 1.11; 95% CI 0.75–1.62; p = 0.609). The MR analyses supported a causal effect of OSA on HF (inverse-variance weighted odds ratio 1.12; 95% CI 1.03–1.23; p = 0.007) and revealed no causal evidence for insomnia.

Conclusions:

Sleep disorders are associated with increased HF risk, which varies by phenotype. Convergent observational and genetic evidence implicates OSA in increased HF risk, whereas insomnia shows no independent or causal association. These findings prioritize OSA screening and targeted prevention in HF risk mitigation.

Introduction

Although heart failure (HF) incidence has stabilized or even modestly declined in recent decades, the global prevalence continues to increase because of demographic transitions, therapeutic advances, and other contributors [1,2,3]. Current estimates suggest that HF affects 1–3% of the global population, with projections indicating a further increase by 2030 [4,5]. As a leading contributor to mortality, disability, and health care burden, research on HF prevention and treatment must urgently prioritize modifiable risk factors.

Research into the link between sleep disorders and HF has increased markedly in recent years. Clinical observations indicate that up to 75% of HF patients report sleep disturbances [6,7,8]. However, whether sleep disorders represent independent risk factors, concomitant complications, or secondary consequences of HF remains unclear. Previous observational studies have shown considerable heterogeneity in results; while some studies have demonstrated that insomnia and obstructive sleep apnea (OSA) increase HF risk, other studies have reported no significant association or even contradictory findings [9,10,11,12,13]. These conflicting results pose interpretational challenges. The inherent limitations of observational studies have hindered definitive conclusions regarding the role of sleep disturbances in the pathogenesis of HF. Within genetic epidemiology, Mendelian randomization (MR) treats inherited genetic variants as instruments for modifiable exposures and allows an estimation of causal effects on outcomes. This framework reduces bias from unmeasured confounding and temporal ambiguity and yields more robust inference [14,15].

We hypothesized that the association between sleep disorders and incident HF varies by sleep-disorder phenotype and population characteristics (such as age, sex, and geographic region), may be affected by residual confounding, and could include a causal component. To clarify these associations, we implemented a two-dimensional evidence synthesis strategy as follows: (1) a systematic meta-analysis of observational studies to summarize the current evidence and quantify the differential effects of sleep disorder subtypes on HF risk; and (2) MR analysis to address residual confounding in traditional observational studies and assess causal relationships.

Methods

Guided by MOOSE and PRISMA 2020 [16,17], the protocol was preregistered in PROSPERO (CRD420251002831). The screening and data extraction were performed in duplicate. Random-effects pooling and phenotype-specific subgrouping were prespecified. Bidirectional two-sample MR with linkage disequilibrium (LD) clumped instruments was used for inverse-variance weighted (IVW) estimation as the primary method, with MR–Egger and weighted-median methods used for sensitivity analyses; the heterogeneity and horizontal pleiotropy diagnostics were reported.

We searched PubMed, Embase, Web of Science, the Cochrane Library, and ClinicalTrials.gov for cohort studies on sleep disorders and incident HF published from inception to Feb 18, 2025. The search was conducted by investigators with expertise in epidemiological research and medical librarians trained in systematic review methodologies to ensure comprehensive coverage of the relevant literature. Reference lists and supplementary materials of the included studies were manually screened to ensure comprehensive retrieval. We applied the PICOS scheme and employed both MeSH vocabulary and free-text phrases spanning four key domains as follows: (1) exposure: sleep disorders; (2) outcome: HF; (3) objective: risk factors and hazard ratios (HRs); and (4) study design: cohort studies. Queries were assembled with Boolean operators (OR/AND). Full search strings are provided in the Supplementary Materials.

This study included adult cohorts free of HF at baseline. Exposures were defined according to the ICSD-3 and included insomnia, sleep-related breathing disorders, hypersomnia, circadian rhythm sleep–wake disorders, and restless legs syndrome [18]. These conditions were diagnosed through self-reports, validated rating scales, clinical history, or objective testing (e.g., polysomnography). The control group consisted of individuals without sleep disorders. The primary outcome was incident HF events meeting guideline-defined diagnostic criteria. Eligible studies needed to provide HRs with 95% confidence intervals (CIs) for binary outcomes or report effect sizes that were convertible to HRs. Only English-language literature was included. The exclusion criteria included noncohort designs (e.g., case–control, cross-sectional, and clinical randomized controlled trials), secondary literature (reviews), case reports, animal studies, nonresearch items (letters, commentaries, and conference abstracts), and studies with incomplete outcome data.

Data, namely, publication year, sample size, geographic region, follow-up duration, participant age and sex, sleep-disorder phenotype, diagnostic methodology, HRs (95% CIs) for incident HF, and adjusted covariates, were extracted in duplicate by two reviewers. Discrepancies were resolved through consensus between two researchers and persistent disagreements were resolved by a third senior investigator. We appraised study quality with the widely adopted, validated, domain-based, methodological Newcastle–Ottawa Scale (NOS), a standardized tool for nonrandomized studies; using predefined cutoffs, scores ≥7 were considered low risk of bias, 5–6 moderate, and ≤4 high.

Meta-analytic synthesis

The meta-analysis was performed in Stata version 16.0. Unless otherwise noted, effect sizes are presented as HRs (95% CIs). To accommodate design-specific features, data synthesis followed the principles below:

Shared control group subgroup data: If a study reported subgroup-specific effect sizes without providing an overall effect size, subgroup data were pooled by constructing a covariance matrix prior to meta-analysis. This approach mitigated statistical bias from repeated utilization of control group information.

Independent control group data: When no overlap existed between control groups across subgroups, conventional meta-analysis methods were directly applied to aggregate effect sizes.

The I² index quantified heterogeneity; random effects were chosen if I² was >50%, with fixed effects used when this threshold was not met. Unexplained across-study heterogeneity was interrogated using advance-specified subgroups in combination with meta-regression analyses of study-level covariates.

Forest plots were constructed to display study-specific and pooled HRs (95% CIs), with annotations indicating study weights and model-selection criteria. Robustness was assessed using a leave-one-out sensitivity analysis. We screened for publication bias by examining funnel plots, applying Egger's linear regression, and treating p < 0.1 as significant.

Certainty of evidence

We appraised the overall certainty of the body of evidence regarding the association between sleep disorders and HF using the GRADE approach. Domains included risk of bias (via the NOS), consistency (quantified by I²), indirectness (applicability of populations and outcomes), imprecision (judged by CI width and sample-size adequacy), and publication bias (assessed with visual, symmetry oriented inspection of the funnel-plot and intercept oriented, symmetry sensitive, validated Egger's linear regression). Evidence certainty was graded (high/moderate/low/very low), and any rater disagreements were reconciled by consensus or a senior, independent, blinded, methodologically trained third reviewer.

Mendelian randomization analysis

Publicly accessible summary-level GWAS datasets were used to assemble independent exposure (sleep disorders and phenotypes) and outcome (HF) cohorts to ensure that no sample overlap occurred. Single-nucleotide polymorphisms (SNPs) served as independent, genome-wide–significant, LD-pruned, relevance-validated instrumental variables and were selected by the two following criteria: (1) a significant association with sleep disorders in exposure GWAS at genome-wide significance thresholds, and (2) variants showing any association with HF in the outcome GWAS (p < 5 × 10^–5) were removed to limit horizontal pleiotropy.

At the genome-wide level, SNPs strongly associated with sleep-disorder phenotypes were selected at p < 5 × 10^–8; when instruments were insufficient for the final MR, the threshold was relaxed to p < 5 × 10^–6. Because LD—nonrandom correlations among variants—can bias causal estimates, we performed LD clumping (using a pairwise LD r² threshold of 0.001 and a predefined, fixed, genome-wide physical 10,000-kb window). Effect alleles were harmonized, and palindromic SNPs with ambiguous strands as well as incompatible variants were excluded. To minimize weak instrument bias, F-statistics were calculated for every candidate SNP; a higher F denotes stronger instruments, and variants with F< 10 were excluded [14].

Causal inference uses the IVW estimator as the primary analysis with MR-Egger and pleiotropy aware, robustness oriented weighted median as complementary analyses. A causal relationship was declared when IVW p < 0.05 and the effect directions were concordant across all three approaches. Heterogeneity among variant-specific effects was assessed with the Cochran Q test. When the Q test was nonsignificant (p ≥ 0.05), we reported fixed-effects IVW; otherwise, the random-effects IVW was used. Sensitivity workups included the MR-Egger's linear regression intercept test (intercept p < 0.05 interpreted as bias-inducing, cross-pathway, instrument-level, directional horizontal pleiotropy) and MR-PRESSO to identify outliers and account for pleiotropy/heterogeneity. Analyses were run with TwoSampleMR 0.6.9 in R 4.4.

Results

A systematic search across five databases identified 1,102 potentially relevant studies. After screening and eligibility assessment, we included 14 studies on sleep disorders and incident HF that encompassed 49,501 exposed participants and 337,317 controls [9,10,11,12,13,19,20,21,22,23,24,25,26,27]. The PRISMA-compliant study selection flow and MR analysis are shown in Fig. 1.

Fig. 1.

Study selection (PRISMA) and MR analysis flowcharts. HR: hazard ratio, IVW, inverse variance weighted, MR: Mendelian randomization, MR-PRESSO: Pleiotropy RESidual Sum and Outlier.

Table 1 profiles the 14 included cohorts and details authorship, study region, population demographics (such as age and sex), methods used to ascertain and subtype the exposure, effect-size estimates with 95% CIs, and follow-up duration. Sleep disorder diagnoses were determined through self-reports and rating scales in eight studies (57.1%) and via ICD-9/10 codes combined with electronic medical records in six studies (42.9%). HF diagnoses were established using clinical guidelines, ICD codes, and medical history records. All studies adjusted for a predefined core set of covariates (such as sex, hypertension, diabetes, smoking, obesity, and additional study-specific variables; see Supplementary Materials Table S1).

Table 1.

Summary exposure and outcome characteristics of the included studies.

Study	Region	Age (mean, year)	Gender, male (%)	Exposure subtype	Diagnosis of exposure	HR (95% CI)	Follow-up time
Adderley et al. (2020) [10]	U.K.	60	75.0%	OSA	Clinical code	1.67 (1.35–2.06)	13
Ben-Joseph et al. (2023) [19]	USA	38	32.9%	Narcolepsy	ICD-9, ICD-10	1.35 (1.03–1.76)	5
Dalgaard et al. (2020) [9]	USA	68	54.0%	OSA	Medical history	1.14 (0.89–1.45)	1.5
Gao et al. (2021) [20]	USA	49	31.0%	Restless legs syndrome	ICD-9	1.90 (1.57–2.31)	5
Ingelsson et al. (2007) [21]	Sweden	50	100%	Unclassified	Self administered questionnaire	1.52 (1.16–1.99)	30
Javaheri et al. (2016) [11]	USA	76	100%	OSA	Polysomnography,	1.42 (0.90–2.24)	7.3
Laugsand et al. (2012) [13]	Norway	44	45.0%	Insomnia	Self administered questionnaire	1.37 (1.06–1.77)	11.3
Liu et al. (2023) [22]	U.K.	60	68.1%	Unclassified	Self-reported	1.41 (1.12–1.75)	11.1
Ljunggren et al. (2016) [23]	Sweden	44	0%	OSA	Self-reported	2.20 (1.10–4.40)	11.4
Mazzotti et al. (2019) [24]	USA	66	45.0%	Excessively sleepy	Questionnaires and scales	1.71 (1.00–2.92)	11.8
Newman et al. (2000) [25]	USA	73	42.4%	Daytime sleepiness	Questionnaire	Male: 1.49 (1.12–1.98) Female: 2.21 (1.64–2.98)	4.85
Wang et al. (2019) [26]	China	53	40.1%	Insomnia	ICD-9	1.37 (1.19–1.78)	13
Westerlund et al. (2013) [12]	Sweden	50	35.5%	Insomnia	Self-report questionnaire	0.55 (0.32–0.96)	13.2
Zhuang et al. (2023) [27]	China	51	79.3%	OSA (Snoring)	Questionnaires and scales	Occasional: 1.32 (1.14–1.52) Frequent: 1.24 (1.06–1.46)	8.8

OSA: obstructive sleep apnea; HR: hazard ratio; CI: confidence interval; ICD: International Classification of Diseases; NA: not applicable; U. K: United Kingdom; USA: United States of America.

The methodological quality was appraised with the NOS across three domains as follows: selection, comparability, and outcome assessment. All studies scored ≥7 stars and indicated high methodological quality (Supplementary Materials, Table S3).

Meta-analysis

Among the 14 included studies, one reported effect sizes exclusively for two subgroups sharing a common control group [27]. We constructed within-study covariance using generalized least squares to handle subgroups sharing the same control cohort. Harmonized estimates were 1.32 (1.14–1.52) vs. 1.24 (1.06–1.46). The other 13 reports provided nonoverlapping control groups.

A heterogeneity assessment across the 14 studies yielded an I² value of 64.9% (>50%), which indicated statistically significant interstudy heterogeneity. A random-effects analysis revealed that sleep disorders were linked to a greater risk of incident HF (1.46 (1.30–1.63); z = 6.49; p < 0.001; Fig. 2). Robustness was maintained in the leave-one-out sensitivity analyses; omitting any single study had a negligible effect on the pooled effect. Publication bias was examined with funnel plots, which appeared as approximately symmetrical, and Egger's regression (p = 0.631) likewise revealed no significant small-study effects. Under the GRADE criteria, evidence quality was deemed low, and reflected substantial heterogeneity (I² = 64.9%) and issues of indirectness and imprecision concerning phenotype and outcome assessment.

Fig. 2.

A forest plot of the association between sleep disorders and HF risk. OSA: obstructive sleep apnea, RLS: restless legs syndrome, HR: hazard ratio, CI: confidence interval.

The 14 studies were grouped into five clinical phenotypes as follows: OSA, insomnia, sleepiness, narcolepsy, and restless legs syndrome (RLS); the corresponding counts were 5, 3, 2, 1, and 1, respectively. On the basis of clinical relevance and study counts, OSA and insomnia were analyzed separately. For OSA, a random-effects model (I² = 52.3%) revealed a significant association with HF risk (1.39 (1.18–1.64); z = 3.92; p < 0.001). In contrast, the insomnia subgroup (I² = 79.4%) was not significantly associated according to a random-effects model (1.11 (0.75–1.62); z = 0.51; p = 0.609). For the remaining phenotypes (excessive daytime sleepiness, narcolepsy, and RLS; I² = 49.0%), a fixed-effects model indicated a significant association with HF risk (1.73 (1.53–1.95); z = 8.83; p<0.001; Fig. 3). Overall, OSA and the other phenotypes were significantly related to increased HF risk, whereas insomnia was not independently associated. Egger's tests revealed no publication bias across subgroups. By GRADE, the certainty for phenotype-specific associations was rated low, which was mainly because of inconsistency (such as insomnia) and imprecision in smaller phenotype strata.

Fig. 3.

A forest plot of the meta-analysis of different sleep disorder phenotypes and the risk of HF.

OSA: obstructive sleep apnea, RLS: restless legs syndrome, HR: hazard ratio, CI: confidence interval.

Exploratory probes of heterogeneity. Prespecified subgroup analyses (such as by region, sex distribution, and follow-up duration) revealed directionally consistent associations and no robust effect modification; full subgroup outputs are reported in the Supplementary Materials Table S2. A univariable meta-regression revealed no effect modification by demographic characteristics, cardiometabolic conditions, or lifestyle factors (all p > 0.05); the coefficients and diagnostics are summarized in Supplementary Materials Table S2.

Mendelian randomization analysis

To assess the causality between sleep disorders and HF, we sourced GWAS summary statistics from the publicly available, large-scale UK Biobank, FinnGen, and HERMES datasets [28,29]. Then, we used the three aforementioned MR approaches, predefined and complementary, to evaluate the putative causal effect of sleep disorder–related phenotypes on HF risk. After LD clumping and harmonization, 9–18 independent SNPs (depending on the phenotype) were retained as instruments for MR, and all instruments had F-statistics >10, which indicated adequate strength.

The principal findings of this analysis are as follows (Fig. 4): i) Overall sleep disorders were significantly linked to increased HF risk (IVW: odds ratio [OR]: 1.24 (1.10–1.42), p = 6.35 × 10^–4). However, the MR–Egger method yielded an effect estimate in the opposite direction, which suggests potential pleiotropic effects. ii) The IVW estimate provided no evidence of a link between insomnia and incident HF (OR: 1.01 (0.97–1.04); p = 0.77). iii) The evidence for a causal link between daytime sleepiness and HF is insufficient and the association showed significant heterogeneity. IVW suggested a positive link (OR: 2.23 (1.13–4.43); p = 0.021), whereas MR–Egger suggested a negative trend.iv) Sleep apnea was positively linked to incident HF (IVW method: OR: 1.12 (1.03–1.23); p = 0.007) with consistent effect directions across all three methods.

Fig. 4.

A forest plot of results of the MR analysis. IVW: inverse variance weighted, MR: Mendelian randomization, HF: heart failure, OR: odds ratio, CI: confidence interval, SNP: single-nucleotide polymorphism.

Sensitivity analyses revealed that the MR–Egger intercept test detected significant pleiotropy in the association between sleepiness and HF (p = 0.034), and the MR-PRESSO correction did not eliminate this bias. Leave-one-out checks revealed that no single SNP dominated the results, which supports the stability of the causal estimates. Reverse-direction MR (HF as the exposure; sleep disorders as outcomes) yielded null effects (all p > 0.05), which argued against reverse causation.

Discussion

For the first time, this study identifies distinct association patterns between sleep disorder phenotypes and HF risk. Observational analyses revealed a 46% overall increased risk of HF associated with sleep disorders, which was driven significantly by OSA, whereas insomnia demonstrated no independent association. The MR analysis corroborated the causal role of OSA in elevating HF risk but revealed no causal evidence for insomnia, which suggests heterogeneity in the biological mechanisms underlying different sleep disorder phenotypes and their impact on HF pathogenesis.

Sleep disorders are highly prevalent among individuals with HF in adult populations, and epidemiological evidence increasingly identifies them as critical drivers of the cardiovascular event cascade, especially in the context of HF progression. Sleep disorders may induce or accelerate HF progression via multiple mechanisms such as the following: i) the metabolic dysfunction pathway [30,31,32], ii) the inflammatory activation pathway [33,34], iii) elevated norepinephrine levels and hemodynamic changes [34,35], and iv) epigenetic and transcriptional alterations in clock genes [36,37]. Some studies have suggested a still-unconfirmed, incompletely explained, mechanistically opaque, yet plausible bidirectional relationship between HF and sleep. The Rotterdam Study demonstrated that HF could predict a decline in sleep quality [38] and partially supports this bidirectional association, although the evidence may lack sufficient power for definitive conclusions. These findings contrast with our findings that the bidirectional MR analysis revealed no reverse causal association between HF and sleep disorders. This discrepancy may arise because observational studies fail to distinguish the temporal sequence between HF symptoms and sleep disorders. Notably, while our MR analysis confirmed a causal relationship (OR = 1.24), the effect size was smaller than that observed in the observational results (HR = 1.46), which suggests that confounding factors may amplify the apparent effect.

Obstructive sleep apnea is widely recognized as an important comorbidity and major, clinically consequential, prognostically relevant, modifiable risk factor for HF with an estimated prevalence in HF populations of 11–38% [39,40]. Sleep apnea contributes to cardiovascular pathogenesis via oxidative stress, inflammation, and endothelial dysfunction, and recent evidence links OSA to alterations in gut microbiota [34,41,42]. Notably, OSA is strongly associated with HF with preserved ejection fraction (HFpEF), as reflected by German registry data showing an OSA prevalence of 69% in patients with HF with reduced ejection fraction (HFrEF) and 81% in patients with HfpEF [43]. Sex disparities further complicate the risk assessment as follows: the Sleep Heart Health Study reported elevated HF risk in male OSA patients, but these findings were not replicated in the MrOS cohort [44]. Compared with central sleep apnea (CSA), OSA in HF patients is more frequently associated with obesity, hypertension, higher left ventricular ejection fractions, and HFpEF predominance, although OSA patients have lower HF hospitalization rates than CSA patients [45]. However, there is insufficient literature on CSA-HF associations, which precluded further subgroup analyses in this study.

In a prospective study involving approximately 500,000 adults 30–79 years old, insomnia was associated with elevated risks of hypertension and coronary heart disease (HR 1.07–1.13) [46]. However, its impact on HF may be limited by confounding factors (such as depression and beta-blocker use) or phenotypic definition heterogeneity. Notably, sleep duration shows a U-shaped relationship with cardiovascular risk and both short sleep (<6–7 hours) and long sleep (>8–10 hours) may exacerbate cardiovascular damage [34]. However, this association did not manifest independently in the context of HF pathogenesis. Our findings suggest that insomnia alone may not sufficiently drive HF progression but likely requires synergistic interactions with other sleep phenotypes such as OSA.

Managing sleep disorders is challenging in routine care, especially among patients with HF [47]. Although recommendations indicate that cognitive behavioral therapy is considered the first-choice intervention for insomnia, uptake in HF patients is constrained by multimorbidity and suboptimal adherence. Emerging melatonin supplementation therapies have shown potential in experimental settings as a way to improve sleep rhythms and exert cardioprotective effects, but high-quality evidence remains lacking [48]. For patients with OSA, continuous positive airway pressure lowers the frequency of breathing disturbances, yet compliance in HF patients is hindered by intrathoracic pressure fluctuations and sleep fragmentation. Phrenic nerve stimulation for CSA decreases apneic episodes but fails to demonstrate survival benefits. This “bidirectional dilemma” in therapeutic strategies, where improving sleep may complicate HF management and optimizing HF control might compromise sleep quality, underscores the need for tailored, multidisciplinary strategies to address the two-way burden of sleep disorders and HF comorbidity.

This study is strengthened by a dual validation strategy that combines meta-analysis with MR. Additionally, only HRs from cohort studies were pooled, which allowed time-to-event modeling to more precisely reflect the long-term, dynamic HF risk linked to sleep disorders. However, several limitations should be mentioned. First, we limited the retrieval to English publications and this may have led to language or publication bias and constrained external validity. Second, although adjusted HRs were used, residual confounding and phenotype misclassification inherent to observational cohorts may persist. Third, MR inference depends on core assumptions; despite adequate instrument strength and sensitivity analyses, unbalanced horizontal pleiotropy cannot be fully excluded. Fourth, available data did not allow a formal dose–response assessment, which limits inferences about exposure thresholds and gradients.

In the conclusion, this study demonstrated an elevated risk of incident HF in the presence of sleep disorders, with phenotype specificity as follows: OSA exhibits robust evidence in both observational associations and causal inference, whereas the independent association between insomnia and HF lacks statistical and causal support. Given these findings, OSA screening should be prioritized in HF prevention strategies. Future research must refine the evidence base for sleep disorder phenotypes to inform precise interventions and mitigate HF risk.

Acknowledgment

We would like to thank American Journal Experts (www.aje.com) for English language editing.

Disclosure

This study was supported by the National Mentorship Program (No. Qngg2022049), Suzhou Science and Technology Development Plan (No. SKYD2023205), and Jiangsu Province Traditional Chinese Medicine Science and Technology Development Plan General Projects (No. MS2023106).

Disclosure of AI tool usage

No AI tools were used in the writing, image creation, data collection, or analysis for this manuscript.