Sleep‐Disordered Breathing and Sleep Fragmentation in Relation to Atrial Fibrillation Risk: A Systematic Review and Meta‐Analysis

ABSTRACT

Background

Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias worldwide. This study aimed to evaluate the association between sleep‐disordered breathing (SDB), primarily assessed using the apnea–hypopnea index (AHI), and AF risk, with additional analysis of sleep fragmentation parameters such as wake after sleep onset (WASO) and arousal index (AI).

Methods

This systematic review and meta‐analysis followed Cochrane Handbook recommendations and PRISMA guidelines, with protocol registration in PROSPERO (No. 127490). Randomized controlled trials (RCTs) and observational cohort studies assessing sleep disturbances and AF were included. MEDLINE, Cochrane Library, and Scopus were searched from inception to February 2026. Extracted data included study characteristics, demographics, cardiovascular risk factors, and sleep metrics. Outcomes were incident and recurrent AF. A random‐effects model was applied.

Results

Twenty‐one cohort studies (611,898 participants) were included. SDB was associated with increased incident AF risk (RR 1.67, 95% CI 1.41–1.97; I ² = 66.5%), with consistent Bayesian estimates. A dose–response relationship showed a 2% increase in AF risk per unit rise in AHI. For AF recurrence, 11 studies (n = 1810) demonstrated a significant association (RR 1.55, 95% CI 1.40–1.72; I ² = 0%), supported by Bayesian analysis.

Conclusion

Sleep disturbances are significantly associated with both incident and recurrent AF and may represent modifiable risk factors.

Sleep‐disordered breathing (SDB) and sleep fragmentation are associated with increased risks of incident and recurrent atrial fibrillation (AF). This meta‐analysis identifies a significant dose–response relationship between the apnea–hypopnea index (AHI) and AF risk, supporting sleep assessment as an important component of AF risk stratification and management.

1. Introduction

Cardiac arrhythmias, particularly atrial fibrillation (AF), are among the most common and clinically consequential cardiovascular disorders worldwide, substantially increasing the risks of stroke, heart failure, and mortality [1]. Using Global Burden of Disease 2021 data, this study assessed global trends in the incidence, prevalence, mortality, and DALYs of atrial fibrillation/flutter (AF/AFL) from 1990 to 2021. In 2021, AF/AFL caused 4.48 million new cases, 52.55 million prevalent cases, and 338 947 deaths, with higher burdens observed in high‐SDI regions. Projections suggest increasing incidence and prevalence through 2050 [2]. Traditional risk factors such as advancing age, hypertension, diabetes, and obesity are well established; however, there is growing recognition that sleep health also plays a critical role in cardiovascular disease pathogenesis [3] Poor sleep, including inadequate duration, irregularity, and disturbed continuity, is increasingly implicated in adverse cardiovascular outcomes [4].

Although the apnea–hypopnea index (AHI) remains the standard metric for diagnosing and grading obstructive sleep apnea (OSA), it does not fully capture the physiological burden of the disease. Complementary polysomnographic measures of sleep fragmentation—including the arousal index (AI), wake after sleep onset (WASO), sleep efficiency (SEI), and sleep fragmentation index (SFI)—quantify recurrent sleep disruption and provide additional information on impaired sleep continuity and sleep quality beyond AHI alone [5, 6, 7]. It is a common feature of OSA and often accompanies respiratory disturbances quantified by the AHI; however, similar patterns of disrupted sleep continuity may also occur independently of respiratory events in conditions such as chronic insomnia, aging, and environmental sleep disruption [5, 7, 8, 9].

Mechanistically, sleep‐disordered breathing (SDB) may contribute to cardiovascular pathophysiology through sympathetic nervous system overactivation, hypothalamic–pituitary–adrenal axis stimulation, endothelial dysfunction, systemic inflammation, and metabolic dysregulation [10]. These processes are implicated in the development of hypertension, atherosclerosis, and arrhythmogenic myocardial substrates, including alterations in autonomic tone and structural remodeling [10, 11, 12]. Such mechanisms have been well described in the context of OSA and increased AF risk, particularly in relation to respiratory disturbances quantified by the AHI. In addition to intermittent hypoxia, OSA is also associated with repeated arousals and disrupted sleep continuity, which may further contribute to arrhythmogenic pathways [13, 14].

Epidemiological evidence linking sleep fragmentation to arrhythmia risk is emerging but heterogeneous. Given these gaps, a rigorous meta‐analysis of longitudinal studies evaluating OSA and arrhythmia risk is warranted. The primary objective of this study was to quantify the association between OSA severity, SDB severity, assessed by the AHI, and the risk of incident and recurrent AF in adult populations. Secondary objectives included examining the relationships between sleep fragmentation metrics (e.g., WASO, AI, and SEI) and arrhythmia outcomes.

2. Methods

2.1. Study Design

This systematic review was designed and conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions. The review was reported following the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines. We included randomized controlled trials (RCTs), prospective and retrospective cohort studies. The protocol for this systematic review was registered with PROSPERO (No. 127490).

2.2. Search Strategy

A comprehensive literature search was conducted in MEDLINE (PubMed), Cochrane Central Library, and Scopus from database inception through February 2026, without language restrictions.

The search strategy combined controlled vocabulary terms (MeSH in PubMed) and free‐text keywords related to cardiac arrhythmias and sleep‐related disturbances. Search terms included variations of AF, atrial flutter, arrhythmia, OSA, sleep apnea, AHI, apnea, hypopnea, sleep fragmentation, sleep quality, insomnia, WASO, AI, SEI, and related terms.

Database‐specific Boolean operators, truncations, and field tags were applied as appropriate for each database. The complete electronic search strategies for all databases are provided in Supporting Information.

We manually searched for references for identified studies. Studies were excluded if they met any of the following criteria: (1) cross‐sectional design, case reports or case series, editorials or commentaries, narrative reviews, or systematic reviews/meta‐analyses (the latter were used only for citation tracking); (2) inclusion of pediatric populations (< 18 years) or non‐human studies; (3) sleep disturbance not assessed using predefined and validated metrics, including AHI, WASO, AI, sleep fragmentation (SF) index, or SEI, or when data were not extractable; (4) arrhythmia outcomes not reported or not objectively ascertained; (5) sleep assessment performed after arrhythmia onset in analyses of incident AF (i.e., absence of temporal sequencing); (6) duplicate or overlapping cohorts, in which case the most comprehensive or recent report was retained; (7) sleep assessed exclusively during acute hospitalization, intensive care, or perioperative settings; (8) cohorts restricted to patients with end‐stage or decompensated heart failure in whom nocturnal symptoms predominated sleep disruption; or (9) studies in which sleep disturbance was clearly secondary to established arrhythmia. The articles collected from the systematic search were transferred to the Rayyan library software, where duplicates were eliminated.

2.3. Study Selection

All identified records were imported into Rayyan and duplicates were removed. Subsequently, two independent reviewers screened titles and abstracts, followed by full‐text reviews of potentially eligible studies; only those that satisfied the abovementioned criteria were included. Discrepancies were resolved through discussion and consultation with a third reviewer. Study selection was summarized using a PRISMA flow diagram. The search strategy, research questions, PICO, MeSH, and keywords were mentioned in Figure S1.

2.4. Data Extraction

The following baseline characteristics were extracted from the included studies: sample size, age, sex, body mass index (BMI), race, hypertension, OSA, coronary artery disease, tobacco and alcohol use, sleep assessment modalities, and sleep fragmentation metrics. Study outcomes, including incident AF and recurrence of AF, were also abstracted.

2.5. Risk of Bias Assessment

The methodological quality of included studies was assessed independently by two reviewers. Observational studies were assessed using the ROBINS‐E tool. Disagreements were resolved by consensus.

2.6. Statistical Analysis

For dichotomous outcomes, risk ratios (RRs) with 95% confidence intervals (CIs) were calculated. Due to the limited number of studies reporting pure sleep fragmentation metrics, the primary meta‐analysis was based on SDB severity (AHI). Fragmentation‐specific measures were analyzed descriptively and in secondary analyses when feasible.

Meta‐analyses were conducted using R (version 4.5.2) with the meta, metafor, metamedian, and bayesmeta packages. Random‐effects models were applied to account for anticipated clinical and methodological heterogeneity. Statistical heterogeneity was assessed using the chi‐square (Cochran Q) test and quantified using the I ² statistic, with values > 50% considered indicative of substantial heterogeneity. Statistical significance was defined as a two‐sided p < 0.05. Publication bias was evaluated through visual inspection of funnel plots.

A two‐stage random‐effects dose–response meta‐analysis was performed using the Greenland and Longnecker method to examine the relationship between increasing sleep apnea severity (measured by the AHI) and the risk of AF. Study‐level risk of bias was assessed using the ROBINS‐E tool.

2.7. Subgroup and Sensitivity Analyses

Subgroup analyses were conducted when data allowed, including modality (PSG versus actigraphy/wearables) and endpoint type (incident versus recurrent arrhythmia). Sensitivity analyses were conducted using leave‐one‐out analyses. All analyses were performed using R to generate forest plots and robustness checks.

3. Results

3.1. Search Results

The search identified 4235 records: 1710 from Scopus, 2342 from PubMed, and 183 from Cochrane. After removing 300 duplicate records, 3935 studies remained for the title and abstract screening. Following the screening process, 3793 studies were excluded based on title and abstract evaluation. One hundred forty‐two studies were assessed for eligibility, of which 21 studies [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34] were included in the qualitative and quantitative synthesis. The detailed screening process is illustrated in Figure 1.

FIGURE 1.

PRISMA flow diagram of study selection.

3.2. Study Characteristics

The baseline characteristics table summarizes 21 cohort studies investigating SDB and AF, including 10 studies evaluating incident AF, of which 8 contributed to the quantitative meta‐analysis, and 11 studies assessing AF recurrence. The studies were conducted across North America, Europe, and Asia and included both prospective and retrospective cohort designs, with follow‐up durations ranging from 0.9 to 11.9 years and sample sizes varying widely from 44 to 579 521 participants. Sleep assessment was predominantly performed using in‐laboratory or home polysomnography (PSG), while several studies used portable sleep apnea testing or ambulatory monitoring. Across studies, the reported AF event rates showed substantial variability, with incidence proportions ranging from 0.013 to 0.142 and recurrence rates ranging from 0.25 to 0.81, reflecting differences in study populations, diagnostic modalities, and clinical settings, as shown in Table 1.

TABLE 1.

Baseline characteristics of included patients.

Study	Country	Study design	Follow‐up (years)	Sample size	Mean Age	Mean BMI	Sleep assessment modality	Fragmentation metrics	Outcome category	AF rate
Wang et al. [14]*	China/USA	Prospective cohort study	5.3	1660	64	29	PSG	SBII, HB, AHI	Incidence	0.137
Chen et al. [15]	China/USA	Prospective cohort study	10	1810	63	28.7	Home PSG	WASO, SEI	Incidence	0.137
May et al. [16]	USA	Prospective cohort study	6.5	843	65	27.1	PSG	AHI, CAI	Incidence	0.142
Chao et al. [17]	Taiwan	Nationwide longitudinal cohort study	9.2	579 521	38.9	NR	PSG	SDB severity, AHI	Incidence	0.013
Costa et al. [18]*	USA	Prospective cohort study	3.83	1858	67	28.7	Home PSG	Heart rate fragmentation	Incidence	0.045
Kendzerska et al. [20]	Canada	Historical cohort study	10	8256	47	28.8	PSG	TST, AHI	Incidence	0.021
Gami et al. [22]	USA	Retrospective cohort study	4.7	3542	49	33	PSG	AHI, AI	Incidence	0.043
Tung et al. [24]	USA	Prospective cohort study	5	2912	62.8	28.2	Portable sleep apnea testing	Sleep‐disordered breathing	Incidence	0.137
Cadby et al. [27]	Australia	Retrospective cohort study	11.9	6841	NR	NR	Portable sleep apnea testing	Events per hour of sleep	Incidence	0.086
Mehra et al. [31]	USA	Prospective cohort study	2	2911	76.35	26.65	Portable sleep apnea testing	AHI	Incidence	0.053
Kanagala et al. [32]	USA	Prospective cohort study	1	43	66	32	PSG	AHI	Recurrence	0.815
Minami et al. [19]	Japan	Prospective cohort study	1.2	94	65.74	24.5	EEG + actigraphy	Fragmentation index	Recurrence	0.364
Trenkwalder et al. [21]	Germany	Prospective cohort study	1	187	60.3	27.7	Ambulatory polygraphy	AHI	Recurrence	0.343
Matiello et al. [23]	Spain	Prospective cohort study	1.42	174	52	27.3	Berlin questionnaire + home study	AHI	Recurrence	0.786
Tang et al. [25]	China	Prospective cohort study	0.9	178	57	25.5	Portable sleep apnea testing	Events per hour of sleep	Recurrence	0.250
Szymanski et al. [26]	Poland	Retrospective cohort study	2.5	251	57.7	29.7	Portable sleep apnea testing	Events per hour of sleep	Recurrence	0.652
Kawakami et al. [28]	Japan	Prospective cohort study	1.1	124	NR	NR	Type‐III portable monitoring	Sleep‐disordered breathing	Recurrence	0.546
Naruse et al. [29]	Japan	Prospective cohort study	1.5	153	NR	NR	Portable sleep apnea testing	Sleep‐disordered breathing	Recurrence	0.371
Mazza et al. [30]	Italy	Prospective cohort study	1	158	70	28.2	Portable sleep apnea testing	AHI	Recurrence	0.694
de Heide et al. [33]	Netherlands	Retrospective cohort study	1	104	59	26.8	WatchPAT	AHI	Recurrence	0.513
Jongnarangsin et al. [34]	USA	Prospective cohort study	1	324	58	32	PSG	AHI	Recurrence	0.594

^*Indicates studies that were included only in the qualitative analysis.

3.3. Risk of Bias Assessment

Using the ROBINS‐E tool, most included studies were judged to have a low risk of bias across several domains, particularly for measurement of exposure, outcome assessment, and selection of the reported results. However, some concerns were identified in multiple studies, mainly related to potential confounding and participant selection into the study or analysis. A high risk of bias was observed in a small number of studies, primarily due to confounding and post‐exposure interventions, whereas the risk of bias related to missing data was generally low. Detailed domain‐level assessments are presented in the Figure S2.

3.4. Outcomes

3.4.1. Primary Outcomes

3.4.1.1. Incident Atrial Fibrillation

Analysis of 8 studies including 606 570 participants demonstrated a statistically significant increased risk in the OSA group compared with controls using a random‐effects model (RR = 1.67, 95% CI 1.41–1.97). Moderate heterogeneity was observed among studies (I ² = 66.5%, τ² = 0.0354, p = 0.0039).

Subgroup analysis according to the cutoff value showed similar effect estimates. Studies using a cut‐off of 15 demonstrated an increased risk (RR = 1.60, 95% CI 1.31–1.95; I ² = 73.8%), while studies using a cut‐off of 5 also showed a significant association (RR = 1.93, 95% CI 1.53–2.44; I ² = 0%). Differences between these subgroups were not statistically significant (p = 0.23).

Subgroup analysis by sleep assessment method showed that studies using portable PSG reported a pooled risk ratio of 1.47 (95% CI 1.14–1.90), whereas studies using standard PSG showed a stronger association (RR = 1.94, 95% CI 1.68–2.25). The single study using mixed assessment methods reported RR = 1.88 (95% CI 1.44–2.46). Differences between sleep assessment subgroups were not statistically significant under the random‐effects model (p = 0.18) (Figure 2). Leave‐one‐out sensitivity analysis demonstrated stable results, with pooled risk ratios ranging from 1.60 to 1.77, indicating that no single study materially influenced the overall effect estimate (Figure S3).

FIGURE 2.

Forest plot of the association between sleep‐disordered breathing and incident atrial fibrillation with subgroup analyses.

A mixed‐effects meta‐regression model was conducted to evaluate the influence of BMI and mean age as potential moderators. Neither BMI (estimate = 0.054, p = 0.578) nor mean age (estimate = 0.008, p = 0.660) significantly explained the between‐study heterogeneity. Residual heterogeneity remained substantial (τ ² = 0.065; I ² = 75.85%), and the moderators did not account for any meaningful proportion of variability across studies (R ² = 0%). The overall test of moderators was not statistically significant (QM = 0.31, p = 0.855), indicating that BMI and age did not significantly modify the association. Residual heterogeneity remained statistically significant (QE = 11.91, p = 0.008).

Assessment of publication bias showed that the funnel plot appeared visually symmetric, suggesting minimal publication bias. This was supported by Egger's regression test (p = 0.64), which did not demonstrate statistically significant small‐study effects (Figure S4).

A Bayesian random‐effects meta‐analysis of 8 studies demonstrated a significant association. The posterior median pooled effect was μ = 0.50 (95% CrI 0.30–0.68), corresponding to an approximate risk ratio of 1.65 (95% CrI 1.36–1.97). The posterior probability that the effect exceeded zero was > 99.9%, indicating strong evidence for a positive association. Moderate between‐study heterogeneity was observed (τ = 0.20, 95% CrI 0.05–0.39), with a posterior median I ² of 66.7%. Bayes factor analysis further supported evidence against the null hypothesis (BF ≈0.002 for μ = 0). Sensitivity analyses using wider priors produced nearly identical results, confirming the robustness of the findings (Figure S5).

The linear dose–response analysis demonstrated a significant association between increasing AHI and AF risk (χ ² = 19.82, p < 0.001). Each 1‐unit increase in AHI was associated with an approximately 2.1% increase in AF risk (RR ≈1.02). Moderate heterogeneity was observed across studies (I ² = 67.5%; Q = 9.23, p = 0.026). To further explore the shape of this association, an exploratory spline analysis was performed. The spline model demonstrated a significant dose–response relationship (F = 23.14, p < 0.001), with both spline terms remaining statistically significant (p = 0.0001 and p = 0.0154). The exploratory spline analysis demonstrated a progressive increase in AF risk with increasing AHI, with the predicted relative risk rising from 1.07 at an AHI of 5 to 1.71 at an AHI of 30, followed by a plateau at higher AHI levels, indicating that the association may not be linear (Figure 3).

FIGURE 3.

Dose–response relationship between apnea–hypopnea index (ahi) and risk of atrial fibrillation a: linear model, cubic spline model.

3.4.1.2. Recurrent Atrial Fibrillation

Eleven studies involving 709 exposed and 1101 control participants evaluated the risk of recurrent AF. The pooled analysis demonstrated that SDB was associated with a significant increase in the risk of recurrent AF (RR = 1.55, 95% CI 1.40–1.72) with no observed heterogeneity (I ² = 0%). Subgroup analysis based on the sleep assessment modality showed consistent findings. Studies using PSG demonstrated a significant increase in the recurrence of AF (RR = 1.51, 95% CI 1.25–1.83) with no heterogeneity (I ² = 0%). Similarly, studies using portable sleep monitors were also associated with a significant increase in recurrence risk (RR = 1.55, 95% CI 1.34–1.80) with low heterogeneity (I ² = 24.9%). The single study using the Berlin questionnaire also demonstrated a significant association (RR = 1.62, 95% CI 1.28–2.05) (Figure 4). Leave‐one‐out sensitivity analysis showed stable results, with pooled risk ratios ranging from 1.53 to 1.61, and all estimates remained statistically significant. This indicates that no single study significantly influenced the overall effect (Figure S6).

FIGURE 4.

Forest plot of the association between sleep‐disordered breathing and recurrent atrial fibrillation with subgroup analyses.

A mixed‐effects meta‐regression model (REML, k = 8) was conducted to evaluate whether mean age and mean BMI moderated the association between exposure and recurrent AF. The overall test of moderators was not statistically significant (p = 0.112).

A Bayesian random‐effects meta‐analysis demonstrated good model convergence (R̂ = 1.00). The pooled posterior median risk ratio (RR) for recurrent AF was 1.57 (95% CrI 1.38–1.79). The posterior probability that RR > 1 was 100%, with a 99.84% probability that RR > 1.2 and 74.84% probability that RR > 1.5. Between‐study heterogeneity was low τ = 0.07 (95% CrI 0.00–0.24). There was no statistically significant evidence of publication bias on Egger's regression test (p = 0.14) and visual inspection of the funnel plot did not suggest substantial asymmetry (Figure S7).

3.4.2. Secondary Outcomes

Analysis of secondary sleep parameters revealed several notable differences between groups. Patients who developed new AF exhibited a significantly higher median AHI (MD = 4.17 events/h, 95% CI: 1.29–7.06), suggesting a greater burden of SDB. Increased WASO suggested greater sleep disruption in the AF group (MD = 12.95 min, 95% CI: 7.05–18.85), while total sleep time (TST) was significantly reduced (MD = −23.86 min, 95% CI: −43.52 to −4.20). In contrast, sleep latency, T90, and the AI did not differ significantly between groups (Figure 5).

FIGURE 5.

Forest plot of secondary sleep parameters associated with atrial fibrillation.

With respect to AF recurrence, pooled analysis demonstrated that patients with recurrent AF had a significantly higher mean AHI compared with those without recurrence (MD = 4.52 events/h, 95% CI: 1.68–7.37), consistent with the observed association between greater sleep apnea severity and AF persistence. However, no significant difference was observed in the AI between the recurrent and non‐recurrent AF groups (Figure 6).

FIGURE 6.

Forest plot of secondary parameters related to atrial fibrillation recurrence.

4. Discussion

In this systematic review and meta‐analysis including 21 longitudinal studies with 665 457 participants, objectively measured sleep apnea was associated with a higher risk of incident AF (RR 1.67, 95% CI 1.41–1.97; I ² = 66.5%). Subgroup analyses showed consistent results across AHI thresholds (≥ 15: RR 1.60, 95% CI 1.31–1.95; ≥ 5: RR 1.93, 95% CI 1.53–2.44). thresholds (≥ 15: RR 1.60, 95% CI 1.31–1.95; ≥ 5: RR 1.93, 95% CI 1.53–2.44). The subgroup differences likely reflect variations in study‐defined AHI thresholds, as studies using an AHI cut‐off of ≥ 5 also included patients with higher AHI values and did not distinguish between mild and more severe OSA. It was also associated with increased AF recurrence (RR 1.55, 95% CI 1.40–1.72), with similar effects across sleep assessment methods. Bayesian analyses produced comparable estimates, supporting the robustness of the findings. Mean age and BMI were not significant moderators of the association between SDB and recurrent AF. However, given the relatively small number of studies included in the meta‐regression (k = 11), the analysis may have been underpowered to detect modest moderator effects.

The role of sleep in cardiovascular health has gained increasing recognition, with the American Heart Association and American College of Cardiology identifying sleep as a fundamental component of cardiometabolic health [35, 36]. While OSA is a well‐established risk factor for AF through mechanisms such as intermittent hypoxia, intrathoracic pressure changes, and autonomic dysfunction [10], the independent contribution of sleep disturbance remains uncertain. Observational evidence, including large population‐based studies such as the Sleep Heart Health Study, suggests that markers of fragmented sleep—such as increased WASO, higher arousal frequency, and reduced SEI—are associated with AF risk, even after accounting for OSA severity [14]. Our findings support an association between sleep disturbance and AF, although heterogeneity likely reflects variability in sleep assessment methods, AF detection strategies, follow‐up duration, and differences in covariate adjustment across studies.

This relationship is biologically plausible, as recurrent sleep disruption can induce sympathetic activation, parasympathetic withdrawal, and alterations in atrial refractoriness, promoting arrhythmogenesis [11, 12]. Additionally, sleep disturbance has been linked to activation of the hypothalamic–pituitary–adrenal axis, systemic inflammation, oxidative stress, and endothelial dysfunction, all of which may contribute to atrial remodeling and fibrosis [11, 12]. Unlike OSA, where intermittent hypoxia predominates, sleep fragmentation may arise from diverse conditions such as insomnia, aging, and circadian misalignment [7, 37]. Together, these findings suggest that impaired sleep continuity may represent an additional pathway influencing AF risk beyond classical SDB.

The subgroup findings suggest that both study size and sleep assessment method may have influenced the observed associations. Among PSG‐based studies, Cadby et al. demonstrated that increasing severity of OSA independently predicted incident AF hospitalization in a large sleep‐clinic cohort [27]. Earlier work by Gami et al. also identified a significant relationship between OSA and AF using PSG [22]. In contrast, Chen et al. using a portable monitor reported an association between sleep apnea risk scores and cardiovascular disease [15], while Tung et al. and Mehra et al. also demonstrated links between SDB and AF or nocturnal arrhythmias in large community‐based cohorts [24, 31]. Although these studies supported a positive association, the overall effect size was generally smaller than in PSG‐based studies. The larger estimate reported by Costa et al. should be interpreted cautiously given its wider confidence interval and lower statistical weight [18].

Some sleep fragmentation metrics, particularly WASO, showed significant associations with AF, whereas others such as the AI did not. This discrepancy may reflect differences in the physiological meaning of these measures, as WASO captures sustained sleep disruption while the AI reflects brief micro‐arousals with less consistent clinical impact. Additionally, variability in measurement methods and the limited number of studies reporting specific fragmentation metrics may have reduced statistical power and contributed to inconsistent findings [38].

The AI was not significantly associated with AF despite its close relationship with SDB severity. This finding is consistent with a recent prospective cohort study in which WASO, but not the AI, remained independently associated with incident AF after adjustment for AHI and other confounders. The authors proposed that WASO more directly reflects sleep continuity and cumulative wakefulness, whereas the AI primarily measures the frequency of brief arousals and may be influenced by different arousal mechanisms, limiting its independent association with AF [14].

Such outcomes carry significant clinical relevance. For one, quantifiable measures of sleep disruption, like WASO and arousal frequency, could act as supporting markers in AF risk assessment. Including sleep continuity evaluation in heart‐related examinations aligns with modern perspectives acknowledging rest as an adjustable aspect of cardiometabolic well‐being [36]. Additionally, the link with AF reappearance proposes that broken sleep may hinder stable atrial function recovery following rhythm‐regulation procedures. Assessing sleep patterns before catheter procedures or electrical cardioversion might help pinpoint patients prone to relapse. Finally, if causation is verified, SDB and fragmentation may represent a potentially modifiable risk factors, pending interventional validation. Approaches including cognitive behavioral therapy for insomnia (CBT‐I), circadian rhythm adjustments, optimized sleep environments, and treating concurrent breathing disorders could enhance sleep consistency. Controlled intervention studies will be crucial in verifying whether correcting fragmentation metrics lowers AF recurrence or return rates.

The linear dose–response analysis showed that each 1‐unit increase in AHI was associated with an approximately 2% increase in AF risk. Exploratory spline analysis suggested a progressive increase in risk that plateaued at higher AHI levels, consistent with cumulative physiological stress [39]. This occurs because each apnea event triggers negative intrathoracic pressure, intermittent hypoxia, and sympathetic surges that cause acute atrial stretching and electrical instability. Over time, this repetitive stress promotes structural remodeling and fibrosis, creating a substrate for arrhythmia. Clinically, this suggests that even mild elevations in AHI carry a modest but significant increase in risk, shifting the focus toward early intervention.

CPAP therapy, a key modifier of AF outcomes, was inconsistently reported and therefore could not be evaluated in the present analysis. This is clinically relevant, as CPAP has been suggested to improve atrial remodeling and reduce AF recurrence; however, evidence remains inconsistent [40]. The absence of CPAP data may have introduced residual confounding.

Incident AF and recurrent AF are distinct clinical entities. Incident AF indicates the initial development of AF, whereas recurrent AF occurs after rhythm‐control interventions and is primarily caused by persistent arrhythmogenic substrate, pulmonary vein reconnection, incomplete lesion formation, and atrial remodeling. The moderate heterogeneity observed in the incident AF analysis (I ² = 66.5%) likely reflects differences in study populations, definitions of SDB (AHI thresholds ≥ 5 versus ≥ 15 events/h), sleep assessment methods (standard PSG, portable monitoring, and mixed approaches), AF ascertainment strategies, follow‐up duration, and adjustment for cardiovascular risk factors. Nevertheless, subgroup, sensitivity, and Bayesian analyses consistently demonstrated a positive association, supporting the robustness of the overall findings despite between‐study variability. In contrast, the association between SDB and recurrent AF was remarkably consistent across studies (I ² = 0%), indicating minimal between‐study variability. This consistency likely reflects the inclusion of more homogeneous patient populations undergoing rhythm‐control interventions and standardized definitions of AF recurrence. The reproducibility of this finding strengthens the evidence that SDB is an important and potentially modifiable determinant of AF recurrence, underscoring the value of systematic screening and optimization of SDB in patients undergoing AF ablation or cardioversion.

This study has some strengths and limitations. This investigation adhered to PRISMA and Cochrane methodology, performing an extensive literature search through February 2026. Subgroup examinations and sensitivity assessments were predetermined, alongside systematic appraisal of bias risk. Both frequentist and Bayesian methods were employed to show analytical robustness. No statistically significant indications of publication bias emerged, though certain constraints warrant acknowledgment. A majority of incorporated investigations featured observational designs, which inherently restricts causal interpretation. Potential residual confounding factors, such as undetected OSA, pharmaceutical utilization, alcohol consumption, or accompanying cardiometabolic conditions, may persist despite adjustments. Standardization of fragmentation measures proved inconsistent, while considerable variability in AF occurrence analyses highlights both clinical and procedural discrepancies. Additionally, investigation of arrhythmia recurrence involved diminished cohort sizes, potentially compromising conventional analytical rigor.

An important point to be mentioned is that this meta‐analysis evaluated SDB rather than the independent effects of OSA and central sleep apnea (CSA). Although some included studies, such as Tung et al., Kendzerska et al., and de Heide et al. [20, 24, 33], distinguished OSA from CSA, others, including Kanagala et al. [32], reported only overall SDB‐ or OSA‐based outcomes without separate analyses of CSA. Consequently, the pooled findings reflect the overall association between SDB and AF rather than the individual contributions of OSA and CSA.

An important limitation of this meta‐analysis is that most included studies reported only the total AHI without distinguishing OSA from CSA. This distinction is clinically relevant because OSA and CSA may promote AF through different pathophysiological mechanisms. OSA is primarily associated with intermittent hypoxia, negative intrathoracic pressure swings, and autonomic imbalance, whereas CSA is more closely linked to unstable ventilatory control, hypocapnia‐related electrophysiological instability, and autonomic dysfunction. Prospective studies, including the Sleep Heart Health Study, have suggested that CSA may be a stronger independent predictor of incident AF than OSA in certain populations. Consequently, pooling undifferentiated AHI data may have obscured subtype‐specific associations. Nevertheless, because community‐based cohorts are predominantly composed of patients with OSA, the pooled estimates in this meta‐analysis likely reflect primarily OSA‐related risk.

Another limitation is the heterogeneity of sleep assessment methods across the included studies. Several studies relied on portable sleep monitoring or questionnaire‐based screening tools rather than attended PSG. These approaches have limited ability to distinguish obstructive from central respiratory events, and commonly used screening questionnaires were developed to detect OSA rather than CSA. Furthermore, home sleep apnea testing reports the respiratory event index (REI), which is based on total recording time rather than TST and may therefore underestimate SDB severity. These methodological differences may have resulted in misclassification of disease severity and underrecognition of CSA, potentially attenuating subtype‐specific associations. Future studies using standardized PSG with separate reporting of obstructive and central respiratory events are needed to better define their independent contributions to AF risk.

Sleep disturbance is associated with an increased risk of incident AF and recurrent AF following catheter ablation. Despite substantial heterogeneity, the association remained consistent across sensitivity and Bayesian analyses. These findings suggest that sleep disturbance may represent a potentially modifiable risk factor influencing both AF development and post‐ablation recurrence.

Funding

The authors have nothing to report.

Disclosure

All authors take responsibility for all aspects of reliability and freedom from bias of the data presented and their discussed interpretation.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Risk of bias assessment using the ROBINS‐E tool.

Figure S2: Funnel plot assessing publication bias for incident atrial fibrillation.

Figure S3: Funnel plot assessing publication bias for recurrent atrial fibrillation.

Figure S4: Leave‐one‐out sensitivity analysis for incident atrial fibrillation.

Figure S5: Leave‐one‐out sensitivity analysis for recurrent atrial fibrillation.

Figure S6: Bayesian posterior distribution for incident atrial fibrillation.

Figure S7: Bayesian posterior distribution for recurrent atrial fibrillation.

Data Availability Statement

The data that supports the findings of this study are available in the supporting information of this article.