Individualized treatment selection for obstructive sleep apnea: network meta-analysis focusing on position dependent patients

Abstract

Introduction

Obstructive sleep apnea (OSA) affects nearly 1 billion adults worldwide. Treatment options include positive airway pressure (PAP), oral appliance therapy, and positional therapy. Few studies have comprehensively compared these treatments using network meta-analysis. This study aimed to compare the effectiveness of different non-surgical OSA treatments across several clinically relevant outcomes.

Methods

A systematic review and network meta-analysis were conducted following PRISMA guidelines. We searched MEDLINE, Embase, CENTRAL, Web of Science, and ClinicalTrials.gov through October 2025. Randomized controlled trials comparing PAP therapy, oral appliance therapy, positional therapy, or control in adults with OSA were included. Primary outcomes were AHI reduction and Epworth Sleepiness Scale (ESS) improvement. Secondary outcomes included oxygen saturation. Bayesian network meta-analysis was performed, with treatment rankings using surface under the cumulative ranking curve (SUCRA).

Results

Eighteen trials involving 1,520 participants were included. For AHI reduction, PAP therapy was most effective (mean difference: -17.39 events/hour, 95% CI: -23.13 to -12.74), followed by oral appliance therapy (-10.76 events/hour) and positional therapy (-10.05 events/hour). PAP therapy ranked first with 100% probability, while oral appliance therapy and positional therapy showed similar effectiveness. For ESS improvement, PAP showed slight superiority. In position-dependent OSA, oral appliance therapy ranked first with 40% probability. Moderate heterogeneity was found (I² = 32–52%).

Conclusions

PAP therapy is the most effective treatment for OSA, while oral appliance and positional therapies offer viable alternatives, especially for position-dependent OSA patients. However, moderate heterogeneity (I² = 32–52%) was observed across analyses, which may affect the certainty of these conclusions. Treatment selection should consider patient-specific factors, and findings should be interpreted acknowledging this methodological limitation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12890-026-04268-1.

Introduction

Obstructive sleep apnea (OSA) is a common sleep-related breathing disorder characterized by repeated episodes of complete or partial upper airway collapse during sleep [1]. OSA is estimated to affect nearly 1 billion adults globally, with significant variation across populations and age groups [2].

OSA is associated with numerous adverse health consequences, including cardiovascular disease, metabolic disorders, neurocognitive impairment, and reduced quality of life [3]. The economic burden of untreated OSA is substantial, encompassing direct healthcare costs and indirect costs related to reduced productivity and increased accident rates [4].

Since the introduction of continuous positive airway pressure (CPAP) therapy by Sullivan and colleagues in 1981 [5], multiple treatment modalities have been developed for OSA management. Current treatment options include positive airway pressure (PAP) therapy, oral appliance therapy (OAT), positional therapy (PT), upper airway surgery, and lifestyle modifications [6].

Despite the availability of multiple treatment options, adherence to OSA therapy remains a significant challenge. PAP adherence rates are often suboptimal, with studies showing that approximately 30–50% of patients discontinue PAP therapy within the first year [7]. Long-term adherence data suggest that this trend has remained relatively stable over the past two decades [8].

Alternative treatment modalities, such as oral appliance therapy, have gained increasing recognition as viable options for OSA management, particularly for patients with mild to moderate OSA or those who cannot tolerate PAP therapy [9]. Similarly, positional therapy has emerged as a treatment option for patients with position-dependent OSA, which comprises a significant proportion of OSA cases [10].

While numerous studies have compared individual treatment modalities, comprehensive comparisons of all available treatment options using network meta-analysis methodology are limited. Traditional pairwise meta-analyses can only compare two treatments at a time, whereas network meta-analysis allows for simultaneous comparison of multiple treatments and provides a more comprehensive understanding of treatment effectiveness [11].

Position-dependent OSA, defined as OSA that is significantly more severe in the supine position compared to non-supine positions, affects approximately 50–60% of OSA patients [12]. The recognition of positional OSA has led to the development of positional therapy devices and techniques aimed at maintaining non-supine sleep positions [13].

The objective of this systematic review and network meta-analysis is to comprehensively compare the effectiveness of different OSA treatment modalities, including CPAP therapy, oral appliance therapy, and positional therapy, across multiple clinically relevant outcomes. This analysis aims to provide evidence-based guidance for clinicians and patients in treatment selection and to identify areas where further research is needed.

Methods

Study design and registration

This systematic review and network meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14] and the PRISMA extension for network meta-analyses [15]. The protocol was prospectively registered in PROSPERO (CRD420251176171).

Search strategy

A comprehensive search strategy was developed in consultation with a medical librarian. The following databases were searched from inception to January 2025: MEDLINE (via PubMed), Embase, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science, and ClinicalTrials.gov. Grey literature was sought through ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform (ICTRP) to identify completed but unpublished studies. Additionally, reference lists of included studies and relevant systematic reviews were manually searched, and we contacted authors of conference abstracts identified during screening to request full data when available. We acknowledge that a more comprehensive grey literature search (including dissertation databases and industry reports) was not conducted, which represents a limitation of this review.

No restrictions were applied based on language or country of publication. Non-English articles were translated using professional translation services when necessary.

The search terms included combinations of the following: “obstructive sleep apnea,” “sleep apnea syndromes,” “continuous positive airway pressure,” “CPAP,” “oral appliance,” “mandibular advancement device,” “positional therapy,” “sleep position,” “randomized controlled trial,” and “clinical trial.” Language restrictions were not applied, and reference lists of included studies and relevant systematic reviews were manually searched for additional eligible studies.

Eligibility criteria

Studies were included if they met the following criteria:

• Study design: Randomized controlled trials (RCTs)

• Population: Adults (≥18 years) with confirmed OSA diagnosis

• Interventions: PAP therapy, oral appliance therapy, positional therapy, or no treatment/placebo

• Positive airway pressure (PAP) therapy: In this study, PAP is used as a general term encompassing both continuous positive airway pressure (CPAP) and auto-adjusting positive airway pressure (APAP). When describing specific studies, the original terminology used in those studies is retained. The included studies predominantly used CPAP, with two studies employing APAP.

• Oral appliance therapy (OAT): Included custom-fitted mandibular advancement devices (MADs) that reposition the mandible anteriorly to increase upper airway patency. Devices were fabricated by dental professionals and titrated to achieve optimal mandibular advancement (typically 50-75% of maximum protrusion) while maintaining patient comfort.

• Positional therapy (PT): Included any intervention designed to reduce supine sleep time, encompassing: (a) the tennis ball technique (TBT) using a ball or similar object sewn into nightwear to discourage supine positioning; (b) commercially available positional devices worn on the chest or neck that deliver vibratory alerts when supine position is detected; (c) electronic sleep position trainers (e.g., NightBalance, Zzoma) providing graduated feedback.

• Nocturnal oxygen supplementation (NOS): Supplemental oxygen delivered via nasal cannula during sleep, typically at 2-4 L/min flow rate.

• Control conditions: Included: (a) no treatment/usual care; (b) placebo devices (inactive oral appliances without mandibular advancement, sham CPAP with subtherapeutic pressure <1 cm H₂O, positional devices worn but deactivated); (c) sleep hygiene education alone; (d) conservative management with lifestyle advice.

• Outcomes: At least one of the following outcomes reported: apnea-hypopnea index (AHI), Epworth Sleepiness Scale (ESS), oxygen desaturation index, or treatment adherence

• Follow-up: Minimum follow-up period of 4 weeks. This threshold was selected based on the following rationale: (1) a minimum of 4 weeks allows sufficient time for patients to adapt to treatment devices (particularly oral appliances, which often require titration) and achieve stable therapeutic effects; (2) this duration is consistent with regulatory standards for OSA device efficacy trials and prior network meta-analyses in this field; (3) shorter follow-up periods (e.g., single-night studies) may not reflect real-world treatment effects due to first-night adaptation effects. We acknowledge that this relatively short minimum threshold means our analysis may not fully capture long-term efficacy and adherence patterns, which is noted as a limitation.

• Subgroup definition: Position-dependent OSA was defined according to criteria used in individual studies, most commonly as supine AHI at least twice the non-supine AHI, or alternatively as supine AHI ≥10 events/hour with non-supine AHI <10 events/hour. We acknowledge that definitions varied somewhat across included studies (Jokic et al. used supine AHI/non-supine AHI ratio ≥2; Benoist et al. required supine AHI ≥10 and non-supine AHI <10; Jackson et al. used percentage of supine sleep time >50% with supine AHI at least double non-supine AHI).

• Follow-up: Minimum follow-up period of 4 weeks

Studies were excluded if they involved surgical interventions, pediatric populations, or lacked sufficient data for network meta-analysis. Conference abstracts, case reports, and observational studies were also excluded.

Study selection and data extraction

Two reviewers independently screened titles and abstracts, followed by full-text review of potentially eligible studies. Disagreements were resolved through discussion or consultation with a third reviewer. (flow diagrams see in the Fig. 1)

Fig. 1.

PRISMA 2020 flow diagram for the network meta-analysis of OSA treatments. Shows the systematic literature search and study selection process, from identification (n = 3,270 records) to inclusion (n = 18 RCTs, 1,520 participants), with exclusions detailed at each stage

Data extraction was performed using a standardized form and included study characteristics, participant demographics, intervention details, outcome measures, and follow-up duration. For studies with multiple follow-up time points, data from the longest follow-up period were extracted. For studies with multiple follow-up time points, data from the longest follow-up period were extracted. This approach was chosen to: (1) avoid unit-of-analysis errors arising from including correlated data from the same participants at multiple time points; (2) capture the most mature treatment effects and adherence patterns; and (3) maintain consistency with conventional network meta-analysis methodology. We recognize this approach has limitations: it may miss important information about the temporal trajectory of treatment effects, including early response patterns or potential waning of efficacy over time. Future research employing component network meta-analysis or multivariate meta-analysis techniques could address this limitation by simultaneously modeling outcomes across multiple time points.

Risk of bias assessment

Risk of bias was assessed using the revised Cochrane Risk of Bias tool (RoB 2) [16]. Each study was evaluated across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of reported results. Overall risk of bias was categorized as low, some concerns, or high risk.

Statistical analysis

Network meta-analysis was conducted using a Bayesian framework implemented via Markov Chain Monte Carlo (MCMC) simulation [17].This study employed a Bayesian framework for network meta-analysis; therefore, 95% credible intervals (CrI) are reported throughout. Unlike frequentist 95% confidence intervals (CI), which indicate that 95% of similarly constructed intervals would contain the true parameter value, Bayesian 95% credible intervals indicate a 95% probability that the true parameter lies within the interval, given the observed data and prior assumptions. For readers more familiar with frequentist interpretation, Bayesian credible intervals with non-informative priors (as used in this study) are numerically similar to frequentist confidence intervals and can be interpreted analogously for practical purposes.

Non-informative prior distributions were specified for all parameters: normal distributions with mean 0 and variance 10,000 for treatment effect parameters, and uniform distributions (0, 5) for between-study standard deviation. Three MCMC chains were run with 100,000 iterations each, discarding the first 50,000 as burn-in, with convergence assessed via trace plots, Brooks-Gelman-Rubin diagnostic (potential scale reduction factor < 1.05), and autocorrelation plots.

Both fixed-effect and random-effects models were fitted. Model selection was based on the deviance information criterion (DIC), with lower DIC indicating better model fit (difference > 3 considered meaningful). The random-effects model was preferred when between-study heterogeneity was substantial (τ² > 0).

Heterogeneity was quantified using the I² statistic, interpreted as low (< 25%), moderate (25–75%), or high (> 75%), and the between-study variance (τ²). Prediction intervals were calculated to illustrate the expected range of treatment effects in future similar studies.

For continuous outcomes (AHI, ESS), mean differences and 95% credible intervals were calculated. For binary outcomes (treatment adherence), odds ratios and 95% credible intervals were estimated. Network connectivity was assessed through visual inspection of network plots to ensure all treatments were connected either directly or indirectly, forming a single connected network without isolated nodes. Network consistency (coherence between direct and indirect evidence) was evaluated using: (1) the node-splitting method, which separates direct and indirect evidence for each comparison and tests for disagreement using a z-test (P < 0.05 indicating significant inconsistency); (2) the design-by-treatment interaction model for global inconsistency assessment [14].

Publication bias and small-study effects were assessed using comparison-adjusted funnel plots, where study-specific effect sizes were plotted against standard errors, centered around comparison-specific pooled effects. Asymmetry was evaluated visually and tested using Egger’s regression test (P < 0.10 suggesting potential bias). Asymmetric funnel plots may indicate publication bias (preferential publication of positive results) but may also reflect genuine heterogeneity or methodological differences between smaller and larger studies.

Treatment rankings were generated using the surface under the cumulative ranking curve (SUCRA) values.

All analyses were performed using R software with the “netmeta” and “gemtc” packages [18]. Statistical significance was defined as 95% credible intervals not crossing zero for continuous outcomes and not crossing one for odds ratios.

Results

Characteristics of included studies

A total of 18 randomized controlled trials involving 1,520 participants were included in this network meta-analysis [19–36], which see in Table 1. These studies were published between 1999 and 2023, with sample sizes ranging from 26 to 240 participants. The mean age reported in individual studies ranged from 43.4 years (Jokic et al. [19]) to 55.7 years (Andrén et al. [27]), with a weighted overall mean age of 49.8 years across all participants, and with male participants predominating in all studies (proportion ranging from 65.8% to 100% across individual studies; weighted overall proportion: 78.3% male). Body mass index (BMI) varied from 24.7 to 31.0 kg/m², indicating predominantly overweight to mild obesity. Baseline apnea-hypopnea index (AHI) ranged from 13 to 62.3 events per hour, representing mild to severe OSA severity.

Table 1.

Summary of the baseline characteristics of the included clinical trials

Authors and year	Design	Sample size	Intervention	Location and name of device	Control	Follow-up/wash-out
Jokic et al., 1999 [19]	Randomized single blind crossover trial	n = 13	PT	Back, backpack with tennis ball	CPAP	2 weeks/no wash-out
Barnes et al., 2004 [20]	Randomized three-way crossover trial	n = 114	CPAP, MAS	Oral; MAS	Placebo tablet	3 months/no wash-out specified (1-week washout between periods)
Gotsopoulos et al., 2004 [21]	Randomized controlled crossover trial	n = 61	PT	Oral; MAS	Control oral appliance	4 weeks/no wash-out
Bardwell et al., 2007 [22]	Randomized placebo-controlled study	n = 38	CPAP, oxygen	nocturnal oxygen supplementation	Placebo CPAP	2 weeks/no wash-out
Skinner et al., 2008 [23]	Randomized crossover trial	n = 20	PT	Chest; TASB (thoracic anti-supine band)	nCPAP	1 month/no wash-out
Permut et al., 2010 [24]	Randomized trial	n = 38	PT	Neck; positional device (PD)	CPAP	1 night intervention/no wash-out
Kohler et al., 2011 [25]	Randomized controlled trial	n = 41	CPAP withdrawal	-; subtherapeutic CPAP	Continued CPAP	2 weeks/no wash-out
van Maanen et al., 2012 [26]	Crossover (ON/OFF)	n = 30	PT	Neck; vibration device (prototype)	Inactive treatment (device OFF)	1 night intervention/no wash-out
Andrén et al., 2013 [27]	Randomized controlled trial	n = 72	PT	Oral; OA (oral appliance with mandibular advancement)	OA without advancement	3 months/no wash-out
Phillips et al., 2013 [28]	Randomized crossover trial	n = 126	CPAP	Oral; MAD	MAD	1 month/no wash-out
Dieltjens et al., 2015 [29]	Randomized crossover trial	n = 20	PT + MAD	Chest; TASB (thoracic anti-supine band)	MAD alone	2 consecutive nights/no wash-out
Jackson et al., 2015 [30]	Randomized controlled parallel trial	n = 86	PT	Back; position modification device	Sleep hygiene	4 weeks/no wash-out
Benoist et al., 2017 [31]	Randomized controlled trial	n = 145	PT	Neck; Sleep Position Trainer (SPT)	OAT	3 months/no wash-out
Diaféria et al., 2017 [32]	Randomized controlled trial	n = 100	Myofunctional therapy + CPAP	Oral; myofunctional exercises	Placebo myofunctional therapy	3 months/3 weeks wash-out
Laub et al., 2017 [33]	Randomized controlled trial	n = 101	PT	Neck; Sleep Position Trainer (SPT)	Sleep hygiene	2 months/no wash-out
de Ruiter et al., 2018 [34]	Randomized controlled trial	n = 99	PT	Neck; Sleep Position Trainer (SPT)	OAT	12 months/no wash-out
Suzuki et al., 2021 [35]	Randomized parallel trial	n = 60	PT	Neck; PTD (position therapy device)	OAT	1 night intervention/no wash-out
Huang et al., 2023 [36]	Randomized multi-arm parallel trial	n = 60	SPT + OAT	Eye; eye mask SPT	SPT alone, OAT alone	6 months/no wash-out

APAP auto-adjusting positive airway pressure, MAD mandibular advancement device, n sample size, PSG polysomnography, PT positional therapy, RCT randomized controlled trials, TBT tennis-ball technique, OAT oral appliance therapy, CPAP continuous positive airway pressure, SPT sleep position therapy, SOT SPT combined with OAT

Follow-up duration varied considerably, from single-night assessments to 12-month follow-up periods. Most studies (n = 12, 63.2%) had follow-up periods of 3 months or less, while seven studies (36.8%) examined longer-term outcomes extending beyond 3 months. The studies were conducted across multiple countries, with the highest representation from the Netherlands (n = 4), followed by the United States (n = 3) and Australia (n = 2) .

Risk of bias assessment

The risk of bias assessment using the RoB2 tool revealed that most included studies had moderate methodological quality. Overall, 2 studies (11.11%) were classified as low risk of bias, 14 studies (77.78%) had some concerns, and 2 studies (11.11%) were classified as high risk of bias (Fig. 2).

Fig. 2.

Risk of bias assessment for included studies using the RoB 2 tool. Evaluates five domains plus overall risk: low (green), some concerns (yellow), high (red). Studies listed by first author and year

In the randomization process domain, most studies (n = 12, 66.67%) demonstrated adequate sequence generation and allocation concealment, resulting in low risk of bias. However, 6 studies (33.37%) showed some concerns primarily due to insufficient reporting of randomization methods.

The domain of deviations from intended interventions presented the greatest challenge, with 16 studies (88.89%) showing some concerns or high risk of bias. This was primarily attributed to the inherent difficulty of implementing effective blinding for physical interventions such as positional therapy devices and oral appliances. Only 2 studies (11.11%) achieved low risk in this domain through successful implementation of sham interventions [22, 25].

For missing outcome data, 8 studies (44.44%) demonstrated low risk of bias with minimal dropout rates and appropriate handling of missing data. 10 studies (55.56%) showed some concerns or high risk, primarily due to substantial dropout rates in long-term follow-up studies without adequate imputation methods.

Measurement of outcomes showed consistently low risk across all studies (n = 18, 100%), as polysomnography represents an objective and standardized assessment method.

Selection of reported results demonstrated low risk in 13 studies (72.22%), with most studies reporting all pre-specified outcomes. Five studies (27.78%) showed some concerns due to selective reporting or insufficient protocol information.

Apnea-hypopnea index reduction

Network meta-analysis of AHI reduction included 17 studies with 1,471 participants. One study (Diaferia et al. [32]) was excluded from the AHI network analysis because it reported AHI as median and interquartile range rather than mean and standard deviation, and the data distribution was significantly skewed, precluding reliable conversion. This study was retained in other analyses where appropriate data were available. The interventions analyzed included positional therapy (PT), oral appliance therapy (OA), positive airway pressure (PAP), and control/no treatment (Fig. 3).

Fig. 3.

Network evidence plots for treatment comparisons. Depicts direct comparisons across outcomes: A AHI (all studies); C ESS (all studies); B and D show subgroup analyses for position-dependent OSA, defined as supine AHI ≥ 2 times non-supine AHI (criteria varied slightly across studies; see Methods for details). These subgroup analyses are exploratory; E SpO2. Node size by study count; edge thickness by direct evidence. Abbreviations: OA, oral appliance; PAP, positive airway pressure; PT, positional therapy; NOS, nocturnal oxygen supplementation

PAP therapy demonstrated the greatest efficacy in reducing AHI (mean difference: -17.39 events/hour, 95% CI: -23.13 to -12.74) compared to control. Oral appliance therapy showed moderate effectiveness (mean difference: -10.76 events/hour, 95% CI: -15.55 to -6.44), while positional therapy demonstrated smaller but significant improvements (mean difference: -10.05 events/hour, 95% CI: -14.62 to -5.89) (Fig. 4).

Fig. 4.

Forest plots of network meta-analysis results. Mean differences with 95% credible intervals are shown. Red circles indicate statistically significant differences (95% CrI excludes zero). Blue squares indicate inconclusive comparisons due to imprecision (wide 95% CrI crossing zero), where clinical equivalence cannot be assumed. Vertical dashed line represents no effect

Direct comparisons revealed that PAP was significantly more effective than oral appliance therapy (mean difference: -6.59 events/hour, 95% CI: -11.65 to -2.39) and positional therapy (mean difference: -7.28 events/hour, 95% CI: -11.98 to -3.65). Oral appliance therapy showed comparable effectiveness to positional therapy (mean difference: 0.70 events/hour, 95% CI: -2.89 to 4.45).

Treatment ranking showed PAP with 100% probability of being the most effective treatment, followed by oral appliance therapy (66% probability of ranking second) and positional therapy (66% probability of ranking third). Statistical heterogeneity was moderate (I² = 52%, P = 0.03), indicating some variability between studies. Assessment of publication bias using comparison-adjusted funnel plots and Egger’s regression test revealed mild asymmetry for AHI outcomes (Egger’s test P = 0.08). Visual inspection of Fig. 5A shows that smaller studies (higher standard errors, upper portion of plot) tended to report larger treatment effects favoring active interventions compared to larger studies. This pattern may reflect: (1) publication bias, where smaller studies with null or negative findings were less likely to be published; (2) genuine heterogeneity, where smaller studies enrolled more selected populations likely to respond to treatment; or (3) methodological differences, where smaller studies had less rigorous protocols. Given the borderline statistical significance (P = 0.08), we cannot definitively confirm or exclude publication bias. If publication bias is present, our pooled estimates may overestimate true treatment effects. Sensitivity analyses excluding the smallest studies (n < 30) did not substantially alter conclusions (PAP vs. control AHI MD: -16.82, 95% CrI: -22.45 to -11.28), providing some reassurance regarding robustness. For ESS outcomes, the funnel plot (Fig. 5B) showed no significant asymmetry (Egger’s test P = 0.31). (Fig. 5A). Several comparisons yielded inconclusive results due to imprecision (wide credible intervals crossing zero): oral appliance therapy versus positional therapy for AHI (MD: 0.70, 95% CrI: -2.89 to 4.45), positional therapy versus control for ESS (MD: -1.36, 95% CrI: -2.75 to 0.07), and all comparisons for oxygen saturation except PAP versus control. These inconclusive findings reflect limited direct evidence and should be interpreted as indicating clinical equipoise rather than equivalence.

Fig. 5.

Funnel plots for publication bias in network meta-analysis. Comparison-adjusted plots: A AHI; B ESS. Points represent study comparisons; dashed line: expected shape without bias; solid line: regression. No asymmetry detected

Epworth sleepiness scale improvement

ESS data were available from 11 studies involving 984 participants. PAP therapy showed the greatest improvement in daytime sleepiness (mean difference: -2.30 points, 95% CI: -3.60 to -1.15) compared to control. Oral appliance therapy demonstrated moderate ESS improvement (mean difference: -2.13 points, 95% CI: -3.46 to -0.86), while positional therapy showed smaller improvements (mean difference: -1.36 points, 95% CI: -2.75 to 0.07).

Direct comparisons between treatments showed minimal differences, with PAP versus oral appliance therapy (mean difference: -0.17 points, 95% CI: -1.48 to 1.06), PAP versus positional therapy (mean difference: -0.93 points, 95% CI: -2.52 to 0.53), and oral appliance therapy versus positional therapy (mean difference: -0.76 points, 95% CI: -2.14 to 0.56).

Treatment ranking showed PAP with 59% probability of being the most effective treatment, oral appliance therapy with 35% probability of ranking first and 53% probability of ranking second, and positional therapy with 78% probability of ranking third. Statistical heterogeneity was mild (I² = 35%, P = 0.15), indicating good consistency across studies. The funnel plot showed minimal asymmetry with Egger’s test non-significant (P = 0.31), suggesting no evidence of publication bias (Fig. 5B).

Oxygen saturation parameters

Analysis of oxygen desaturation index (ODI, defined as the number of ≥ 3% or ≥ 4% oxygen desaturation events per hour of sleep) included 4 studies with 476 participants. PAP therapy showed the greatest improvement in ODI (mean difference: -6.59 events/hour, 95% CrI: -11.55 to -1.53) compared to control, indicating approximately 6.6 fewer desaturation events per hour. Oral appliance therapy demonstrated moderate improvement (mean difference: -2.20 events/hour, 95% CrI: -7.12 to 2.87), while positional therapy showed comparable effects (mean difference: -6.64 events/hour, 95% CrI: -15.52 to 2.16). For mean oxygen saturation (SpO₂), limited data from 3 studies showed PAP therapy improved mean SpO₂ by 2.1% points (95% CrI: 0.8 to 3.4) compared to control. Clinically, an ODI reduction of > 5 events/hour and SpO₂ improvement of > 2% are generally considered meaningful thresholds associated with reduced cardiovascular risk.

Treatment ranking showed positional therapy with 39% probability of being most effective, followed by PAP (28% probability of ranking first), and NOS (31% probability of ranking first). The wide confidence intervals reflect the limited number of studies and heterogeneous patient populations. Statistical heterogeneity was moderate (I² = 43%, P = 0.11), indicating acceptable consistency among the limited number of studies. Due to the small number of studies, formal assessment of publication bias was not performed, though visual inspection of the funnel plot did not suggest obvious asymmetry.

Subgroup analysis: position-dependent obstructive sleep apnea

This subgroup analysis highlights phenotype-specific efficacy, a key advance in OSA management.

AHI reduction in position-dependent OSA

Subgroup analysis for position-dependent OSA included 8 studies with 496 participants. PAP therapy remained most effective (mean difference: -9.58 events/hour, 95% CI: -17.37 to -4.08) compared to control. Oral appliance therapy showed moderate effectiveness (mean difference: -8.99 events/hour, 95% CI: -18.66 to 0.33), while positional therapy demonstrated enhanced effectiveness in this subgroup (mean difference: -6.62 events/hour, 95% CI: -12.08 to -1.69).

Treatment ranking in position-dependent patients showed PAP with 59% probability of being most effective, oral appliance therapy with 40% probability of ranking first and 41% probability of ranking second, and positional therapy with 78% probability of ranking third. Statistical heterogeneity within the position-dependent subgroup was moderate (I² = 46%, P = 0.08), suggesting reasonable consistency despite different study populations and positional therapy devices.

ESS improvement in position-dependent OSA

ESS data for position-dependent OSA were available from 5 studies with 284 participants. PAP therapy showed the greatest improvement (mean difference: -2.73 points, 95% CI: -6.08 to 1.16), followed by positional therapy (mean difference: -1.55 points, 95% CI: -3.83 to 0.51) and oral appliance therapy (mean difference: -1.53 points, 95% CI: -4.73 to 1.81).

Treatment ranking in this subgroup showed PAP with 67% probability of being most effective, positional therapy with 50% probability of ranking second, and oral appliance therapy with 36% probability of ranking second. Heterogeneity was mild (I² = 32%, P = 0.21), indicating good consistency across studies in this specific patient population.

Detailed SUCRA values and treatment rankings for all outcomes are summarized in Table 2.

Table 2.

SUCRA values and treatment rankings for all outcomes

Treatment	AHI Reduction		ESS Improvement		ODI Reduction
	SUCRA (%)	Rank	SUCRA (%)	Rank	SUCRA (%)	Rank
PAP	99.8	1	72.5	1	68.3	2
Oral Appliance	58.2	2	65.8	2	45.2	3
Positional Therapy	52.1	3	35.4	3	78.5	1
NOS	-	-	-	-	38.0	4
Control	0.3	4	1.8	4	5.0	5

Treatment	Position-Dependent OSA: AHI		Position-Dependent OSA: ESS
	SUCRA (%)	Rank	SUCRA (%)	Rank
PAP	82.3	1	78.5	1
Oral Appliance	68.5	2	42.3	3
Positional Therapy	45.2	3	55.8	2
Control	4.0	4	8.5	4

Higher SUCRA values indicate higher probability of being the best treatment. SUCRA = 100% indicates the treatment is certainly the best, SUCRA = 0% indicates the treatment is certainly the worst. Rankings should be interpreted with caution when SUCRA values are similar, or confidence intervals overlap substantially

SUCRA surface under the cumulative ranking curve

Discussion

This network meta-analysis provides comprehensive evidence comparing the effectiveness of different non-surgical interventions for obstructive sleep apnea across 18 studies involving 1,520 participants. Our findings demonstrate a clear hierarchy of treatment effectiveness, with PAP therapy showing superior efficacy for AHI reduction, followed by oral appliance therapy and positional therapy showing comparable moderate effectiveness.

PAP therapy achieved the greatest AHI reduction compared to control, confirming its established role as the gold standard treatment for OSA [37]. Beyond statistical significance, clinical meaningfulness of observed treatment effects warrants consideration. For AHI, a reduction of ≥ 5 events/hour or ≥ 50% from baseline is generally considered clinically significant, associated with reduced cardiovascular risk and improved quality of life. By this criterion, PAP therapy (mean reduction: 17.39 events/hour vs. control) clearly exceeds clinical significance thresholds, while oral appliance therapy (10.76 events/hour) and positional therapy (10.05 events/hour) also demonstrate clinically meaningful effects.

For ESS, the minimal clinically important difference (MCID) is generally accepted as 2–3 points on the 24-point scale. PAP therapy achieved a mean ESS reduction of 2.30 points compared to control, which reaches but does not substantially exceed the lower bound of the MCID range. Oral appliance therapy (2.13 points) similarly approached the MCID threshold, while positional therapy (1.36 points) fell below this threshold. This suggests that while all active treatments improve objective sleep parameters, the subjective improvement in daytime sleepiness may be more modest and variable.

For oxygen desaturation index, reductions of > 5 events/hour are considered clinically relevant. Both PAP therapy (6.59 events/hour reduction) and positional therapy (6.64 events/hour reduction) met this threshold, though confidence intervals were wide due to limited data.

These findings underscore that statistical significance does not guarantee clinical meaningfulness, and treatment decisions should incorporate patient-reported outcomes and individual treatment goals alongside objective polysomnographic parameters. Sources of between-study and between-intervention heterogeneity merit explicit discussion. Within positional therapy, interventions ranged from low-cost, low-technology approaches (tennis ball technique) to sophisticated electronic devices (vibrating sleep position trainers), which may have differential efficacy, tolerability, and adherence profiles. The tennis ball technique, while inexpensive, is often poorly tolerated long-term due to discomfort, whereas electronic trainers provide more gentle feedback and may achieve better adherence. However, insufficient studies were available to conduct meaningful subgroup analyses by positional device type.

Similarly, oral appliances varied in design characteristics, including whether they were custom-fitted or prefabricated, titratable or fixed, and the degree of mandibular advancement achieved. Custom-fitted, titratable devices are generally more effective but also more expensive and require dental expertise for fitting and adjustment.

Study populations also contributed to heterogeneity. Baseline OSA severity ranged from mild (AHI 13 events/hour) to severe (AHI 62 events/hour), and treatment effects may differ across severity spectrum. Patient characteristics including age, BMI, and craniofacial anatomy may also modify treatment response but could not be explored due to study-level data availability.

These sources of heterogeneity may introduce inconsistency into the network and limit the precision of pooled estimates. Future individual patient data meta-analyses could better address these limitations by enabling exploration of treatment effect modifiers. However, the moderate effectiveness of oral appliance therapy and positional therapy suggests these alternatives may be viable options for selected patients, particularly given the clinical challenges associated with PAP adherence [38].

The comparable effectiveness between oral appliance therapy and positional therapy has important clinical implications. For patients who cannot tolerate PAP therapy, both alternatives offer meaningful therapeutic benefits. The choice between these modalities may depend on patient-specific factors such as positional dependency of OSA, dental suitability for oral appliances, and patient preferences [39].

Our ESS results revealed relatively modest differences between treatments, with PAP showing only slight superiority over oral appliance therapy and positional therapy. This suggests that while PAP remains most effective for objective sleep parameters, the subjective experience of daytime sleepiness improvement may be more comparable across treatments [40].

The subgroup analysis of position-dependent OSA patients revealed particularly relevant findings. In this population, positional therapy maintained its effectiveness while showing enhanced relative performance compared to the overall OSA population. This supports the concept of precision medicine approaches in OSA management, where treatment selection is tailored to specific patient phenotypes [41].

The treatment ranking in position-dependent patients showed oral appliance therapy with a 40% probability of being most effective, highlighting its potential as a first-line alternative to PAP in carefully selected patients. This finding aligns with current recommendations for considering patient-specific factors in treatment selection [9].Furthermore, given the observed heterogeneity in the definitions of position dependent OSA across trials, the findings from our subgroup analysis should be considered exploratory, and results should be interpreted with appropriate caution.

The oxygen saturation analysis revealed interesting patterns, with positional therapy showing the highest probability of being most effective for improving oxygen desaturation indices. This finding suggests that positional interventions may have particular benefits for preventing oxygen desaturations, which are closely linked to cardiovascular outcomes in OSA [42].

The inclusion of nocturnal oxygen supplementation in our network provides additional therapeutic options for patients with persistent hypoxemia despite other interventions. While not superior to mechanical treatments for AHI reduction, oxygen supplementation may offer specific benefits for patients with cardiovascular comorbidities [43].

Our treatment ranking analysis provides a framework for clinical decision-making. PAP therapy’s consistent ranking as most effective supports its continued use as first-line therapy. However, the competitive performance of oral appliance therapy and positional therapy provides evidence-based alternatives for patients unable to use PAP effectively [44]. Caution is warranted when interpreting SUCRA-based treatment rankings. While these rankings provide an intuitive hierarchy, they should not be overinterpreted when effect size differences are modest and credible intervals overlap substantially. For instance, the comparison between oral appliance therapy and positional therapy for AHI reduction yielded a small mean difference of 0.70 events/hour (95% CrI: -2.89 to 4.45), indicating that these treatments likely have comparable clinical effectiveness despite different ranking probabilities. Similarly, for ESS improvement, the differences between PAP, oral appliance, and positional therapy were modest (all within approximately 1 point), which may not represent clinically perceptible differences for patients. Therefore, clinical decision-making should integrate patient-specific factors, preferences, and practical considerations rather than relying solely on treatment rankings.

The relatively narrow confidence intervals for treatment rankings, particularly between oral appliance and positional therapy, suggest that patient preferences and individual circumstances should play important roles in treatment selection beyond purely efficacy considerations [38].

Several limitations should be acknowledged. The heterogeneity in study populations, OSA severity, and outcome measurement approaches (explicitly, the variations between 3% and 4% thresholds for defining the oxygen desaturation index) may limit the precision of treatment effect estimates and affect the comparability across trials. The risk of bias assessment revealed that most included studies (77.78%) were rated as having ‘some concerns,’ primarily in the domain of deviations from intended interventions. This is largely attributable to the inherent difficulty of blinding participants and personnel to physical interventions such as positional therapy devices and oral appliances. Unlike pharmacological trials where placebo controls can achieve effective blinding, sham devices for positional therapy or oral appliances are challenging to implement convincingly. This lack of blinding may introduce performance bias (patients expecting benefit may report greater subjective improvement) and detection bias (unblinded outcome assessors may be influenced by knowledge of treatment allocation). Consequently, the certainty of our conclusions, particularly for subjective outcomes such as ESS, should be interpreted with this methodological limitation in mind. The objective measurement of AHI through polysomnography provides some protection against these biases for our primary outcome. Several sources of clinical heterogeneity should be acknowledged when interpreting our findings. First, follow-up duration varied considerably across studies, ranging from single-night assessments to 12-month follow-up periods, which may capture different phases of treatment response and adherence. Second, baseline OSA severity differed substantially (AHI range: 13 to 62.3 events/hour), representing patients from mild to severe OSA, and treatment effects may vary across severity levels. Third, positional therapy interventions were heterogeneous, including simple techniques (tennis ball technique), vibrating positional devices, and sophisticated electronic sleep position trainers, which may have differential efficacy. Fourth, oral appliances varied in design and degree of mandibular advancement. These sources of clinical variability may limit the precision of pooled estimates and should be considered when applying these findings to individual patients. Future research should focus on identifying patient characteristics that predict differential treatment responses, enabling more personalized treatment selection [45].

The limited long-term follow-up data in many included studies restricts our understanding of treatment durability and adherence patterns over time. Longitudinal studies comparing these interventions with extended follow-up periods would provide valuable insights into real-world effectiveness [46].

These findings support a tiered approach to OSA management, with PAP therapy remaining the most effective option for AHI reduction, but with oral appliance therapy and positional therapy providing viable alternatives with comparable effectiveness profiles. Clinicians should consider patient-specific factors including OSA phenotype, comorbidities, and treatment preferences when selecting interventions.

The enhanced effectiveness of positional therapy in position-dependent OSA patients supports the importance of phenotyping approaches in treatment selection. Integration of sleep study findings with patient characteristics may optimize treatment outcomes while improving patient satisfaction and adherence [47].

Conclusion

PAP therapy remains the most effective treatment for OSA across objective outcomes. However, oral appliance therapy and positional therapy provide viable alternatives with comparable moderate effectiveness, particularly for position-dependent OSA patients. These conclusions should be interpreted with caution given the moderate statistical heterogeneity observed (I² = 32–52%), variations in study definitions (such as varying ODI criteria) and protocols, and the predominance of studies with ‘some concerns’ regarding risk of bias. Treatment selection should consider patient-specific factors including OSA phenotype, treatment tolerance, and individual preferences to optimize outcomes and adherence. Further high-quality, adequately powered trials with standardized outcome definitions and longer follow-up periods are needed to strengthen these recommendations.

Authors’ contributions

All authors contributed to study conception, design, data analysis, and manuscript preparation. All authors reviewed and approved the final manuscript.

Funding

This research received support from Huzhou City Science and Technology Plan Project, No.2023GY75.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.