Obstructive sleep apnea (OSA) is the most common sleep-disordered breathing, with a prevalence of 9–38% (Senaratna et al., 2017). One of the main risk factors for OSA is obesity (Tuomilehto et al., 2013). Obesity is a global epidemic with a steady increase in the incidence rates in recent years. Its incidence has tripled since 1975 according to the World Health Organization (WHO) (The Lancet Gastroenterology Hepatology, 2021, Congdon and Amugsi, 2022). The bidirectional relationship between OSA and obesity is complex. An increase in body mass index (BMI) and fat mass contributes to the deposition of fat tissues in the upper airway, including fat pads, uvula, and the base of the tongue. This deposition leads to an upper airway crowdedness, predisposing it to repetitive episodes of collapse during sleep (Turnbull et al., 2018; Yanari et al., 2022). Alternatively, intermittent hypoxia and sleep fragmentation can increase visceral adipose tissue (Harada et al., 2014; Zheng et al., 2022). It is estimated that 40% of people with obesity have notable OSA that warrants treatment (Wolk et al., 2003). Both OSA and obesity are independent risk factors for many co-morbid conditions (such as cardiovascular, cerebrovascular, metabolic, and even neoplastic diseases) (Wolk et al., 2003).

Positive airway pressure (PAP) therapy is considered the treatment of choice for OSA (Epstein et al., 2009). However, adherence to PAP therapy can range between 34 and 50% (Roecklein et al., 2010; Rotenberg et al., 2016). Furthermore, there is evidence of age and sex disparities in PAP therapy adherence that can range between 17 and 71% (Patel et al., 2021). Therefore, alternative therapies are available to patients with poor adherence to PAP therapy. One of these options is weight loss, either conservatively via lifestyle modification or surgically via bariatric surgery. The American Academy of Sleep Medicine (AASM) strongly recommends surgical consultation for bariatric surgery in all patients with OSA and class II/III obesity (BMI ≥ 35 kg/m²) (Kent et al., 2021). Similarly, the National Institute of Health (NIH) recommends referral to bariatric surgery in patients with morbid obesity and OSA regardless of PAP compliance (NHLBI, 2000; Kent et al., 2021). Weight loss surgery usually reduces the severity of OSA; however, a complete resolution is unusual (Greenburg et al., 2009; Wong et al., 2018).

Recently, two systematic reviews and meta-analyses (SRMA) were published to synthesize the evidence related to bariatric surgery's impact on OSA (Wong et al., 2018; Zhang et al., 2019). Given the publication of several additional studies since the last SRMA in 2019, we conducted and updated a systematic review of the literature for the quality of the evidence and then conducted a meta-analysis to synthesize the evidence to date.

The aim of this systematic review and meta-analysis is to address the following question:

Studies were selected according to the inclusion criteria outlined below:

Since the goal of this systematic review and meta-analysis is to study the impact of bariatric surgery on OSA in the absence of other treatment options, all studies using positive airway pressure therapy or non-invasive ventilation after bariatric surgery were excluded.

The literature search was conducted until 31 March 2023 using the following keywords: “sleep,” “study,” “apnea,” “obstructive,” “bariatric,” “surgery,” “sleeve gastrectomy,” “gastric bypass,” and “gastric banding”. There were no date limits to the search. The search was conducted in Ovid MEDLINE (R); Elsevier Embase (1947–2022); and Elsevier Scopus (1823–2022). Screening was conducted in two stages. In stage one, the authors (SM and AR) independently conducted an initial screening of the titles and abstracts. In stage two, the full text of the records included in stage one of the screening was obtained by authors (SM, DC, MK, SH, PR, and KH) to ensure they met eligibility criteria. Any disagreements were resolved by a discussion between the reviewers. The study selection process is illustrated in Figure 1.

The quality of included studies was assessed using the following tools:

Samples of these quality assessment tools are included as Supplementary Tables S1, S3.

Two authors (SM and AR) independently assessed the certainty of the evidence. Five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) were used to assess the certainty of the body of evidence as they are related to the studies that contributed data to the meta-analyses for the prespecified outcomes. The certainty of evidence was reported as high, moderate, low, or very low. We used the methods and recommendations described in Sections 8.5 and 8.7 and Chapters 11 and 12 of the Cochrane Handbook for Systematic Reviews of Interventions. We justified all decisions to downgrade or upgrade the certainty of studies using footnotes, and we provided comments to aid the reader's understanding of the results where necessary (Mudano et al., 2019).

Data were pooled for all studies that examined the effect of surgical weight loss on OSA severity and BMI. The pooled estimate of the mean difference was used as the primary outcome measure and 95% confidence intervals were calculated. A random effects model by DerSimonian and Laird was chosen to synthesize the data throughout because it allows within- and between-study variations, which are applicable to this meta-analysis that includes mostly observational cross-sectional studies with inherently more variability (Dersimonian and Laird, 1986).

Heterogeneity between studies was tested using both the I² statistic, where I²-values of 25, 50, and 75% were defined as mild, moderate, and high heterogeneity, respectively. Publication bias was assessed by visual inspection of funnel plots to examine possible asymmetry and by using the Egger regression asymmetric test (Bowden et al., 2015). To explain the heterogeneity between studies and to examine the influence of various factors, we performed a meta-regression. The following factors were studied: baseline AHI and BMI, age, study design (observational vs. randomized control trial), type of bariatric surgery, the continent where the study was done, study results used more than once, and duration between bariatric surgery and follow-up sleep study. Leave-one-out sensitivity analyses were performed to further explore the changes in our findings by iteratively removing one influential study at a time. All analyses were conducted in STATA software version 17.0 (Stata Corp., College Station, TX). The significance was set at two-tailed p-values of 0.05.

A total of 9,797 articles were identified by searching the different databases. Following the removal of duplicate articles (n = 473), articles unreadable by automation tools (n = 4,306), and articles excluded for other reasons (n = 3,912), a total of 1,106 articles were screened using the title and the abstract. The details of the included/excluded articles were added to the PRISMA flow diagram. This yielded 69 articles that were fully reviewed. After applying the inclusion/exclusion criteria, a total of 30 articles were deemed eligible for both qualitative and quantitative (meta-analysis) synthesis. Two of these articles were RCTs (Aguiar et al., 2014; Bakker et al., 2018). All the non-randomized controlled trials except two studies [one case-control (Busetto et al., 2005) and one cross-sectional (Lage-Hansen et al., 2018)] were observational (i.e., 26 studies). Out of these 26 observational studies, 19 were prospective (Pillar et al., 1994; Valencia-Flores et al., 2004; Pallayova et al., 2011; Krieger et al., 2012; Bakker et al., 2013, 2014; Fredheim et al., 2013; Bae et al., 2014; Karaköse et al., 2014; Zou et al., 2015; Del Genio et al., 2016; Jiao et al., 2016; Shaarawy et al., 2016; Xu et al., 2016; Peromaa-Haavisto et al., 2017; Tirado et al., 2017; Al-Jumaily et al., 2018; Chierakul et al., 2020; Yilmaz Kara et al., 2021) and 7 retrospective (Peiser et al., 1984; Fritscher et al., 2007; Morong et al., 2014; Obeidat et al., 2020; Song et al., 2021; Wu et al., 2022; Yanari et al., 2022) in study design. Figure 1, the PRISMA flow diagram, demonstrates the flow of records from the initial search through the selection process, the number of records included, and the reasons for exclusion. Tables 1, 2 illustrate the characteristics of the included studies. The quality assessment of the included studies is shown in Supplementary Tables S2 (“non-RCT” and S4 “RCT”). The quality of evidence across all studies included was very low (Supplementary Table S5). This is related to several factors, including a high risk of bias and publication bias.

Geographically, out of 30 studies, 14 were conducted in Asia, 7 in North America, 6 in Europe, and 3 in South America. A total of 1,369 participants were included in the analysis. They were middle-aged (the range of mean ages was 35–51 years) and primarily women (66% with a range of 0–100% in each study). Individual study sample sizes ranged from 10 to 162 participants, with most enrolling between 23 and 56 participants. The participants underwent laparoscopic SG (5 studies), RYGB (13 studies), and AGB (3 studies). Seven studies included patients who had more than one type of surgery, and two studies did not specify the type of surgery and mentioned only bariatric surgery. Most studies included patients who underwent standard in-laboratory polysomnography (27 studies), and three studies included patients who underwent HSAT. The following parameters were recorded: AHI, ODI, mean SpO₂, TST < 90, and L SpO₂. Of the 30 studies, 22 used the American Academy of Sleep Medicine scoring criteria to score obstructive hypopneas, and three studies did not include details on scoring criteria (Table 2).

The heterogeneity (I²) values were high for all parameters obtained except T-90. Accordingly, a sensitivity analysis was conducted to assess potential sources of heterogeneity. Potential sources evaluated included the study design, the study population (i.e., the geographical location by continent), the mean age, the baseline BMI, the baseline AHI, the follow-up after surgery, the type of bariatric surgery, and the multiple use of the same study. A meta-regression analysis was conducted to reduce or resolve the heterogeneity.

There was evidence of publication bias in the following parameters: BMI and AHI, ODI, and mean SpO₂. This is most likely related to the following potential reasons. First, when we searched the databases we excluded non-English language articles, which can contribute to publication bias. Second, the funnel plots showed an asymmetrical distribution of the included studies across the midline, and finally, the Egger's test was statistically significant (BMI, p = 0.0001; AHI, p = 0.0008; ODI, p = 0.03; mean SpO₂, p = 0.006), supporting the assumption that publication bias is very likely. There was no publication bias in T-90, p = 0.59 and L SpO₂, p = 0.67. The funnel plots of these parameters are illustrated in Supplementary Figures S1A–F.