The association between obstructive sleep apnea and metabolic syndrome: a systematic review and meta-analysis

Obstructive sleep apnea (OSA) is a condition in which there is repetitive obstruction of the upper airway during sleep, resulting in hypopnea (reduced airflow during sleep) or apnea (complete cessation of airflow during sleep) [1, 2]. Patients with OSA may experience symptoms, including loud souring, frequent arousals, sleep fragmentation, and daytime sleepiness, which characterize obstructive sleep apnea syndrome [2, 3]. Studies have shown that OSA is associated with increased risks of hypertension, stroke, type 2 diabetes mellitus, and other cardiovascular diseases [4‐7]. Approximately 4 % of men and 2 % of women in the general population were considered as having OSA in 1993 [8, 9]. With the increasing epidemic of obesity, the prevalence of OSA among adults is estimated to be up to 17 % [10].

Metabolic syndrome (MS) is a cluster of metabolic risk factors for diabetes and cardiovascular diseases, including central obesity, hypertension, hyperglycemia, and dyslipidemia [11‐13]. The prevalence of MS varies between 20 and 40 % worldwide and tends to increase in parallel with the epidemic of obesity [14, 15]. In the past decade, numerous studies have indicated a relationship between OSA and MS, but the results remain debatable [16]. Coughlin et al. found that OSA was independently associated with an increased risk of MS, irrespective of age and body mass index (BMI) [17]. However, Papanas et al. showed that the strong association between the presence of OSA and MS became non-significant when BMI was considered [18].

Based on these findings, and the fact that no meta-analysis has been conducted on this relationship, we aimed to perform a systematic review and meta-analysis to evaluate the association between OSA and MS.

Overall, 5648 references were identified and 20 studies were finally included in the review [17, 18, 26‐43]. All of the included studies were reviewed by full text. Of the included studies, 15 were cross-sectional [18, 26‐33, 35‐37, 40, 42, 43], five were case–control [17, 34, 38, 39, 41], and none were cohort studies. In total, 2456 patients with OSA and 1705 subjects with no OSA in cross-sectional studies, together with 1156 OSA patients and 404 controls in case–control studies, were included in the meta-analysis for the association of OSA with MS (Fig. 1).

Fourteen studies were hospital-based studies [17, 18, 28‐34, 37‐41], five were community-based studies [26, 35, 36, 42, 43], and one was a population-based study [27]. All of the studies used the AHI based on an overnight PSG to diagnose OSA, whereas the AHI thresholds differed between studies. Two studies [44, 45], of which one used the minimal patient contact sleep diagnosis system and another used a pulse-oximeter to assess OSA, were excluded. Almost all of the studies adopted the National Cholesterol Education Program Adult Treatment Panel III criteria to define MS, except for one study [39] in which the International Diabetes Federation criteria were used (Table 1).

The NOS [22] was used to evaluate the methodological quality of the included studies (Table 2). The mean NOS score for the cross-sectional and case–control studies was 6.7 and 6.8, respectively, indicating that the overall methodological quality was generally good. Fourteen hospital-based studies had relatively low scores in the selection category, mainly because the controls who were derived from a hospitalized population (not the same community) may have resulted in potential selection bias. In the comparability category, obesity was the most important factor, but was adjusted for in only seven studies [17, 26, 32, 34, 35, 39, 43]. In the exposure category, all of the studies had perfect scores for the ideal and same method of ascertainment for cases and controls. Venous blood was obtained to assess MS in the fasting state in the morning after performing PSG in almost all of the prospective studies. Therefore, we were cautious that they had nearly a 100 % response rate for both groups. Two studies did not receive stars in the item of non-response rate [26, 28]. One of these studies was a retrospective study in which the authors mentioned that 22 patients were excluded for insufficient data, but they did not report the non-response rate in each group separately [28]. The other study was a community-based study in which a subset of participants from the Wisconsin Sleep Cohort Study were invited to measure metabolic parameters; therefore, the non-response rate for each group should have been described [26].

As mentioned above, the adjusted ORs were preferred for the meta-analysis. For the remaining studies without adjusted ORs, the crude or calculated ORs were adopted for the meta-analysis. Sensitivity analysis was then performed based on the exclusion of studies that provided unadjusted data only. Two studies separately reported the adjusted ORs for men and women [39] or for mild and moderate-to-severe OSA [26]. The results were initially pooled within the study and then a single estimate was included in the meta-analysis. Overall, for cross-sectional studies, the pooled OR of MS in individuals with OSA was 2.87 (95 % CI: 2.41–3.42), with unimportant between-study heterogeneity of the analysis (P = 0.23, I ²  = 20 %). Subgroup meta-analyses based on the criteria of OSA (AHI > 5 events/h or > 15 events/h) showed similar pooled ORs (2.89 [95 % CI: 2.39–3.50] and 2.73 [95 % CI: 1.74–4.28], respectively). There was no evidence of substantial heterogeneity in the subgroup analyses (P = 0.13, I ²  = 31 % and P = 0.91, I ²  = 0 %, respectively) (Fig. 2). For case–control studies, the pooled OR of MS in individuals with OSA was 2.56 (95 % CI: 1.98–3.31), with moderate between-study heterogeneity of the analysis (P = 0.19, I ²  = 35 %). Subgroup meta-analyses based on the criteria of OSA (AHI > 5 events/h or > 10 events/h [15 events/h]) showed similar pooled ORs (2.11 [95 % CI: 1.50–2.96] and 3.29 [95 % CI: 2.23–4.85], respectively). The I ² was 0 % (P = 0.96) and 39 % (P = 0.19), respectively (Fig. 3). In addition, mixed effect, multilevel meta-analyses showed no effects of mean BMI, mean patient’s age, percentage of male sex, adjusted (yes/no), and publication year on the overall results (P > 0.05 for all analyses).

Only some of the cross-sectional studies and one case–control study provided data on severity of OSA related to MS. Therefore, meta-analysis on the relationship between MS and the severity of OSA was conducted in all of the eligible studies together. The pooled crude ORs of MS in individuals with mild and moderate-to-severe OSA were 2.39 (95 % CI: 1.65–3.46) and 3.45 (95 % CI: 2.33–5.12), respectively. There was substantial heterogeneity in the meta-analyses (P = 0.02, I ²  = 53 % and P = 0.004, I ²  = 63 %, respectively) (Fig. 4).

To verify the robustness of the results, as well as the potential sources of heterogeneity, subgroup analyses were performed. We examined exclusion of studies with obesity as the most important unadjusted factor (Model 1), those where the Adult Treatment Panel III criteria were not used to define MS (Model 2), and those with a special population (children, young men, morbidly obese individuals, and non-obese people) (Model 3). These subgroup analyses only resulted in a marginal change in the pooled ORs (Table 3).

No evidence of publication bias was detected by using Begg’s test (z = 1.18, P = 0.232) and Egger’s test (z = 1.18, P = 0.232). Visual inspection of the funnel plots showed that they were symmetrical (Fig. 5).

We also performed meta-analyses that included two conference reports that were not subsequently published [46, 47]. The results were similar to those that did not include conference reports, although there was substantial heterogeneity in the meta-analyses. The Galbraith plot showed that the conference reports greatly contributed to between-group heterogeneity (Additional file 2: Figure S1, Additional file 3: Figure S2, and Additional file 4: Figure S3)

Our meta-analysis, which included 15 cross-sectional (2456 patients with OSA and 1705 subjects with no OSA) and five case–control studies (1156 OSA patients and 404 controls), identified a significant relationship between OSA and MS. This meta-analysis also showed that the pooled ORs of MS in individuals with OSA for the cross-sectional and case–control studies were 2.87 (95 % CI: 2.41–3.42) and 2.56 (95 % CI: 1.98–3.31), respectively. Although there were substantial differences among studies (e.g., the diagnosis of OSA and definition of MS), subgroup analyses confirmed the robust relationship.

Obesity is strongly linked to MS and is also a well-known risk factor for OSA [48]. With the escalating prevalence of obesity, the association between OSA and MS has been increasingly recognized over the past few years [16]. Several possible mechanisms have been suggested to explain the biological plausibility of OSA, independent of obesity, increasing the risk of MS [16]. These mechanisms may be related to intermittent hypoxia (IH), oxidative stress, cytokines, and selective activation of systematic inflammation [49, 50]. Recurrent obstructive events result in IH in OSA. Repaid reoxygenation of transiently ischemic tissues can damage tissues and release reactive oxygen, the culprit of oxidative stress [51]. IH and resultant oxidative stress have been proposed as a pathogenetic pathway between OSA and disturbance of glucose homeostasis [51], insulin resistance [52], hypercholesterolemia [53], and hyperlipidemia [54]. Similarly, inflammatory cytokines (e.g., tumor necrosis factor-α and interleukin-6) that are triggered by IH and sleep fragmentation have been postulated as a putative mechanism of MS. Inflammatory cytokines may also impair insulin action in peripheral tissues and increase insulin resistance, dyslipidemia, and hypertension in OSA [55]. In conclusion, OSA is linked with metabolic dysregulation in the complex human biological system. IH and sleep fragmentation are suggested to trigger an array of downstream effects (i.e., sympathetic activation, neurohumeral changes, inflammation, and oxidative stress), which are pathophysiological cascades that are common to the pathogenesis of cardiometabolic diseases [16]. However, a specific weakness of our study must be mentioned with respect to pathophysiological aspects, which were not the main focus of the review.

The shared relationship of OSA and MS with obesity should be addressed. Obesity is a major risk factor for OSA because it directly or indirectly contributes to upper airway narrowing during sleep. An example of this situation is by promoting enlargement of soft tissue structures within and surrounding the airway, increasing abdominal fat mass, and recumbent posture [56]. Additionally, obesity, in particular abdominal obesity, is thought to be connected with MS [12]. Our previous study reported that the prevalence of MS in participants with obesity was significantly higher than that in the general population [57]. Therefore, obesity may play a major potential confounding role and should be adjusted in the relationship between OSA and MS. Our meta-analysis showed that obesity, as well as age and sex, could be adjusted as confounding factors in only 12 of 20 included studies. However, we found that all of the studies evaluated obesity only by measuring and calculating BMI, and other accurate imaging examinations (e.g., computed tomography, magnetic resonance imaging) were not performed. We also found that only four studies further adjusted for smoking and alcohol drinking [17, 33‐35]. More importantly, none of the studies adjusted for other important risk factors of MS (e.g., educational level, family or personal income, and family history of diabetes or hypertension) [58], which would be expected to reduce the positive association between OSA and MS towards the null value.

We observed a substantial difference among the included studies. An example of this difference was that the normal limit of AHI for diagnosing OSA varied between studies. Most studies used a threshold of over five events/h, and some used 10 events/h [28, 34, 41] or 15 events/h [17, 32, 36]. Furthermore, because MS is a new concept with debatable criteria over a decade [11], two definitions with individual waist circumference criteria to define abdominal obesity in different ethnics were adopted in our meta-analysis. Additionally, adjusted ORs were preferred in the meta-analysis. However, because of the fact that the relationship between OSA and MS was not the primary outcome in some studies, the adjusted ORs were not provided. Therefore, we could only use raw data to calculate the crude ORs in some included studies. In addition, although Papanas et al. [18] suggested that a strong association became non-significant when confounders were considered, the adjusted ORs were not provided and the crude ORs, which exhibited a significant difference, were thus adopted. However, the results of the subgroup analyses confirmed the robustness of association between OSA and MS.

Our study has several potential limitations. First, the included studies were cross-sectional or case–control studies, and none were cohort studies, which could not prove a cause-effect relationship between OSA and MS. However, a beneficial effect of continuous positive airway pressure (CPAP) therapy on MS in patients with OSA could provide more information about the causal relationship [59]. Future randomized controlled trials are required to investigate causality. Second, although the adjusted ORs from each included study were preferred in the meta-analysis, we could not exclude the possibility that our results may have been influenced because adjustment for potential confounders differed in each study. Third, selective outcome reporting remains a possibility. Some studies were speculated to be related to the topic of OSA and MS, but adjusted ORs or raw data were not obtained, despite attempting to contact authors for additional published or unpublished data. Fourth, the two methods that we adopted for assessing the severity of OSA might not be appropriate because they were not formally tested. This could be the possible reason for the large heterogeneity in the meta-analysis of severity of OSA on the risk of MS. Finally, the search was limited to English-language studies only, which had the potential of not including studies in other languages.

In summary, this meta-analysis of cross-sectional and case–control studies confirms a positive association between OSA and MS, although causality between these two factors has not been demonstrated yet. OSA and the MS are important cardiovascular risk factors and they may act synergistically. With the rapidly growing health problem of OSA and MS, further population- or community-based cohort and randomized controlled studies with adequate adjustment for multiple major confounding factors are required.