Tumor necrosis factor alpha is a promising circulating biomarker for the development of obstructive sleep apnea syndrome: a meta-analysis

Obstructive sleep apnea syndrome (OSAS) is a chronic inflammatory disorder. The relationship between tumor necrosis factor alpha (TNF-alpha) and OSAS has been widely evaluated, but the results thus far remain inconclusive. We thereby decided to quantify the changes of TNF-alpha between OSAS patients and controls by a meta-analysis. This study complies with the MOOSE guidelines. Two reviewers independently searched articles and abstracted relevant data. In total, 47 articles (59 studies) were analyzed, including 2857 OSAS patients and 2115 controls. Overall, OSAS patients had a significantly higher level of circulating TNF-alpha than controls (weighted mean difference [WMD]: 9.66 pg/mL, 95% confidence interval [CI]: 8.66 to 11.24, P<0.001), but with significant heterogeneity (I2: 99.7%). After adjusting for potential missing studies, the overall estimate was weakened but still significant (filled WMD: 2.63 pg/mL, 95% CI: 2.56 to 2.70, P<0.001). When studies were stratified by OSAS severity, the changes in circulating TNF-alpha between patients and controls increased gradually with the more severe grades of OSAS. In patients with mild, mild-to-moderate, moderate, moderate-to-severe and severe OSAS, circulating TNF-alpha was higher than respective controls by 0.99, 1.48. 7.79, 10.08 and 8.85 pg/mL, with significant heterogeneity (I2: 91.2%, 74.5%, 97.6%, 99.0% and 98.1%). In conclusion, our findings demonstrated that circulating TNF-alpha was significantly higher in OSAS patients than in controls, and this difference became more pronounced with the more severe grades of OSAS, indicating that TNF-alpha might be a promising circulating biomarker for the development of OSAS.


INTRODUCTION
Obstructive sleep apnea syndrome (OSAS) is a chronic inflammatory disorder featured by recurrent bouts of partial or complete upper airway obstruction during sleeping [1]. OSAS poses a major burden on individual and public health, as it respectively affects 10% and 17% of middle-aged (30-49 years old) and aged  years old) men, and 3% and 9% of middle-aged and aged women [2]. It is worth noting that affected individuals are more likely to suffer cardio-and cerebro-vascular diseases, such as hypertension, heart failure and stroke [3,4]. At present, continuous positive airway pressure (CPAP) ranks as the main treatment option for patients with moderate or severe OSAS, and it can assist in reducing systematic inflammation in the airways of OSAS patients [5]. Hence, understanding the inflammation process may offer a possible clue to understanding the molecular mechanisms behind the pathogenesis of OSAS.
Several lines of evidence from animal experiments and clinical investigations have indicated that the presence of OSAS is associated with the increased production of inflammatory mediators [6,7]. Tumor necrosis factor alpha (TNF-alpha) is a key modulator of systematic inflammation [8][9][10], and TNF inhibition has proven to ameliorate the progression of OSAS [11]. Moreover, some researchers have observed a significant high level of circulating TNF-alpha in OSAS patients vis-à-vis healthy individuals [12][13][14][15][16][17][18], whereas others did not [19,20]. The probable causes are multifaceted, relating to

RESULTS
After searching three public databases, a total of 171 articles written in English language were indexed. After reviewing the title and abstract of each article, 99 were excluded for definitive reasons. After reading the full text of the rest 72 potential articles, 25 were further excluded, leaving 47 qualified articles in this metaanalysis according to the preset inclusive criteria . Because 9 articles provided data by OSAS severity, 1 article by hypertension and 1 article by obesity, there were a total of 59 independent studies involving 2857 OSAS patients and 2115 controls in the final analysis. The baseline characteristics of 59 studies are summarized in Table 1 and Supplementray Table 1.
Of 59 qualified studies, 25 were from Asian countries, 21 from European countries, 5 from North American countries, 3 from South American countries, 3 from cross-continent countries, 1 respectively from Australia and Latin America. 13 studies involved only male individuals, and 7 studies involved underage individuals. Age was reportedly matched between patients and controls by 23 studies, and there were 35 studies involving individuals free of hypertension and diabetes mellitus. There were 51 and 8 cross-sectional and nested case-control studies, respectively. OSAS was diagnosed by polysomnography by 51 studies. As for OSAS severity, mild OSAS was reported by 6 studies, mild-to-moderate OSAS by 3 studies, moderate OSAS by 5 studies, moderate-to-severe OSAS by 14 studies and severe OSAS by 8 studies.
as high as 99.7% and the probability of Egger's test was 0.012. The filled funnel plot indicated that there were 11 missing studies with negative findings (Figure 2), and after adjusting for these missing studies, overall estimate was weakened but still significant (filled WMD: 2.63 pg/mL, 95% CI: 2.56 to 2.70, P < 0.001).  In the upper plot, the "md_tnf" in the y-axis is the mean difference of circulating TNF-alpha in pg/mL, and the "s.e. of: md_tnf" in the x-axis is the standard error of mean difference in circulating TNF-alpha. In the lower plot, the "theta" is the mean difference of circulating TNF-alpha in pg/mL, and the "s.e. of: theta" is the standard error of mean difference in circulating TNF-alpha.
Stratified analyses according to age, gender, country, hypertension, diabetes mellitus, research type, matched condition, diagnostic criteria of controls, diagnostic criteria of OSAS and OSAS grade are shown in Table 2.
In the following stratified analyses, only subgroups involving 3 or more studies were displayed. By country, OSAS patients vis-à-vis controls had remarkably high circulating TNF-alpha in China (WMD: 58.59 pg/mL, P < 0.001). When grouping studies by development, the changes in circulating TNF-alpha were strongly potentiated in developing countries (WMD: 17.17 pg/mL) than in developed countries (WMD: 2.37 pg/mL). Further by continent, the change was the highest in Asia (WMD: 29.84 pg/mL), followed by North America (WMD: 6.00 pg/mL) and Europe (WMD: 1.28 pg/mL). By research type, this change in cross-sectional case-control studies Abbreviations: WMD, weighted mean difference; 95% CI, 95% confidence interval; I 2 , inconsistency index.
(WMD: 10.41 pg/mL) was overwhelming relative to nested case-control studies (WMD: 5.10 pg/mL). When studies were stratified by OSAS severity, the changes in circulating TNF-alpha between patients and controls increased gradually with the more severe grades of OSAS. In patients with mild, mild-to-moderate, moderate, moderate-to-severe and severe OSAS, circulating TNFalpha was higher than respective controls by 0.99, 1.48. 7.79, 10.08 and 8.85 pg/mL. In spite of the above stratified analyses, there was no immediate improvement in between-study heterogeneity. A meta-regression analysis was hence conducted to see the impact of other confounding factors on the changes of circulating TNF-alpha between OSAS patients and controls. After regressing all possible confounders as mentioned in the Methods, only abdomen circumference and IL-6 were found to exert a significant impact on the changes of circulating TNF-alpha (abdomen circumference: P < 0.001 in patients and P = 0.026 in controls; IL-6: P = 0.001 in patients and P = 0.003 in controls). No significance was found for the other confounders (data not shown). In view of this significant finding, correlation analysis was conducted to test the relationship of circulating TNF-alpha with abdomen circumference and IL-6. The correlation of circulating TNF-alpha with abdomen circumference was marginal (P = 0.078), while the correlation with IL-6 was remarkably significant (P < 0.001).

DISCUSSION
On the basis of 59 studies and 4972 individuals, this meta-analysis aimed to quantify the changes of circulating TNF-alpha between OSAS patients and controls. Our results illustrated that circulating TNF-alpha was significantly higher in OSAS patients than in controls, and this difference became more pronounced with the more severe grades of OSAS, indicating that TNF-alpha might be a promising circulating biomarker for the development of OSAS.
There is strong evidence that TNF-alpha is a central regulator of inflammation, and its antagonists have proven to be efficacious in treating inflammatory diseases [59,60]. OSAS is a chronic inflammatory disorder, and its presence can lead to the increased production of some inflammatory mediators in circulation, including TNFalpha. An animal study found that the excessive sleepiness incurred by recurrent arousals during sleep might be due to the activation of TNF-alpha-depended inflammatory pathways [61,62]. In addition, expression data showed that TNF-alpha was highly expressed in the heaviest OSAS patients relative to the less obese OSAS patients and non-apneic snorers [63]. The association of circulating TNF-alpha with OSAS risk has been widely evaluated, while no consensus exists in up-to-date literature [19,[51][52][53][54]. Based on these observations, it is reasonable to postulate that circulating TNF-alpha might be a clinical useful indicator for predicting OSAS risk. To shed some light on this postulation, we comprehensively analyzed the results of 59 studies through a meta-analysis and aimed to derive a reliable estimate between circulating TNF-alpha and OSAS.
A previous meta-analysis of 19 studies by Nadeem et al demonstrated that OSAS patients had higher circulating TNF-alpha than controls by 1.03 pg/mL, and this difference was confused by obvious heterogeneity that remained unexplored in their study [64]. The present metaanalysis by pooling the results of 59 studies confirmed and strengthened this significant difference by deriving an unbiased estimate of 2.63 pg/mL for circulating TNFalpha in the trim-and-fill analysis. As with a majority of meta-analyses, we should be circumspect about the impact of between-study heterogeneity, as not every study's methodological and clinical aspects are identical [65]. In light of the differences in OSAS severity, research type, matched condition and so forth in the present metaanalysis, we can at least expound on some degree of heterogeneity, which accounted for part of conflicting findings in the literature. As it turns out, our stratified analyses demonstrated that the country, research type and OSAS severity might be possible sources of heterogeneity. It is worth mentioning that with the more severe grades of OSAS defined by AHI, circulating TNF-alpha was much higher in patients than in controls. Although the observational nature of all involved studies in this meta-analysis precluded the causal-effect exploration between circulating TNF-alpha and OSAS, our findings may provide indirect evidence that TNF-alpha might be a promising circulating biomarker for the development of OSAS. We concede that whether elevated circulating TNF-alpha is the cause or the effect of OSAS remains an open question. In the future, clinical trials are warranted to dissect this relation.
In spite of clear strengths including a large number of qualified studies and a comprehensive exploration on heterogeneity, it should be realized that there are several limitations to association studies included in this metaanalysis. First, selection bias might be possible given that only English articles were indexed. Although there was a significant probability of publication bias, the filled effect estimate after adjusting for missing studies was still significant in circulating TNF-alpha between OSAS patients and controls. Second, the results of this meta-analysis were based on 59 studies, while the total sample was not large enough. The power to reject the null hypothesis is very limited in some subgroup analyses. Third, between-study heterogeneity cannot be fully accounted for, in spite of a wide panel of stratified analyses conducted. It will be encouraging to explore the other sources of methodological and clinical aspects to mitigate heterogeneity. Moreover, this meta-analysis was undertaken with summary data, and to thoroughly account for heterogeneity one usually needs to perform a meta-analysis based on individual participant data, which are not always feasible. Fourth, the impact of obesity on the relationship between circulating TNF-alpha and OSAS cannot be solved due to the lack of necessary data, although it is increasingly recognized that obesity is an established risk factor for OSAS.
In sum, this meta-analysis of 59 studies and 4972 individuals demonstrated that circulating TNF-alpha was significantly higher in OSAS patients than in controls, and this difference became more pronounced with the more severe grades of OSAS, indicating that TNF-alpha might be a promising circulating biomarker for the development of OSAS. Our results, as are of consequence, deserve to be tested through relevant biological means and validated in large, well-designed prospective studies.

MATERIALS AND METHODS
This is a systematic meta-analysis on observational data, and its conduct complies with the guidelines enacted by the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group [66].
Using public databases of PubMed, Embase and Web of Science, articles that reported the changes of circulating TNF-alpha between OSAS patients and controls were indexed on November 3, 2016. Research content was confined to materials written in English language only. Included articles had to meet the following criteria: (i) OSAS as the clinical endpoint diagnosed by standard methods; (ii) case-control study design; (iii) availability of serum or plasma TNF-alpha levels expressed as mean or median value along with standard deviation or standard error or 95% confidence interval (95% CI) or interquartile range or range in both OSAS patients and controls.
Exclusion process of candidate articles was accomplished with two steps: first, the title and abstract were reviewed to remove articles that were clearly irrelevant, such as animal experiments or clinical interventions; second, the full text of the remaining articles was evaluated according to the inclusive criteria, and meanwhile the reference list of each qualified article was also inspected to avoid possible loss of candidates. Two reviewers (Qingsheng Li and Xin Zheng) independently implemented literature search and exclusion process, and they settled all inconsistencies by discussion.
Statistical analyses were handled using the STATA software (11 th version). The changes of circulating TNFalpha were expressed with the weighted mean difference (WMD) along with its 95% CI. Heterogeneity is measured by the I 2 statistic, which is calculated as 100%×(Q -d.f.)/Q (here Q is the Cochran's heterogeneity statistic and d.f. is the degree of freedom) and describes the percentage of total variation across studies that results from heterogeneity rather than from chance [67]. In case of no heterogeneity (the I 2 statistic < 50%), a fixed-effects model was adopted to calculate the WMD and 95% CI. Otherwise, a random-effects model was adopted.
The Begg's funnel plot was created to illustrate the likelihood of publication bias, which was statistically evaluated by the Egger's test. In addition, a filled funnel plot by the fill-and-trim method was also created to determine the number of missing studies with negative findings and filled effect estimates were derived accordingly.

Author contributions
QL and XZ conceived and carried out the analysis. QL and XZ searched literature and abstract data. XZ performed statistical analyses. QL provided assistance in tables and figures. QL and XZ wrote the manuscript. All authors reviewed the manuscript.

CONFLICTS OF INTERESTS
There are no competing interests among authors.