Serum and Plasma Tumor Necrosis Factor Alpha Levels in Individuals with Obstructive Sleep Apnea Syndrome: A Meta-Analysis and Meta-Regression

Background: Obstructive sleep apnea syndrome (OSAS) is associated with a variety of inflammatory factors. Specifically, proinflammatory cytokines appear to be associated with the pathogenesis of OSAS. Methods: For the present meta-analysis and meta-regression on serum and plasma tumor necrosis factor alpha (TNF-α) levels in individuals with and without OSAS, we performed a systematic search without any restrictions of the PubMed/Medline, Scopus, Cochrane Library, and Web of Science databases to find relevant articles published up to 1 February 2020. Results: Fifty-five (adults: 29 studies on serum and 17 studies on plasma; children: 4 studies on serum and 5 studies on plasma) were included and analyzed. Always compared to age-matched healthy controls, the pooled MDs were as follows: adults, serum: 10.22 pg/mL (95% CI = 8.86, 11.58; p < 0.00001); adults, plasma: 5.90 pg/mL (95% CI = 4.00, 7.80; p < 0.00001); children, serum: 0.21 pg/mL (95% CI = 0.05, 0.37; p = 0.01); children, plasma: 5.90 pg/mL (95% CI = 4.00, 7.80; p < 0.00001). Conclusions: Compared to healthy and age-matched controls, adult individuals with OSAS had significantly higher serum/plasma TNF-α levels. For children with OSAS, significantly higher levels were observed for TNF-α in serum but not in plasma.

Life 2020, 10, x FOR PEER REVIEW 8 of 28
Life 2020, 10, x FOR PEER REVIEW 9 of 28

Ethnicity
Subgroup analysis of serum TNF-α concentrations in adult samples are reported in Table 2.

Mean BMI of Individuals with OSAS
For serum TNF-α concentrations, and compared to healthy controls, adult participants with OSAS and a BMI > 30 km/m 2 showed a pooled MD of 1.27 (95% CI: 0.67, 1.86; p < 0.0001). For adult participants with OSAS and a BMI ≤ 30 km/m 2 , the pooled MD was 23.07 pg/mL (95% CI: 20.18, 25.96; p < 0.00001). Thus, results showed that serum TNF-α concentrations were 18.2 times higher participants with OSAS and a BMI > 30 kg/m 2 , when compared to those participants with a mean BMI ≤30 kg/m 2 .
For plasma TNF-α concentrations, and compared to healthy controls, adult participants with OSAS and a BMI > 30 km/m 2 showed a pooled MD of 1.06 pg/mL (95% CI: 0.34, 1.78; p = 0.004). For adult participants with OSAS and a BMI ≤ 30 km/m 2 , the pooled MD was 10.82 pg/mL (95% CI: 7.14, 14.49; p < 0.00001). Thus, results showed that plasma TNF-α concentrations were 10.2 times higher in adults with OSAS and a BMI > 30 kg/m 2 than plasma TNF-α concentrations of adults with OSAS and a BMI ≤ 30 kg/m 2 and always compared to healthy controls.

Mean BMI of Controls
For serum TNF-α levels, compared to controls, individuals with OSAS and with a BMI > 30 kg/m 2 showed a pooled MD of 2.28 pg/mL (95% CI: 0.86, 3.70; p = 0.002); while compared to controls, individuals with OSAS and with a BMI ≤ 30 kg/m 2 showed a pooled MD of 18.64 pg/mL (95% CI: 16.33, 20.95; p < 0.00001). Thus, serum TNF-α levels were 8.2 times higher in studies with participants with OSAS and a BMI >30 kg/m 2 than serum TNF-α levels with participants with OSAS and a BMI ≤ 30 kg/m 2 .

Mean BMI of Individuals with OSAS
For serum TNF-α levels, studies with Caucasian participants with OSAS and a BMI > 30 kg/m 2 showed the pooled MD of 2.58 pg/mL (95% CI: 1.07, 4.10; p = 0.0008), whereas for studies with participants with a BMI ≤ 30 kg/m 2 serum TNF-α levels did not statistically differ between participants with or without OSAS (Table 3).
For plasma TNF-α levels, studies with Caucasian participants with OSAS and a BMI ≤ 30 kg/m 2 showed the pooled MD of 7.99 pg/mL (95% CI: 1.84, 14.15; p = 0.01), whereas no significant MDs were observed in studies including Caucasian participants with OSAS and a BMI > 30 kg/m 2 .

Mean BMI of Controls
For studies with serum TNF-α levels and Caucasian controls with a mean BMI > 30 kg/m 2 , the pooled MD was 2.58 pg/mL (95% CI: 1.07, 4.10; p = 0.0008). There was no statistically significant MD for serum TNF-α levels between Caucasians with or without OSAS and a mean ≤ 30 kg/m 2 .

Total Number of Participants
Serum TNF-α levels had a significant difference in studies with ≤ 100 cases (MD = 3.21 pg/mL; 95% CI: 1.37, 5.05; p = 0.0006), whereas there was no significant difference for studies with >100 cases.

Mean BMI of Controls
For studies with serum TNF-α levels and controls with BMI of controls >30 kg/m 2 , there was a significant difference for serum TNF-α levels (MD = 2.28 pg/mL; 95% CI: 0.86, 3.70; p = 0.002), compared to studies with controls with BMI of controls ≤ 30 kg/m 2 (Table 5).

Total Number of Participants
For serum TNF-α levels and in studies with more than 100 participants, the pooled MD was 63.51 pg/mL (95% CI: 53.65, 73.45; p < 0.00001), while in studies with ≤ 100 participants, the pooled MD was 4.42 pg/mL (95% CI: 3.10, 5.74; p < 0.00001). Thus, serum TNF-α levels were 14.4 times higher in studies with more than 100 participants, compared to serum TNF-α levels in studies with ≤100 participants.

Meta-Regression
The results of meta-regression showed that age, the publication year, the mean BMI, the mean AHI, and the number of participants did not systematically change serum and plasma TNF-α levels ( Table 7). Table 7. Meta-regression analysis based on some variables for serum and plasma levels of tumor necrosis factor alpha in obstructive sleep apnea syndrome patients compared with controls in adult participants.

Year of
BMI, Body mass index; OSAS, Obstructive sleep apnea syndrome. R: Correlation coefficient.

Quality Assessment
The quality score of each study included in the meta-analysis is illustrated in Table 8. Table 8. Quality assessment scores of the studies involved in the meta-analysis.

Sensitivity Analysis
The "cumulative analysis" and the "one study removed" as two sensitivity analyses showed the stability of the results. In addition, excluding statistically the studies with outliers data did not change pooled analysis of serum (MD = 4.53 pg/mL, p < 0.00001) and plasma (MD = 4.18 pg/mL, p < 0.00001) TNF-α levels ( Table 9).  Figure 6 shows the funnel plots of the analysis of serum and plasma TNF-α levels and Table 10 illustrates the results of trim-and-fill method on bias. For serum levels, Egger's and Begg's tests (p = 0.02431 and p = 0.04677, respectively) revealed a publication bias. For plasma TNF-α levels, Begg's test (p = 0.03943) revealed a publication bias, but Egger's test (p = 0.19315) did not show any bias between and across the studies.

Publication Bias
For serum TNF-α levels and 15 imputed studies, under the fixed-effects model, the point estimate and pseudo 95% CI for the combined studies was 0.410 (0.365, 0.455); using the trim-fill method, the imputed point estimate was 0.323 (0.278, 0.368). In addition, under the random-effects model, the point estimate and 95% CI for the combined studies was 8.245 (7.006, 9.483   The overall effect sizes on serum and plasma TNF-α levels reported in the forest plot appeared valid, with trivial publication bias effect based on fixed-effects model, because the observed estimates were similar to the adjusted estimates. In contrast, the overall effect sizes on serum and plasma TNF-α levels reported in the forest plot appeared invalid, with significant publication bias effect based on random-effects model, because the observed estimates had high difference with the adjusted estimates.

Discussion
The present meta-analysis with 55 studies evaluated the serum TNF-α levels (29 studies of adults and 4 studies of children) and plasma TNF-α levels (17 studies of adults and 5 studies of children) in individuals with OSAS, compared to controls. The results showed that plasma and serum TNF-α levels in adult individuals with OSAS were significantly higher than the corresponding levels of control. For children with OSAS serum TNF-α levels, but not plasma TNF-α levels, were statistically significantly higher than those of controls. The present results have clinical importance because elevated serum and plasma TNF-α levels can be a risk factor for the development of further systemic diseases in adults with OSAS. Further, more generally, inflammation may play a crucial role in the pathogenesis of OSAS, and there is evidence that individuals with OSAS have elevated interleukin TNF-α [16,29,40,41,45,48,62,78]. These conclusions also have practical importance because, in addition to routine checks on sleep and sleep-disordered breathing, both children and adults with OSAS need a thorough monitoring of their immune systems.
Elevated TNF-α levels have been associated with the pathophysiology of reoxygenation injury, myocarditis, cardiac allograft vasculopathy, heart failure progression [81], arthritis, diabetes, Crohn's disease, and also cachexia which correlated with terminal malignancy and AIDS [82]. Further, TNF-α is a well-known inflammatory marker, besides being related to atherosclerosis in males [83]. It is also confirmed that hypoxia raises the expression of proinflammatory molecules such TNF-α [84,85].
The present pattern of results also sheds some light on further new findings: More specifically, for adults, it turned out that compared to healthy controls, TNF-α levels were higher both in blood serum and blood plasma. It follows that at least for adults with OSAS, assessing either blood plasma or blood serum does not appear to be of clinical importance. In contrast, for pediatric samples and compared to healthy controls, higher TNF-α levels were observed for blood serum but not for blood plasma. The quality of the data at hand does not allow a deeper understanding of the underlying physiological mechanisms. Thus, for want of such data, the following admittedly speculative assumptions are advanced. First, it is conceivable that for pediatric samples, the number of four studies on plasma TNF-α was too small, and therefore, the statistical variance of plasma TNF-α was blurred by sample size inconsistencies. Similarly, second, at least among adults, Hirotsu et al. [70] showed that OSAS and inflammatory markers such as IL-6 and TNF-α were associated with gender in a very complex fashion. A closer inspection of the four studies on plasma TNF-α among pediatric samples showed that gender was not introduced as a specific factor or confounder. Given this lack of analyses in these publications, possible gender effects could neither be confirmed, nor declined. Third, Imani et al. [86] showed that for pediatric samples, the definition of OSAS was not applied as strict and consistent as among adult samples. It follows that the zero-differences of plasma TNF-α between children with and without OSAS might mirror inconsistencies as regards the definition of control samples. Fourth, Cameron et al. [87] observed that besides the disease status, in pediatric samples with Inflammatory Bowel Disease, the outcome of an antitumor necrosis factor (TNF) therapy depended from participants' pubertal stage. Given this, it is conceivable that the zero-difference of plasma TNF-α between pediatric samples with and without OSAS might have been blurred by the pubertal stage. In the same vein, fifth, it is conceivable that pubertal stage and its implicit growth spurt in height and weight might have obscured a clear-cut pattern of results as regards TNF-α levels between children with and without OSAS. Last, even if highly unlikely, one cannot rule out that the pattern of results is an accidental finding. The pattern of results described in the present meta-analysis and systematic review consistently showed that higher TNF-α levels were associated with pathologically higher BMI scores as a proxy of obesity. Given that above all childhood overweight and childhood obesity is increasing worldwide and is a major concern of public health [88], the association between overweight/obesity, TNF-α levels and OSAS demand particular attention. Following Popa et al. [13], TNF-alpha appeared to be of upmost importance in the development and maintenance of metabolic diseases in which a shift toward a proatherogenic lipid profile and impaired glucose tolerance appeared to occur. Further, investigations assessing the impact of anti-TNF agents on intermediary metabolism seemed to show that a TNF-alpha blockade may improve insulin resistance and lipid profiles, which in turn appeared to be associated with overweight and obesity. Likewise, Ciftci et al. [31] showed in their study that compared to obese males with no OSAS, in obese males with OSAS TNF-alpha levels were significantly higher. Ciftci et al. [31] concluded from their results, that the association between cardiovascular morbidity and OSAS appeared to be best described by the coexistence of other cardiovascular risk factors such as circulating IL-6 and TNF-alpha levels. In the same vein, Sahlman et al. [47] showed in their study that pro-inflammatory markers such as TNF-alpha were markedly increased in patients with mild OSAS. To find a physiological explanation for such processes, Steiropoulos et al. [48] summarized in their introduction that besides the role of energy depot adipose tissue is also an active endocrine organ, releasing proinflammatory cytokines such as Tumor Necrosis Factor-α (TNF-α) and Interleukin-6 (IL-6) which modulate blood pressure, and lipid-and glucose-metabolism. Further, the proinflammatory transcription factor NF-κB appears to be upregulated in OSAS as a result of alterations between hypoxia and reoxygenation and as a result of sleep deprivation. Additionally, NF-κB regulates the expression of inflammatory genes. Given these physiological processes, it appears plausible that OSAS, obesity and poor sleep appear to be highly interrelated.
The results of one study [70] showed that IL-6, triglycerides, and AHI were positively associated with TNF-α, while sex, ghrelin and total cholesterol had a negative association. Further, postmenopausal women had higher TNF-α levels than those of premenopausal women [70]. One study showed that higher TNF-α levels were significantly related to some neurocognitive deficits in children with OSAS [14].
Next, TNF-α appears to mediate both a somnogenic activity and fatigue related to excessive daytime somnolence in obese patients with OSAS [14,28,30]. TNF-α levels were related to the severity of OSAS [31]. One study reported that BMI was weakly related to TNF-α levels [32]. However, Kanbay et al. [41] showed that in individuals with OSAS and obesity serum TNF-α levels were significantly higher than those without obesity. Not surprisingly, there was also a positive albeit modest correlation (r: 0.181; p: 0.034) between higher BMI scores and higher TNF-α levels [16].
As regards AHI and TNF-α levels, Kanbay et al. [41] and others [16,78] reported a significant positive correlation between a higher AHI and higher serum TNF-α levels.
Next, Minoguchi et al. [16] reported in their meta-analysis higher plasma and serum TNF-α levels in adults, compared to non-adults.
Despite the novelty of the results, the following limitations should be considered: (1) In all studies, results have not been adjusted for possible confounding factors such as obesity, smoking, or alcohol consumption. (2) The results of the funnel plots showed a publication bias across the studies; it follows that a systematic bias in the data presentation cannot be ruled out. (3) Studies with a small sample size (less than100 cases) had an inadequate power to detect possibly meaningful associations. (4) There was a high heterogeneity among studies in some analyses. (5) Studies reported different cut-off AHI values, which made comparisons between the studies difficult. (6) In some studies, TNF-α levels were considered as secondary outcome. (7) In some studies, the existence of mixed ethnicities might have blurred the associations between the ethnicity and TNF-α levels.
By contrast, the strengths of the meta-analysis were as follows: (1) There were sufficient studies to allow the subgroup analyses. (2) The sensitivity analysis showed a stability of the results. (3) The studies written in other languages than English were included in the meta-analysis.

Conclusions
The result of the meta-analysis and meta-regression confirmed that compared to healthy controls, individuals with OSAS (adults) had significant higher serum/plasma TNF-α levels, whereas for children with OSAS, this pattern of results was observed for serum TNF-α levels, but not for plasma TNF-α levels. Last, the results of the present meta-analysis showed that elevated TNF-α levels in individuals with OSAS appeared to be related to the severity of the disease. Future studies might investigate if and to what extent interventions on OSAS (e.g., using CPAP devices) favorably impact on TNF-α levels and possibly also on weight regulation.