The effect of obstructive sleep apnea therapy on cardiovascular autonomic function: a systematic review and meta-analysis

Abstract Study Objectives Autonomic function is impaired in obstructive sleep apnea (OSA) and may mediate the association between OSA and cardiovascular risk. We investigated the effect of OSA therapy on autonomic function through a systematic review and meta-analysis of intervention studies. Methods A systematic search using three databases (Medline, Embase, and Scopus) was performed up to December 9, 2020. Studies of OSA patients ≥ 18 years with autonomic function assessed before and after treatment with positive airway pressure, oral appliance, positional therapy, weight loss, or surgical intervention were included for review. Random effects meta-analysis was carried out for five groups of autonomic function indices. Risk of bias was assessed using the Cochrane Collaboration tool. Results Forty-three eligible studies were reviewed with 39 included in the meta-analysis. OSA treatment led to large decreases in muscle sympathetic nerve activity (Hedges’ g = −1.08; 95% CI −1.50, −0.65, n = 8) and moderate decreases in catecholamines (−0.60; −0.94, −0.27, n = 3) and radio nucleotide imaging (−0.61; −0.99, −0.24, n = 2). OSA therapy had no significant effect on baroreflex function (Hedges’ g = 0.15; 95% CI −0.09, 0.39, n = 6) or heart rate variability (0.02; −0.32, 0.36, n = 14). There was a significant risk of bias due to studies being primarily non-randomized trials. Conclusions OSA therapy selectively improves autonomic function measures. The strongest evidence for the effect of OSA therapy on autonomic function was seen in reduced sympathetic activity as assessed by microneurography, but without increased improvement in parasympathetic function. OSA therapy may reduce the risk of cardiovascular disease in OSA through reduced sympathetic activity.


Introduction
Obstructive sleep apnea (OSA) is a highly prevalent sleep disorder, estimated to affect 1 billion people globally [1]. It is characterized by repetitive periods of upper airway collapsed during sleep, resulting in intrathoracic pressure swings, repetitive oxygen desaturation, and sleep fragmentation [2]. The perturbations caused by OSA has detrimental effects on the cardiovascular system through altered cardio-metabolic pathways. Specifically, altered autonomic function plays a key role in mediating cardiovascular risk in OSA [3]. The autonomic nervous system (ANS) is a major regulator of the cardiovascular system, fundamentally through reflex arcs that modulate heart rate and blood pressure. The parasympathetic nervous system decreases heart rate and cardiac contractility, whilst sympathetic modulation opposes these effects and increases peripheral vasoconstriction [3].
Patients with OSA display a cyclical pattern of heart rate and blood pressure surges associated with activation of the sympathetic and parasympathetic nervous system from repetitive apnea events. Altered baroreceptor and chemoreceptor reflexes associated with elevated sympathetic nerve activity may contribute to increased cardiovascular risk in OSA. Several methods are utilized to measure autonomic function in OSA, each with advantages and limitations. Our recent systematic review, found evidence of altered autonomic function in OSA, specifically elevated sympathetic nerve activity [4]. Elevated sympathetic nerve activity is a hallmark of hypertension, a major risk factor for cardiovascular disease. Altered autonomic function may be a key mediator linking OSA and cardiovascular disease and mitigating these autonomic changes by treatment of OSA may reduce cardiovascular risk.
Positive airway pressure (PAP), oral appliance (OA), and positional therapy are the recommended treatment options for OSA. Whereas surgical interventions such as bariatric surgery, oro-pharyngeal surgery, and hypoglossal nerve stimulation are studied in highly selected OSA patients, lifestyle interventions such dietary control and exercise are recommended to mostly overweight or obese patients with OSA. PAP and OA have been shown to ameliorate OSA and improve health-related quality of life and blood pressure, however the effect of OSA therapy on autonomic function remains unclear [5]. Accordingly, the aim of this systematic review is to synthesize evidence from available studies on the effectiveness of obstructive sleep apnea therapy on autonomic function.

Protocol and registration
The design of this systematic review and meta-analysis is registered in PROSPERO (CRD42019146171), the international prospective register of systematic reviews.

Eligibility criteria
Original research studies were considered eligible for inclusion in this review if they met the following criteria: Population: adults ≥ 18 years of age with a diagnosis of OSA, defined as an apnea hypopnea index (AHI) ≥ 5/h or respiratory disturbance index (RDI) ≥ 5/h. Intervention: Any intervention for OSA, including CPAP, OA, positional therapy, weight loss, and surgical interventions to improve upper airway patency.
Comparison: Presence of a control or comparison group without OSA treatment. This could include participants at baseline before the intervention (within group comparison), or a comparison group of participants who did not receive intervention (between group).
Outcome/s: Autonomic function outcomes included were measures of Heart Rate Variability (HRV), Blood Pressure Variability (BPV), Baroreceptor function, Catecholamines, Muscle Sympathetic Nerve Activity (MSNA), and Skin Sympathetic Response (SSR). Autonomic function measurements could be made in the wake resting subject, during sleep, or resting wakefulness before or after sleep. Furthermore, we included autonomic function measurements across varying lengths of analysis, specifically for HRV. Definitions and indices of different aspects of autonomic function and typical associations with health outcomes are described in Appendix 1. Study design: Randomized control trials (RCTs) and before and after studies among patients who received OSA therapy were eligible for inclusion. The primary outcome of interest was change in autonomic function associated with intervention.
Language: English language.

Data sources and search strategy
A literature search using the databases of Medline via OvidSP, Embase classic via OvidSP, and Scopus was performed up to December 9, 2020 and all prior years of publication were considered. The search strategy involved the following keywords: sleep apnea, obstructive sleep apnea, hypopnea, disordered breathing, OSA, and OSAS syndrome, autonomic nervous system, sympathetic, parasympathetic, electrocardiography, heart rate, baroreflex, pressoreceptors, and blood pressure. Sleep apnea related keywords included alternate spellings and were combined with autonomic function related keywords. Search results were filtered for studies in adult humans only. The exact search strategy for each database can be found in Appendix 2. All search results were exported into Endnote literature management software, Endnote X9 (Clarivate Analytics, Philadelphia). Duplicate records were excluded using the automated function in Endnote based on matching authors, article title, and journal title. Review articles were excluded through filtering by "Article Type" within Endnote.

Study selection
Study titles and abstracts were screened by two independent reviewers (KS and SU) and any discrepancies resolved by a third reviewer (HD). Full-text review was conducted by a third reviewer (HD). Study selection is described in Figure 1.

Data extraction
Full-text review and data extraction from each article was completed by one reviewer (HD) and independently checked by another (YSB). Data extracted for meta-analysis included index of autonomic function, unit of measurement, time of measurement (e.g. during sleep stage or during wake), mean and standard deviation (SD) with and without OSA treatment, the number of patients in treated and untreated groups, treatment characteristics (type, dose, duration, and adherence), and characteristics of study participants (baseline OSA severity as indicated by AHI or RDI, proportion of male participants, and mean age of participants). Medians and interquartile ranges, standard errors, and 95% confidence intervals were extracted if means and SDs were not reported. For cross-over studies involving two active interventions, data from both interventions were included for analysis. Where studies reported results for two groups of participants e.g. those adherent to treatment and those not adherent, both groups were included for analysis.

Risk of bias assessment
The Cochrane Collaboration tool for assessing risk of bias was utilized. This is a two-part tool, addressing the seven specific domains for risk of bias, namely, sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting. The first part of the tool describes what was reported to have happened in the study, in sufficient detail to support a judgement about the risk of bias. The second part of the tool assigns a judgement relating to the risk of bias for that entry. This is achieved by assigning a judgement of "Low risk" of bias, "High risk" of bias, or "Unclear risk" of bias. Risk of bias assessment was carried out by two independent reviewers (KS, YSB) with any disagreement independently resolved by a third reviewer (HD). The results of the risk of bias assessment are included in Appendix 3.

Statistical analysis
Meta-analysis was carried out using Stata SE 16.1 for Windows (StataCorp, College Station, TX, United States). Effect sizes (Hedges' g) was calculated using the data from included studies. Given that most studies assessed autonomic function using more than one metric, we grouped the outcomes and conducted meta-analysis for each group of outcomes separately: (1) Heart rate variability (HRV), (2) Baroreceptor function, (3) Catecholamines, (4) Muscle sympathetic nerve activity (MSNA), and (5) I-Metaiodobenzylguanidine (MIBG). No meta-analysis was conducted for measures of mechanical effects of respiration since there was only one study reporting this outcome [6]. The heterogeneity of effect sizes was evaluated using Cochran's Q statistic (p < .05) and the I 2 statistic, with values of 25%, 50%, and 75% representing low, medium, and high heterogeneity respectively [7]. Effect sizes were pooled using a random effects model (DerSimonian-Laird). Forest plots were produced to visually summarize the results. A funnel plot was used in conjunction with Eggers' test to examine the possibility of publication bias [8] (Appendix 4). We did not rely on Eggers' test which has lower power to detect biases, especially when there are fewer than 10 studies. The trim-and-fill method was used to balance funnel plot asymmetry [9].
We planned to examine the impact of mode of therapy, duration of therapy, adherence to therapy, and baseline OSA severity on the outcomes through a meta-regression, however, we found this was not possible after data extraction due to a lack of suitable data. First, the autonomic function measures are heterogeneous and cannot be justifiably combined into one overall outcome. Second, many studies reported multiple autonomic function outcomes and are already included in more than one of the outcome-specific analyses. Third, many studies did not report adherence to therapy (n = 15) and together with missing data on duration of therapy, baseline OSA, and patient demographics, less than half of the relevant studies for each outcome had complete data for multivariate meta-regression (Appendix 5). We report the results for univariate and multivariate meta-regression analyses in Appendix 5 but note that these results could be misleading.

Study characteristics
A total of 43 studies were included in this review (Table 1). Of these, 31 studies compared autonomic function in 649 patients before and after PAP therapy, two studies were conducted in 22 patients before and after MAD [10,11] and one study in 40 patients before and after uvulo-palato-pharyngoplasty or Z-palatopharyngoplasty surgery [12].
Three randomized control trials were included: two studies that compared CPAP vs placebo in 79 patients [13,14] and one study that compared CPAP vs no CPAP in 33 patients [15]. Furthermore, two randomized cross-over trials were included: one study of MAD and CPAP 40 patients [16] and one study of MAD and placebo MAD in 23 patients [17].
Additionally, there was one nonrandomized trial of sibutramine vs CPAP in 40 patients [18], three prospective cohort studies were included, one study of CPAP and uvulopalatopharyngoplasty surgery in seven patients [19], one study of weight loss surgery and CPAP in 27 patients [20], one study of physical activity program vs no physical activity program in 74 patients [21] Risk of bias assessment of included studies Risk of bias assessment is shown in Supplementary Materials (Supplementary Figure  S1a and b). We      found a high risk of bias in allocation concealment and random sequence generation because most of the studies reviewed were not randomized trials. Furthermore, studies that were randomized trials, the method used for randomization or how the randomization sequence was concealed was not described. We found low risk of bias for the blinding of participants and personnel and for the blinding of outcome assessment. This was predominantly driven by the nature of our outcome measures, that is, objective measures that are scored by automated algorithms and which are unlikely influenced by performance or detection bias. Furthermore, we found low risk of bias for incomplete outcome data and selective reporting indicating little evidence for attrition bias and reporting bias.

Meta-analysis
Forty-three studies were included in this review however, only n = 38 studies were considered for meta-analysis due to four studies including only graphical results [13,15,22,23] and one study not reporting standard deviations for the outcome data [24].

Heart rate variability
Heart rate variability (HRV) is a noninvasive measure of cardiac autonomic function which can be evaluated by a number of analysis methods including; time domain, frequency domain, and non-linear methods [25]. The analysis of the beat-to-beat variation in heart beats provide an indirect measure of sympathetic and parasympathetic modulation of the sinus node [25]. There were 24 studies investigating the impact of OSA treatment on HRV using one or more of these methods. Since studies typically included multiple measures of HRV evaluated by a number of methods. To avoid double-counting of results within a study, we grouped HRV outcomes into the five categories: (i) global (n = 14 studies), (ii) sympathetic (n = 14), (iii) parasympathetic (n = 19), (iv) low frequency (n = 11), and (v) very low-frequency measures (n = 6); and selected the most common or the most directly comparable measure in each group for comparison across studies. These categories are further defined in Appendix 1.

(ii) Sympathetic HRV
Low frequency to high frequency (LF:HF) ratio of HRV is a frequency domain measure of cardiac sympathovagal balance and an elevated LF:HF ratio suggests sympathetic predominance [25], similarly SD1:SD2 ratio is a nonlinear measure of HRV and is correlated with the LF:HF ratio [26]. The low frequency (LF) component when expressed in normalized units (nu) is regard as a marker of sympathetic modulation [25]. Therefore, elevated  measures of LF:HF, SD1:SD2, and LF(nu) suggests sympathetic predominance. 17 studies used one of these variables to determine sympathetic modulation of HRV before and after OSA treatment. Figure 3 shows there was a tendency towards a small decrease in sympathetic measures of HRV with OSA treatment, although this was not statistically significant (Hedges' g = −0.16, 95% CI −0.37, 0.06). There was moderate heterogeneity between studies (Q = 42.89, df = 19, p < .001; I 2 = 55.70%). Asymmetry in the funnel plot (Supplementary Figure S2) and Egger's test suggests there is evidence of publication bias (p = .02). The trim-and-fill method suggests that the overall effect is zero once six missing studies and their effect studies are imputed and included (Hedges' g = 0.04, 95% CI −0.19, 0.27). The limited meta-regression of n = 8 studies suggests treatment adherence may modify the overall effect of treatment, however, this is in the opposite direction than expected (β = 2.81; 95% CI −0.40-6.02; p = .09), possibly indicating that better controlled and better reported studies generally report smaller effect sizes. There were no significant effects of other treatment or patient characteristics (Appendix 5).

(iv) Low frequency HRV and SD2
The interpretation of the Low frequency (LF) component is controversial, and it is regarded as a marker that includes both sympathetic and vagal influences [25]. Similarly, SD2, a non-linear measure of HRV, is highly correlated with the LF. Thirteen studies measured LF or SD1 to determine both sympathetic and parasympathetic modulation of HRV before and after OSA treatment. Figure 5 shows there was no change in low-frequency measures of HRV with OSA treatment (Hedges' g = −0.12, 95% CI −0.35, 0.10). There was low to moderate heterogeneity between studies (Q = 20.15, df = 12, p = .06; I 2 = 41.31%). Symmetry in the funnel plot (Supplementary Figure S2) and Egger's test suggests no significant publication bias (p = 0.16). Univariate meta-regression provides evidence for an impact of treatment duration, that is, increasing duration of treatment is associated with larger decrements in HRV (β = −0.29; 95% CI −0.58, −0.01, p = .04). Treatment adherence, baseline OSA, patient age, and sex did not appear to influence the effect of treatment. However, this must be cautiously interpreted due to the small number of studies and the fact that only univariate regression was possible (Appendix 5).

(v) Very low frequency HRV
The physiological correlates of Very low frequency (VLF) of HRV are unknown. Five studies evaluated VLF before and after OSA treatment [25]. VLF HRV decreased with OSA treatment as shown in Figure 6, with a moderate effect size (Hedges' g = −0.51, 95% CI −0.93, −0.09). There was low heterogeneity between studies (Q = 6.79, df = 4, p = 0.15; I 2 = 41.05%). Symmetry in the funnel plot (Supplementary Figure S2) and Egger's test shows no evidence of publication bias (p = .56). Univariate meta-regression showed that increasing treatment duration is correlated with a larger decrease in VLF HRV (β = −0.31; 95% CI −0.61, −0.02, p = .04) and that older age appears to reduce the benefit of treatment (β = 0.07, 95% CI 0.00, 0.15, p = .05). However, these results will need to be replicated as more primary studies become available.

Baroreceptor function
BRS, sBRS, and baroreflex gain, are all measures of baroreceptor function. Low baroreceptor function or sensitivity is an unfavorable prognostic marker of autonomic dysfunction. There were six studies investigating the impact of OSA treatment on baroreceptor function with two studies each contributing two comparisons between patients treated with OSA and MAS and patients who were adherent to CPAP and those who were not, resulting in eight comparisons in total. Meta-analysis demonstrated no significant effect of treatment on baroreceptor function (Figure 7), although the effect size tended towards an increase (Hedges' g = 0.15, 95% CI −0.09, 0.39). There was little heterogeneity between studies (Q = 8.47, df = 7, p = .29; I 2 = 17.33%). Egger's test suggests no significant publication bias (p = .27) although there is some asymmetry in the funnel plot (Supplementary Figure S3). Trim-and-fill imputed two Univariate meta-regressions shows that more severe OSA is correlated with a larger effect of treatment (β = 0.01; 95% CI 0.00, 0.03, p = .04) and a larger proportion of men in the study was also associated with a larger treatment effect (β = 3.70; 95% CI 0.31, 7.06, p = .03). On multivariate meta-regression, the effect seems attributable mostly to sex differences, but this is based only on seven studies and does not reach statistical significance (Appendix 5).

Catecholamines
Circulating catecholamines are both a hormone and a postganglionic sympathetic nervous system mediator modulated by the autonomic nervous systems, and elevated levels are associated with sympathoexcitation [28]. There were only three studies investigating the impact of OSA treatment on catecholamines, specifically adrenaline (epinephrine) and noradrenaline (norepinephrine) (Figure 8)

Discussion
This meta-analysis examined the effect of OSA therapy on autonomic function, as assessed by HRV, baroreflex function, catecholamines, MSNA, and MIBG. Our results found improvements in catecholamines, MSNA, and MIBG measures of ANS function. However, there was no effect of OSA therapy on HRV and baroreflex measures of ANS function.
A large number of studies show a clear association between OSA and CVD prevalence and epidemiological studies show that OSA is associated with increased incidences of CVD [2]. However, three randomized controlled trials (RCT) tested the effect of CPAP on cardiovascular risk reduction for OSA patients, and  all showed no improvement in cardiovascular end points [30]. We believe a vitally important step is to study intermediatory mechanisms of cardiovascular risk to establish mechanistic plausibility of the treatment. As such, assessment of autonomic function has played an important role in understanding the disease pathology in both clinical and research settings and increased sympathetic activity is a hallmark of hypertension, major risk factor for CVD. It is widely hypothesized that OSA therapy reduces sympathetic overactivity and improves parasympathetic activity, restoring the impaired sympathovagal balance in OSA patients, and that this is a potential mechanism reducing the risk of cardiovascular diseases. However, the impact of OSA therapy on autonomic function is debated and clear mechanisms are unknown.
HRV is a noninvasive measure of cardiac autonomic control and was the most common method utilized to assess autonomic  function. It is the analysis of the beat-to-beat variability in heart rate modified by the sympathetic and parasympathetic nervous system [25]. Higher measures of global HRV are indicative of a more adaptable cardiac autonomic function and reduced global measures of HRV are associated with increased morbidity and mortality [25]. The majority of reviewed studies assessed global measures of HRV, however there was no significant effect of OSA treatment on global HRV. The lack of significant effects despite a large number of studies on HRV is likely due to heterogeneity between studies, particularly the variation in the specific index of global HRV used in each study, combined with variation in the results of each study due to the sensitivity of HRV to external influences. Time domain analysis of HRV used to assess global HRV is strongly influenced by the condition and length of recording [25]. The most common method of assessment of HRV in OSA patients is during sleep due to the easily accessible ECG signal during sleep studies. However, assessment of global HRV during sleep, especially in patients with OSA with disordered respiratory rhythm will minimize the sensitivity of HRV as a tool for assessing the efficacy of OSA therapy on global measures of HRV [31]. Long duration of data collection utilizing 24-h Holter monitoring to assess SDNN as a marker of global HRV show the strongest association with cardiovascular risk and favor HRV assessment, accounting for metabolic and circadian variability [31]. Future studies in the sleep field, utilizing HRV as a tool to assess autonomic function should look to standardize measurements over longer duration of data collection.
Our results indicate statistically significant improvements in three out of the four measures of sympathetic function, namely Frequency domain analysis of HRV (LF:HF ratio), catecholamine, MSNA and radio nucleotide imaging. Specifically, OSA treatment moderately decreased catecholamines and MIBG but these results were based on few studies and require replication. OSA treatment led to a large decrease in indices of MSNA and there was a tendency towards a small decrease in the LF:HF ratio of HRV with OSA treatment, although this was not statistically significant. A recent meta-analysis showed CPAP therapy led to a small reduction in LF:HF ratio of HRV during sleep [32]. A number of possible pathophysiologic mechanisms may explain the relationship between OSA therapy and sympathetic function. It is hypothesized that ameliorating hypoxia and hypercapnia may diminish the cyclical pattern of heart rate and blood pressure surges through decreased chemoreflex sensitivity and increased baroreflex sensitivity, further reducing circulating catecholamines [33], collectively decreasing sympathetic activity and improving parasympathetic activity. Results from our metaanalysis demonstrated no significant effect of OSA treatment on baroreflex function, although the effect size tended towards an increase in baroreflex function and more research in this field is needed. Furthermore, parasympathetic function assessed by time and frequency domain measures of HRV was unchanged with OSA therapy. Similarly, a recent meta-analysis showed CPAP therapy had no effect on the high frequency component on of HRV during sleep [33]. Despite a large number of studies, we found a lack of significant effects in parasympathetic function in our meta-analysis. This may be explained by heterogeneity between studies and poor control of the conditions under which HRV was assessed.
A potential source of heterogeneity in autonomic response to therapy is variations in the pathophysiological causes of OSA within specific studies. A range of non-anatomical pathophysiological endotypes have been recognized, including low arousal threshold, poor muscle responsiveness, and high loop gain [34,35] An unstable respiratory control system is one of the major non-anatomical mechanisms and can be quantified using loop gain (LG). LG or responsiveness is the theoretical product of the chemoreceptors. The carotid chemoreceptors detect oxygen in the body, exerting a reflex-mediated increase in ventilation and powerful stimulation of sympathetic vasoconstrictor outflow to the skeletal muscle, renal and mesenteric vascular beds. Respectively, CPAP is effective in ameliorating hypoxia, thus lowering LG and consequently MSNA, in part explaining the reduction of MSNA in a number of studies in this review. A low arousal threshold in OSA is thought to be another key driver of hemodynamic instability, contributing to changes in HRV. The arousal threshold can be manipulated pharmacologically with caution, using sedative and hypnotic agents such as trazodone and eszopiclone, however, studies show lack of a significant effect on the arousal threshold in response to CPAP [36], consequently, explaining the marginal changes seen in HRV with CPAP, in this review. Furthermore, the impact of OSA treatment on the various perturbations of OSA (intermittent hypoxia, sleep fragmentation, and intrathoracic pressure swings) may vary from patient to patient, producing variation in autonomic responses.
Adherence to OSA therapy is another important factor when considering effects of OSA therapy on cardiovascular function and outcomes. Sub-analysis of the randomized intervention with CPAP in coronary artery disease and OSA trial showed a cardiovascular risk reduction in patients who used CPAP for ≥ 4 h/night [37]. Although we planned to examine the influence of treatment adherence on autonomic function, there was insufficient data on treatment adherence for this to be carried out, especially given already small numbers of studies for the primary analysis, a limitation in this review. Similarly, our ability to examine the impact of mode of therapy, duration of therapy, baseline OSA severity, and patient demographics was limited and this will need to be explored in future reviews when more primary studies with full information on treatment and patient characteristics become available. The meta-regression analysis found some evidence of a dose-response effect of treatment, most reliably for MSNA, and potential sex differences in the impact of treatment. However, these results need to be interpreted with caution since all-together less than half of the reviewed studies were able to be included in meta-regression.

Conclusion
The location of the autonomic nervous system renders it inaccessible to easy acquisition of direct physiological testing, thus relying on noninvasive, indirect measures. In this review, the strongest evidence for the effect of OSA therapy on autonomic function is seen in reduced sympathetic activity, as assessed by direct measures of MSNA, however, we did not see the expected improvement in parasympathetic function, assessed by HRV. Reduced sympathetic activity maybe one mechanism by which OSA therapy may reduce the risk of cardiovascular disease in OSA. Future studies with standardized protocols for noninvasive assessment of autonomic function are needed and blood biomarkers of sympathetic activation may enable a better understanding of anatomic function and its role in cardiovascular risk and approaches to therapy with OSA.