Can Continuous Positive Airway Pressure Reduce the Risk of Stroke in Obstructive Sleep Apnea Patients? A Systematic Review and Meta-Analysis

Background and Purpose Obstructive sleep apnea (OSA) has been shown to increase the risk of stroke. Although continuous positive airway pressure (CPAP) is considered the treatment of choice for OSA, whether treating OSA with CPAP reduces the risk of stroke remains unclear. We aimed to evaluate the effects of CPAP on incidence of stroke in patients with OSA. Materials and Methods We conducted a systematic review and meta-analysis of all published studies that provided the number of incident strokes in OSA patients in light of their treatment status with CPAP. Results We identified 8 relevant studies: one randomized controlled study (RCT), 5 cohort studies, and 2 studies using administrative health data. The two overlapping cohort studies in women and the elderly and the 2 studies using administrative health data had analyzed the impact of CPAP on stroke apart from cardiac events, whereas the others had focused on the overall cardiovascular events. Based on a meta-analysis of the cohort studies, treatment with CPAP was associated with a lower incidence of stroke and cardiac events with relative risks of 0.27 [0.14–0.53], and 0.54 [0.38–0.75], respectively, although this could not be reproduced in the RCT and the studies using administrative data. Conclusions Treating with CPAP in patients with OSA might decrease the risk of stroke, although there is some conflicting evidence. Such effect was more pronounced in stroke than in cardiac events. Future studies analyzing stroke apart from cardiac disease would be of interest.


Introduction
It is well-known that obstructive sleep apnea(OSA) is associated with cardiovascular diseases (CVD) such as hypertension, arrhythmia, heart failure, coronary artery disease and stroke [1]. Continuous positive airway pressure (CPAP) is currently the most effective treatment for OSA [2], and has been shown to decrease blood pressure among those with OSA [3]. However, whether or not CPAP can improve cardiovascular outcomes remains unclear [4]. A bidirectional relationship between OSA and CVD has been suggested, with OSA merely being an epiphenomenon of CVD [5]. Moreover, one cannot assume treating OSA with CPAP alone would decrease the impact of OSA on CVD, especially in cases of long-standing OSA.
In terms of stroke, OSA increases the risk of stroke and the association is stronger than in other CVD [6][7][8][9]. This might be due to underlying mechanisms which are unique to both stroke and CVD. Therefore, treating OSA might impact stroke differently compared to other CVD.
The effects of CPAP after an acute stroke had been recently reviewed [10,11]. The 2014 AHA/ ASA guidelines [10] suggested that treatment with CPAP might be considered for patients with acute ischemic stroke or TIA and sleep apnea, based on improved short-term functional outcomes, although a systematic review pointed out a lack of evidence with regard to the recurrence of cardiovascular events (CVE) [12]. Out of 3 studies which addressed the effects of CPAP on CVE [11], one RCT showed a delay in the appearance of CVE, although the incidence of CVE was not different over 24 month follow-up period [13], and a cohort study reported a significantly lower incidence of CVE in the CPAP group after a 7 year follow-up period [14], whereas the other cohort study lacked power to demonstrate any effects due to low levels of adherence [15].
As for primary prevention of stroke, the effects of CPAP have not been systematically reviewed except the AHA/ASA guidelines on the primary prevention of stroke (2011) which reported that there were no prospective studies on this topic [16].
Therefore we investigated whether CPAP for OSA reduces the risk of stroke based on a systematic literature review and meta-analysis. Our secondary objectives were to compare the effects with those in cardiac events and to identify factors that may affect the effects of CPAP, such as severity, age, sex, and level of adherence.

Methods
The review of literature was performed according to the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) statement [17].
we screened the titles and the abstracts of the identified studies. Eligibility was considered if they addressed cardiovascular outcomes of OSA. At the second level, we screened the full manuscript. Studies were included in the systematic review if stroke or cardiovascular outcomes were compared between treated and untreated groups. Studies which focused on outcomes after acute stroke or CVE, or did not comment on stroke as an outcome specifically were excluded.

Data items
We extracted data from the studies selected using a data extraction form as follows: the first author, the year of publication, the country where the study conducted, study design, the number of participants in treated and untreated groups, sex (% men) of the participants, age, inclusion criteria for OSA, intervention, adherence, follow-up duration, incidence and mortality from stroke, cardiac events, and source of funding.

Assessment of methodological quality
Two reviewers independently assessed the methodological qualities for each study using the Cochrane risk of bias for RCT [18] and RoBANS [19] for observational studies. Any unresolved disagreements between the two reviewers were resolved through discussion or with a review from a third reviewer. The risk of bias for each study was assessed at the study level and the information was not used in data synthesis.

Statistical analysis
We specified the number of each stroke, cardiac events and overall CVE (the sum of both stroke and cardiac events), and calculated the relative risk (RR) as the primary outcome. We used Review Manager (RevMan) software version 5.2 for the analysis. A meta-analysis was performed on the cohort studies which provided the number of outcomes of interest, while excluding cohort studies with overlapping participants, RCTs, and studies using health administrative data. To estimate heterogeneity, we estimated the proportion of between-study inconsistency due to true differences between studies, rather than differences due to random error or chance, using the I 2 statistic, with values of 0 to 40%, 30 to 60%, 50 to 90%, and 75 to 100% considered to represent insignificant, moderate, substantial and marked levels of heterogeneity, respectively [20]. Q statistics were also obtained and we considered p<0.05 as significant heterogeneity. A Forest plot was generated for each meta-analysis.

Identification and characteristics of the studies
Ultimately we identified 8 relevant articles from 6 study groups: a RCT [21], 5 cohort studies [22][23][24][25][26], and 2 studies using administrative health data (Fig 1) [27,28]. Three studies from the same cohort reported different outcomes such as incidence [26] and mortality [24,25] in different subgroups such as women [24,26] and the elderly [25] respectively, with overlap of participants. Four of the cohort studies were conducted prospectively [23][24][25][26], and the fifth was conducted mostly prospectively including small number of retrospectively recruited subjects [22]. The other two studies retrospectively analyzed health administrative data-Danish National Patient Registry records [27] and VA Inpatient and Outpatient Medical SAS data for US veterans [28].
The number of subjects ranged from 168 to 33,274, which was a total of 60,186 subjects, which consisted of 18,293 treated and 41,893 untreated patients with OSA.
Two studies from a same cohort included only women [24,26] while the others consisted of predominantly men (60~96%). The age of the subjects mostly ranged from their 40s to 60s, except for one cohort study, which included only elderly patients (aged 65 and above) [25]. Patients with previous CVD were excluded in the RCT study [21] and one of the cohort studies [26] but was included in the other 6 studies [22-25, 27, 28].
Diagnosis of OSA and recommendation of CPAP was based on sleep study in the RCT [21] and 5 cohort studies [22][23][24][25][26], whereas the 2 studies which used administrative data identified OSA patients based on their ICD-codes [27,28]. The apnea-hypopnea index (AHI) criteria for diagnosis of OSA and CPAP treatment varied from 5 to 20 events/ hour.
In the study by Buchner et al, subjects were treated with CPAP (81.3%), BiPAP (13.1%) and intraoral appliance (5.5%) with analysis of the outcome being performed for entire modality [23], whereas CPAP was used exclusively in the rest of the studies. The three cohort studies which involved women [24,26] and the elderly [25] regarded those who used CPAP for less than average 4hour as being untreated, whereas the others considered all CPAP patients as being treated regardless of adherence.
Seven studies were conducted in Europe and one in the US [28]. Details of all the studies included are summarized in Table 1.

Quality of the studies (Risk of bias)
The quality of the studies is summarized in Table 2. The RCT had a low risk of bias [21]. In cohort studies, the untreated group consisted of individuals who refused or could not tolerate CPAP, or were not prescribed CPAP at all because of the lack of EDS in mild to moderate cases, which indicates potential selection bias [22][23][24][25][26]. The degree of OSA at baseline was significantly more severe in the treated group in four studies [21,23,24,26] as well as the degree of EDS in the female cohort [24,26]. Otherwise, the baseline differences between the treated and untreated groups were seldom and adjusted.
Studies using the Danish National Patient Registry reported positive predictive values (PPV) of ICD-codes indicating OSA at 82% but did not mention negative predictive values (NPV) [27]. The VA study did not validate the diagnostic performance of these codes for OSA [28]. Moreover, the severity of OSA was not included in both studies [27,28]. In the validation sample of Danish study, CPAP treated group had more severe OSA than untreated group [27].
In VA study, they considered the possibility of misclassification of treated and untreated group in case of patients who received CPAP treatment outside the VA system [28].
Adherence to CPAP treatment in the included studies which reported adherence [21,[23][24][25][26] was within the range of or higher than the adherence generally reported [2]. CPAP adherence data were not available in a cohort study [22] and two studies using administrative DB [27,28].
Only RCT [21] and one cohort study [26] reported that they were blinded for outcome assessment. Outcomes were assessed by routine outpatient visits in the RCT and by multiple sources of information from clinic visit, reviewing medical records, computerized database, death certificates, the contacting primary physician or telephone calls in the cohort studies. The two studies which used administrative data [27,28] identified outcomes through diagnostic codes for CVE. This process was validated in the Danish study, but they assessed only the PPV (97%) and their searching criteria could have led to missed events that did not require hospitalization [27]. The VA study did not validate the diagnostic codes for CVE [28].
The RCT [21] and the Danish study [27] were funded by the companies manufacturing the CPAP machine. Publication bias cannot be fully excluded.

The effects of CPAP on stroke
Effects on the incidence of stroke. We identified 6 studies [21][22][23][26][27][28] that provided the number of incident stroke events depending on treatment and the results of individual studies are summarized in Table 3. One cohort study consisting of a women [26] and two  [26]. However, CPAP was not shown to have effect on the incidence of either stroke or coronary heart disease in two studies using administrative data [27,28]. On the incidence of overall CVE, RCT could not find significant effect of CPAP. However, a post hoc analysis of the RCT found that treating with CPAP for 4 hours per night or longer was associated with lower incidence of hypertension and CVE (adjusted IDR 0.69 [0.50,0.94])  [23] and Doherty et al. also found that newly developed hypertension and CVE are significantly less common in treated group [22].
Effects on mortality associated with stroke. We identified 3 studies that provided the number of deaths caused by stroke depending on whether CPAP was used or not [22,24,25]. Two studies, one which focused on cardiovascular mortality in women and the other in the elderly [24,25] were from a same cohort with the women cohort study which reported incidence of CVE including both fatal and nonfatal cases [26]. The RCT and Buchner's cohort study provided the number of death from overall CVE without classification into stroke or cardiac events.
Only the elderly cohort study analyzed the effects of CPAP on mortality from stroke apart from cardiac diseases. Untreated severe OSA was associated with an increased risk of death from stroke ( [22][23][24] excepting overlapping studies [25,26] (Fig 3). The degree of heterogeneity between studies was moderate for overall CVE (Q statistic, P = 0.15, I 2 = 48%).
Factors that may affect the effects of CPAP. None of the studies analyzed how clinical factors might influence the effects of CPAP in stroke apart from the overall CVE.
The risk reduction with CPAP for overall CVE was demonstrated in both women [24,26] and men, and in both middle-aged patients and the elderly, even those 75 years of age or older [25].
About severity of OSA, three studies included subgroup analyses depending on OSA severity [21,23,24]. The RCT couldn't find a significant effect of CPAP on overall CVE, which was not different across subgroups of AHI or the percentage of time with SaO2 less than 90% [21]. Women cohort study found the reduction of mortality from overall CVE in both groups with AHI of 10-29 and AHI30, although the effect was not statistically significant in group with AHI of 10-29 after adjustment [24]. In one of the cohort studies which consisted of predominantly men, a subgroup analysis showed that CPAP reduced the overall CVE also in the group with mild to moderate OSA with an AHI 5 to <30 (adjusted HR 0.37 [0.17-0.78]) as well as those with severe OSA [23].

Discussion
There are few studies which have investigated the effects of CPAP on stroke in patients with OSA. The cohort studies generally supported the hypothesis that CPAP may reduce the risk of stroke [22][23][24][25][26], whereas the RCT and studies using administrative data did not [21,27,28].
The RCT had shorter follow-up period than cohort studies and limited power to assess the effect of CPAP on stroke separately [21]. The studies using administrative data have critical  Effect of CPAP on Incidence of Stroke in Obstructive Sleep Apnea Patients issues of validity and bias. Most of all, administrative database didn't have the information of sleep study results [27,28]. CPAP treated group might have more severe OSA than untreated group, which was supported by the finding from the validation sample in Danish study, so the beneficial effect of CPAP could be more pronounced than indicated by those studies [27].
However, biases do exist in cohort studies as well. The untreated group in general consisted of individuals who declined or could not tolerate CPAP [22][23][24][25][26], causing potential bias of better health behavior and better socioeconomic status in the treated group. This might have led to an overestimation of the benefits of CPAP.
On the other hand, baseline differences in the degree of OSA and poor adherence to CPAP might have led to an underestimation the effects of CPAP. The degree of OSA was more severe in most cohort studies [23,24,26] and even in the RCT [21]. All of the studies except the Spanish cohort studies [24][25][26] included individuals with poor adherence to CPAP in the treated group, which may have not truly reflected the efficacy of CPAP. The RCT showed risk reduction of hypertension and CVE in subgroup analysis comparing adherent patients with untreated patients. However, healthy user effect cannot be ruled out in this analysis as well as cohort studies, because the subgroups could not have been randomized based on adherence [21].
Based on a meta-analysis from cohort studies, we found that the treatment with CPAP was associated with lower risk of stroke among patients with OSA with an overall risk reduction of 73%, which was greater than that seen in cardiac disease (46%). This may be due to the difference in mechanisms between stroke and cardiac disease associated with OSA. For example, snoring-associated vibration which may increase carotid artery atherosclerosis [29] and paradoxical embolism in patients with patent foramen ovale during apnea events may be risk factors of stroke but not of cardiac disease [30].
It remains unclear as to whether CPAP has a positive effect on stroke particularly in a certain subgroup. Besides one Spanish cohort study which showed decrease in risk on overall CVE in women [24,26] and the elderly [25], all the other studies consisted of mostly middleaged men hence subgroup analysis could not be performed. CPAP seems effective to decrease the risk of overall CVE not only in patients with severe OSA but also those with mild-to-moderate OSA based on a cohort study which consisted of predominantly men [23], but this was not proved in the women cohort study [24]. The association between OSA and stroke or cardiac disease was reported to be stronger in men [7]. The incidence of stroke increased in men with an AHI higher than 19 and women with an AHI higher than 25 [31], and the incidence of coronary disease and heart failure increased in men with OSA but not in women [32]. There was paucity of data for patients younger than 40. In the previous review, the relative mortality rate was highest in their twenties and then decreased with age [33]. This may be due to agerelated differences in pathophysiology of OSA or healthy survivor effects [32]. Such phenomena raise the need for analyzing the effects of CPAP for each subgroup. In addition, studies on people outside Europe and US are lacking.
We performed meta-analysis on a limited number of studies that were heterogeneous in terms of inclusion criteria. The number of stroke events was also small and was not shown in detail in each study to allow for any subgroup analysis depending on severity, age, and sex. A larger-scale RCT of longer follow-up should provide more accurate evidence. However, based on the strong evidence for the effect of CPAP on daytime function [34], and some favorable evidence supporting the effects of CPAP on hypertension [3], randomly allocating patients to sham devices for long time could be unethical. For this reason, various strategies are needed to estimate the impact of CPAP on stroke through RCT. The RCT included in this review had studied patients with OSA but without excessive sleepiness. Studying high risk patients could shorten the length of follow-up to show the efficacy as in ongoing RCTs, but obtaining further evidence for primary prevention for stroke from those studies would be challenging. Therefore, in spite of the limitation, this study will provide useful information for treating general people with OSA.

Conclusion
In spite of the lack of RCTs and somewhat conflicting results among the studies, the data from well-designed cohort studies suggest that CPAP treatment may reduce the risk of stroke in patients with OSA. Such results were more pronounced in stroke than in cardiac disease. Future studies should analyze the effects of CPAP on stroke apart from overall CVE, depending on certain subgroups as according to age, gender, and the severity of OSA.