Obstructive sleep apnea and risk of stroke: A meta-analysis of prospective studies

Abstract

Stroke is the second leading cause of death worldwide [1]. 78% of strokes are first attacks [2], and about 780,000 Americans experience a new or recurrent stroke each year, on average, one stroke every 40 s [3]. Any possible means to prevent stroke should, therefore, be a key public health priority. Obesity is one of the leading risk factors for stroke and a target for stroke prevention [4,5]. Interestingly, human observational studies strongly suggest that obesity is the most important risk factor for obstructive sleep apnea (OSA), which is characterized by repeated obstructions of the upper airway during sleep which result in oxygen desaturations and alterations in blood pressure and cerebral blood flow [6]. Whether OSA independently increases stroke incidences, or whether this relationship is confounded by this population's prevalent cardiovascular risk factors, remains debated. The latest published metaanalysis of 12 prospective cohort studies showed a significant positive association between OSA and fatal and non-fatal stroke (pooled relative risk = 2.15, 95% confidence interval: 1.42–3.24) [7]. However, that meta-analysis found insufficient evidence in the subgroup meta-analyses by type of prevention, study design, geographical area, duration of follow-up, and methodological quality. Therefore, we carried out a meta-analysis of cohort studies to systematically identify whether OSA independently increases the risk of fatal or non-fatal stroke. We followed standard criteria for the performing and reporting of the meta-analyses of observational studies [8]. A systematic search of published articles (through 26 September 2013) was performed by using electronic databases including PubMed, Cochrane Library, and ISI Web of Science databases. We used the following keywords: sleep disordered breathing, obstructive sleep apnea, OSA, sleep apnea, stroke, cerebral infarction, cerebrovascular disease, hemorrhage, cardiovascular disease, coronary artery disease, prospective study, cohort study, and follow-up study. There were no language restrictions. Studies were included for the metaanalysis if they fulfilled the following criteria: (1) the study of adult patients had a community-based or population-based or clinicbased, prospective cohort design; (2) the exposed population was patients with OSA; (3) reported quantitative estimates of the multivariate-adjusted relative risk (RRs) and 95% confidence intervals (CIs) for stroke associated with OSA, hazard ratio/odds ratios were considered equivalent to RRs; and (4) longer than 1 year of follow-up. Studies were excluded if they fulfilled the following criteria: (1) the study design was a nonprospective cohort (for example, cross-sectional and retrospective case–control studies); (2) unadjusted RRs and 95% CIs were reported; and (3) shorter than 1 year of follow-up. The Newcastle–Ottawa Scale (NOS) was used to assess the quality of studies [9]. A maximum of 9 points can be given for each study in the categories of selection, outcome, and comparability. We defined studies of high or low quality based on the median overall score among all studies. The RRs were pooled using the random-effects model [10]. All the statistical analyses were performed in Stata 11 (StataCorp LP, College Station, TX). P values were 2-sided and p b 0.05 was considered statistically significant. Ten cohort studies met the inclusion criteria [11–20] (Fig. 1). The main characteristics of studies in the meta-analysis were presented in Tables 1A and 1B. A total of eleven comparisons (one study did have sex specific data) investigated the association between OSA and risk of stroke. The ascertainment of stroke, duration of follow up, study design, assessment of OSA, and methodological quality varied across studies. Pooling all 11 comparisons, OSAwas associated with a significantly increased risk of fatal or non-fatal stroke (RR, 2.10; 95% CI, 1.50 to 2.93; p = 0.000) (Fig. 2). Fig. 3 showed the pooled RR for stroke stratified by history of stroke, study design, geographical area, duration of follow-up, and methodological quality. The RR for primary prevention studies was 2.06 (95% CI 1.53–2.79; p = 0.000). Increases in stroke events were also found in the subgroup of study design (clinic-based study: RR 2.44, 95% CI 1.24– 4.80, p = 0.010; and population-based study: RR 2.13, 95% CI 1.45– 3.13, p = 0.000, respectively). In addition, we found significant