Obstructive sleep apnea (OSA) is a chronic disorder that leads to early mortality, stroke, and cardiovascular disease (CVD) (1–5). OSA affects 34% of men and 17% of women of all ages, with an estimated cost of $149.6 billion in 2015 from untreated OSA (6–9). Obstruction of the upper airway during sleep causes hypoxemia, sleep fragmentation, and sympathetic nervous system activation (6–8). OSA may lead to stroke through its associations with potent vascular risk factors, such as hypertension, diabetes mellitus, obesity, and atrial fibrillation. OSA may also increase stroke and CVD risk through reduction in cerebral blood flow, altered cerebral autoregulation, impaired endothelial function, increased platelet activation, inflammation, and oxidative stress related to the intermittent hypoxemia–reoxygenation and arousals associated to increased sympathetic tone (10). The treatment of OSA has been considered an important intervention for reducing the morbidity and mortality associated with stroke and CVD (11). However, treatment of OSA has not consistently reduced cardiovascular risk (12, 13); results partly explained by methodological limitations, such as suboptimal adherence to positive airway pressure therapy. Importantly, some limitations may lie on the need to better identify the subset of individuals who respond more favorably to OSA therapy, with the potential of a greater reduction in adverse outcomes, for further review consider the article by Drager et al. (14). Recently, the emphasis on personalized medical care introduces the possibility of a unique approach to the treatment of OSA based on its pathophysiological mechanisms and/or clinical presentation (15, 16). Personalized medicine can lead to targeted clinical and public health strategies for reducing stroke and CVD risks associated with OSA.

Historically, OSA has been considered a uniform condition defined by an apnea-hypopnea index (AHI) ≥5 events per hour of sleep (17). The AHI combines apneas and hypopneas associated with hypoxemia. Yet, other sleep metrics (i.e., oxygen nadir, arousal frequency), along with clinical symptoms and molecular markers could be better predictors of stroke and CVD outcomes in OSA (17–19). The heterogeneity of clinical presentations leads to the evaluation of OSA phenotypes as a strategy to identify at-risk individuals for stroke and CVD (18, 19). As defined in the article by Zinchuk et al., a phenotype is “a category of patients with OSA distinguished from others by a single or combination of disease features, in relation to clinically meaningful attributes (symptoms, response to therapy, health outcomes, quality of life)” (20). The authors further define clinical phenotypes based on a priori categories according to the presence or absence of symptoms (e.g., daytime sleepiness). Molecular-genetic phenotypes are based on molecular features (i.e., DNA and RNA). Empirical or data-driven phenotypes are based on statistical analyses (e.g., cluster) examining a multitude of sleep symptoms, demographics, clinical, and physiological data to define distinct OSA subtypes (20).

The high prevalence of undiagnosed OSA and its resultant stroke and CVD risk provide a strong impetus to develop better risk-stratification strategies (18, 19). There is a need to further define the OSA phenotypes that may differentially affect stroke and or CVD risk. Thus, the focus of this systematic review is to summarize what is known about OSA and its phenotypes, in relation to the intermediate markers that increase stroke and CVD risk in OSA. We further highlight gaps in the literature that frame a research agenda.

We performed a systematic review by searching the PubMed and Embase databases using the PRISMA guidelines (21). The search strategy was constructed using MeSH terms. The initial search was conducted with the terms “obstructive sleep apnea” or “sleep disordered breathing” and “phenotype” or “phenotypes.” Also articles were searched with the key word “obstructive sleep apnea phenotypes.” The first author evaluated the abstract for articles that included “cardiovascular disease,” or “stroke,” or “cerebrovascular disease.” The search included articles written from January 1, 2007 until November 1, 2017.

To be included in the review, articles had to be primary research studies, published in peer-reviewed journals, written in English and included adults (≥18 years old). Articles were excluded if they were (1) animal studies; (2) a pediatric population; (3) written in languages other than English; (4) conference abstracts; (5) case reports or case series; and (6) commentary or review articles.

A total of 153 articles were retrieved and the abstract were reviewed by the lead author for inclusion and exclusion criteria. A total of 14 articles met the inclusion criteria and were independently reviewed for additional information by the authors (Alberto R. Ramos, Pedro Figueredo, and Douglas M. Wallace). The lead author conducted ancestry search of all retrieved articles’ reference lists. The authors (Alberto R. Ramos, Douglas M. Wallace) also searched the major sleep journals including Sleep Medicine Reviews, Sleep, the Journal of Clinical Sleep Medicine, Sleep Medicine, CHEST, Thorax, and Sleep and Breathing. This resulted in two additional articles. Article tracking software (COVIDENCE) was used to manage the retrieved literature.

The majority of articles used clinical symptoms (n = 8) to define OSA phenotypes followed by experimental or data-driven approach (n = 5) and by molecular-genetic studies (n = 4). Most of the articles were cross-sectional analyses, case–control, or retrospective studies from sleep disorder centers (22–29). Two studies, Patel et al. (30) and Luyster et al. (23) were from community or population-based samples. The study from, Saaresranta et al. (28) and Zinchuk et al. (31) were longitudinal studies. Most studies comprised male participants, except for the two population based studies, Patel et al. (30) and Luyster et al. (23), and Masa et al. (32), which had 55 and 65, and 65% of females, respectively (Table 1).

The purpose of this review was to systematically evaluate manuscripts with concurrent definition of sleep apnea phenotypes and measures of vascular disease. As previously described, a priori symptom classification showed that the OSA-excessively sleepiness (EDS) phenotype, had increased mortality and adverse vascular outcomes. Interestingly, the OSA-EDS phenotype was associated with increased inflammatory markers, such as CRP and interleukin-6. Similarly, increased inflammation is associated with atherogenesis, hence increased inflammation could mediate the association between OSA-EDS phenotype and atherosclerotic disease. Epidemiological studies show that EDS is observed in up to 50% of patients from sleep centers; therefore, many patients do not endorse EDS. While these is not fully understood, the study from Chen et al. (27), suggest that hypomethylation of genes involved in the natriuretic peptide receptor-2 pathways could lead to EDS in OSA (27). These findings, coupled with the increased in CVD among the OSA-EDS phenotype, suggest a genetic susceptibility to the effects of hypoxia (26). Interestingly, the study from Masa et al. (32) suggests that severe hypoxemia could be protective at least in certain populations (obesity hypoventilation). A possible explanation is ischemic precondition, an adaptive mechanism where low levels of hypoxemia/ischemia leads to vascular and endothelial changes that “protects” against worse vascular events and possibly explaining the lower mortality rates of older age groups with OSA (38).

Interestingly, studies using empirically or data-driven analytical methods such as cluster analysis, observed increased cardiovascular outcomes in patients without daytime sleepiness, but rather clusters that were defined as having “poor,” “fragmented” sleep predominantly with insomnia symptoms or periodic limb movements had increased cardiovascular outcomes. Using a data-driven approach provide an avenue to further identify at at-risk individuals that otherwise may not be treated using AHI cutoffs. As noted in the study by Zinchuk et al. (31), the OSA using the AHI cuttoffs did not predict cardiovascular outcomes. This is consistent with other studies were the AHI did not correlate strongly with most relevant OSA comorbidities, such as hypertension (39). However, as noted in these studies (22, 26, 29, 37), the “classic” OSA patient (snoring, obese, daytime sleepiness) is not the most common phenotype.

In our systematic review, most studies sampled middle-aged males from sleep centers. Of importance, studies derived solely from sleep centers are of limited external validity and generalizability due to possible selection bias. Consequently, there is a need to develop tools for health-care professionals, sleep specialist, and scientist to further recognize and risk-stratify different OSA phenotypes; particularly in older-adults and females (40, 41). Importantly, population-based studies can help further evaluate age, sex and ethnic differences in OSA phenotypes; allowing clinicians and researchers to recognize participants an increased risk of stroke and CVD (40, 41), especially in populations with large burden of CVD that may benefit from inclusion in treatment studies.

Most of studies did not measured the physiologic OSA phenotypes or endotypes, as described by Eckert (42). An anatomical predisposition for upper airway collapse is necessary for OSA; however, these phenotypic classification describe non-anatomical pathways to upper airway obstruction, such a low respiratory arousal threshold, and instability of ventilator control causing an increase in “loop gain.” Of interest, a case–control study compared differences in cardiac autonomic tone in three different groups, OSA patients with hypoxia, OSA without hypoxia and controls (AHI < 5) (24). The authors observed an increased sympathetic contribution in patients with OSA and hypoxia. As noted by the authors, patients with increased loop gain can have partial airway obstruction leading to “apneas” without hypoxia. In contrast, patients with decreased loop gain coupled with factors that worsens upper airway collapse, could increase the length of apneas and arousal thresholds leading to periods of hypoxia.

We may have not included all studies describing OSA phenotypes. We only considered studies with OSA phenotypes, which also had information on risk factors, vascular measures or comorbidities (e.g., hypertension, diabetes) known to increase the risk of stroke and CVD. In addition, articles describing OSA phenotypes, but did not label the study as such, may have not being included in this review as part of our search strategy.

This review provides a framework to further understand clinical, molecular and genetic information using data-driven approaches to reveal OSA subtypes. There are several important clinical phenotypes that have not been studied comprehensively. For example, the clinical presentation of OSA in pre- and post-menopausal women has been described to be significantly different than that of men (18). Similarly, race–ethnic differences may exist for OSA but have yet to be explored (43). Novel methods for diagnosis of sleep apnea (home sleep test) and use of commercially available wearables could provide further venues to define and identify patients at risk. In addition, phenotyping body position and sleep architecture needs to be incorporated into the current paradigms, as this may differently affect vascular disease and provide different intervention strategies such as positional therapy (22). There is a paucity of longitudinal studies examining whether OSA subtypes are stable over time. In addition, there are no studies determining whether OSA subtypes respond differentially to OSA treatments. Future studies should consider different OSA phenotypes with the intent to evaluate them against clinical outcomes, especially to optimize the use of clinical trials (14). Determining and evaluating the OSA phenotypes at risk for stroke or CVD can provide the structure for clinical trials in the treatment of OSA and reduction of CVD morbidity and mortality. Importantly, novel genetic and molecular markers are necessary. For example, micro-RNAs have been associated to increased atherosclerosis and CVD and recently have been shown to cause endothelial dysfunction associated to the intermittent hypoxia of patients with OSA (44). Future efforts should consider analyzing large-scale cohorts, in collaboration with geneticist and use of expertise in informatics and big data science.

Drafting/revising the manuscript for content, including medical writing for content, study concept or design, analysis and interpretation of data.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.