Summary of main results
To our knowledge, this study is the first systematic review evaluating the prevalence and risk of OSA in patients with MCI. In total, we found five studies reporting the prevalence of OSA in MCI showing considerable variations (11–71%). Two studies that used PSG [27, 28] to diagnose OSA showed that the prevalence of OSA in patients with MCI is high, 70%, while there is no significant association between having OSA and MCI. Furthermore, the prevalence rates were influenced by OSA diagnostic method and patient recruitment location (community or clinic based). Our findings suggest that OSA may be prevalent in individuals with MCI. Due to the cross-sectional nature of the included studies, we were unable to evaluate a temporal relationship between the conditions (i.e., the occurrence of OSA before or after the development of MCI in patient population). Nevertheless, the clinical impact inferred due to the additive burden of these two disorders demands a closer look into their relationship.
Population recruitment locations
The studies were conducted in six different countries and enrolled a patient population from a community, general (i.e., public health centers) or specialty neurology clinics that likely contributed to the variations in OSA rates. A clinic-based elderly population is more likely to have individuals with undiagnosed OSA with accompanying major comorbidities that will primarily prompt these patients to seek help. In turn, OSA may remain undiagnosed in this population due to OSA symptoms of, for example [29], memory and concentration being falsely attributed to the aging process or to other disorders by clinicians, hence, transiently decreasing the OSA point prevalence in the clinic-based sample. Alternatively, OSA prevalence rates in MCI are more discernable in a community-based sample and are likely a better representative of the target population. The study utilizing such population reported an OSA prevalence of 27 and 26% in patients with and without MCI, respectively [24]. This rate closely resembled the OSA prevalence (AHI ≥ 5) of 30% estimated in elderly patients between the ages of 50 and 70 years, in a recent epidemiological study [30]. Hence, the high OSA prevalence noted across studies using a clinic-based sample may not be an appropriate representation of the target population (MCI or controls) and may in fact be the result of selection bias.
Method of OSA diagnosis or screening
The five studies included in this review used different diagnostic methods and criteria to diagnose OSA, which could partially explain the considerable variation in the prevalence rates of OSA. A diagnosis of OSA is made based on an AHI ≥ 5 events/hour for patients reporting symptoms of OSA (e.g. snoring, daytime sleepiness). The prevalence of OSA in MCI for PSG studies using an AHI ≥ 5 was 70 and 71%, respectively [27, 28]. The use of differing definitions for hypopneas in PSG studies has been shown to result in significant variations in the AHI value, which can drastically alter OSA prevalence rates [31]. Both studies, however, had similar apnea and hypopnea definitions according to the American Academy of Sleep Medicine (AASM) guidelines [32]. Diagnostic testing can also be performed using types II-IV portable sleep monitors. One of the five studies used the ApneaLink device and an AHI ≥ 15 events/ hour to diagnose moderate-to-severe OSA demonstrating a prevalence of 27% among the 230 participants with MCI [24]. ApneaLink is a type III portable monitor commonly used for home sleep testing to screen OSA. In adults with moderate-to-high severity of OSA, the ApneaLink has a sensitivity of 75% and specificity of 87% [33]. The specificity of ApneaLink drops to 62% with an AHI cutoff value of 5 that results in mild OSA being undiagnosed [33]. The exclusion of patients with mild OSA in this study population would have resulted in a lower OSA prevalence rate. The Berlin questionnaire [32] is used to screen for high risk patients with OSA and has a pooled sensitivity and specificity of 76 and 45%, respectively, to identify patients with an AHI ≥ 5 events/hour. The study that used the Berlin questionnaire reported a prevalence of 59% among the 138 individuals with MCI [26]. Finally, the study with the lowest prevalence rate employed a patient population with a self-reported OSA diagnosis [2]. The type of OSA diagnostic metric used was not reported. The use of self-reported symptoms would result in significant underestimation of OSA as patients with OSA may be asymptomatic, hence the comparatively low prevalence rate observed in this study. Furthermore, the study used data from the Alzheimer’s disease Neuroimaging Initiative (ADNI) cohort that enrolled only aMCI patients, hence, the associated memory impairments could have partially accounted for the lack in reported information about a previous OSA diagnosis leading to an underestimation of the true prevalence of OSA. Therefore, in memory clinics, a more standardized approach, preferably using objective sleep measurements, needs to be taken when estimating the prevalence of OSA.
Evidence on association between OSA and MCI
Although several prospective cohort studies [1, 16] have demonstrated that patients with OSA have greater neurocognitive deficits, the risk of OSA and subsequent onset of MCI is seldom explored. In the above mentioned ADNI cohort database, patients with OSA had a younger age onset of MCI by a decade compared to those without OSA, even after adjusting for possible confounding variables [2]. Moreover, continuous positive airway pressure (CPAP) therapy conferred a protective effect, essentially delaying the onset of MCI in those individuals being treated for OSA. Similarly, a number of studies have demonstrated a partial reversibility in cognitive dysfunction with CPAP therapy in individuals with OSA, particularly in the domains of attention, vigilance, executive function and memory [34,35,36]. Finally, a meta-analysis of cross-sectional studies demonstrated that individuals with AD had a 5-fold risk of OSA compared to healthy age-matched controls [17]. Contrary to this, with the exception of one study [2], there were no significant differences in the risk of OSA among individuals with MCI vs. controls. Perhaps, the additive pathological processes and severity of AD makes these individuals prone to OSA development, which may not be present in those with MCI or early AD (i.e. reverse causation). Nonetheless, the results of these studies signify the importance of early recognition and treatment of OSA in possibly diminishing or delaying the future risk of MCI.
Several mechanisms may contribute to the neurocognitive decline in individuals with OSA, including disturbances in oxidative stress, sympathetic activation, endothelial dysfunction, and systemic and vascular inflammation [37, 38]. Long-standing OSA leads to recurrent intermittent hypoxia and alters sleep architecture, which may lead gradually to brain neurodegenerative processes [39]. A recent review proposed several possible mechanisms linking OSA to dementia, highlighting the important roles chronic sleep architecture impairments may play in neurogenesis, synaptic plasticity and memory consolidation [39]. A neuroimaging meta-analysis assessing the neuro-structural differences between patients with OSA and healthy controls reported significant grey matter reductions in the bilateral parahippocampus, left temporal and right frontal lobes of OSA patients [40]. Whilst there is an adverse impact of OSA on the healthy young brain, and this is greater with the aging middle-aged brain [41], the natural assumption is that the additive burden of OSA may exert greater deleterious effects especially to the elderly brain. Untreated OSA can potentially accentuate the progression of MCI and Alzheimer’s disease [2] in cognitively intact individuals due the accumulation of Alzheimer’s disease biomarkers (amyloid beta and tau proteins) [42, 43], through hypoxic insults and/or disrupted sleep architecture [39].
Limitations
Our results should be interpreted with caution since this study has some limitations. First, most of the included studies in our review had a small sample size. A small sampling population can lead to an overestimation of the magnitude of an association and ultimately produce high false-positives. Moreover, it may be difficult to interpret the results from studies with a small sample size due to a wider 95% CI that may lead to an imprecise estimate of the effect. Second, we only looked at studies in English language that may have limited our final study count. Third, with a small number of studies and individuals representing this population, difficulties can arise when attempting to conduct a pooled analysis (i.e. meta-analysis), while adjusting for confounding factors, which can lead to unreliable results. Finally, due to the cross-sectional design of the included studies, evaluating a temporal relationship and associations identified are difficult to interpret.