© 2002 American Thoracic Society
Epidemiology of Obstructive Sleep ApneaA Population Health PerspectiveDepartment of Population Health Sciences, University of Wisconsin—Madison, Madison, Wisconsin; and Department of Medicine, Boston University School of Medicine, Boston, Massachusetts Correspondence and requests for reprints should be addressed to Terry Young, Ph.D., Department of Population Health Sciences, University of Wisconsin—Madison, 502 N. Walnut Street, Madison, WI 53705. E-mail: tbyoung{at}facstaff.wisc.edu
Population-based epidemiologic studies have uncovered the high prevalence and wide severity spectrum of undiagnosed obstructive sleep apnea, and have consistently found that even mild obstructive sleep apnea is associated with significant morbidity. Evidence from methodologically strong cohort studies indicates that undiagnosed obstructive sleep apnea, with or without symptoms, is independently associated with increased likelihood of hypertension, cardiovascular disease, stroke, daytime sleepiness, motor vehicle accidents, and diminished quality of life. Strategies to decrease the high prevalence and associated morbidity of obstructive sleep apnea are critically needed. The reduction or elimination of risk factors through public health initiatives with clinical support holds promise. Potentially modifiable risk factors considered in this review include overweight and obesity, alcohol, smoking, nasal congestion, and estrogen depletion in menopause. Data suggest that obstructive sleep apnea is associated with all these factors, but at present the only intervention strategy supported with adequate evidence is weight loss. A focus on weight control is especially important given the expanding epidemic of overweight and obesity in the United States. Primary care providers will be central to clinical approaches for addressing the burden and the development of cost-effective case-finding strategies and feasible treatment for mild obstructive sleep apnea warrants high priority.
Key Words: obstructive sleep apnea • sleep-disordered breathing • epidemiology • sleep disorder
Brief History of Undiagnosed Obstructive Sleep Apnea Interpretation of Observational Studies of Obstructive Sleep Apnea Occurrence of Obstructive Sleep Apnea Prevalence Incidence and Progression Outcomes: What Is the Cost of Obstructive Sleep Apnea in the Population? Hypertension Cardiovascular Morbidity and Mortality Sleepiness Cognitive Function Health-related Quality of Life Motor Vehicle Crashes and Occupational Accidents Impact on Pregnancy Risk Factors with Particular Population Health Significance Excess Body Weight Alcohol Smoking Nasal Congestion Menopause Obstructive Sleep Apnea in Children Prevalence Risk Factors Consequences Conclusion Appendix The basic epidemiological features of obstructive sleep apnea (OSA), a condition characterized by repeated episodes of apnea and hypopnea during sleep, are well established. Undiagnosed OSA is common in adults, the severity spectrum is wide, and the cardiovascular and behavioral morbidity seen in patients with OSA is also associated with undiagnosed OSA (1–3). There is no controversy regarding the need for better recognition and treatment of severe, symptomatic OSA, but critical questions remain regarding what clinical and public health approaches are needed to address mild to moderate OSA, for which the prevalence is particularly high (3–6). Should case finding or population screening be encouraged? Are risk factor intervention strategies feasible means of reducing the incidence and progression of mild OSA? Formulating answers to these questions requires a greater understanding of the natural history and morbidity attributable to OSA over its severity spectrum and identification of modifiable factors that initiate OSA and affect its progression. In this review we focus on data from population-based epidemiology studies, with the goal of addressing some of the questions raised by the high prevalence of undiagnosed OSA.
Brief History of Undiagnosed Obstructive Sleep Apnea Findings from early clinical and community cross-sectional studies of OSA and hypertension, myocardial infarction, and other cardiovascular disease were mixed and generated considerable controversy (6, 12, 13). It was clear from the low clinical recognition of sleep apnea that only a small fraction of all cases even of severe OSA was being diagnosed (1, 14), so selection bias in studies of patients with sleep apnea was a serious concern. Furthermore, the validity of most studies was questioned because of small samples, inadequate control for obesity and other potential confounding factors, and other methodological limitations. The need for population-based, longitudinal studies of the natural history and adverse health consequences of OSA was widely recognized and large cohort studies were initiated (8, 10, 11, 15–19). Findings from these studies relevant to the significance of undiagnosed OSA are now emerging, and progress has been made toward overcoming methodological drawbacks of case-based and cross-sectional investigations of this condition. The major cohort studies discussed in this review are described briefly in the APPENDIX.
Interpretation of Observational Studies of Obstructive Sleep Apnea In most epidemiology studies, and accordingly in this review, OSA is defined by the number of obstructive apnea and hypopnea episodes per hour of sleep (apnea–hypopnea index, AHI), reflecting the degree of departure from the normal physiology of breathing during sleep. The term "OSA syndrome" will be used to indicate a clinical entity defined by an elevated AHI in conjunction with hypersomnolence or related problems in daytime function and is synonymous with the term "obstructive sleep apnea–hypopnea syndrome." The spectrum of sleep-related obstructed breathing is considered by many researchers to include increased upper airway resistance manifested as snoring without frank apnea or hypopnea events (20) and episodic flow limitation terminating in central nervous system arousals—often called "upper airway resistance syndrome" (21). These conditions may represent the earliest stages of OSA and may be important in investigating disease progression or may be distinct conditions, but to date there are few epidemiological data based on objectively measured nonapneic snoring or on respiratory effort-related arousals. Although a detailed discussion of the pathophysiology of airflow obstruction in OSA is outside the scope of this review, it is clear that upper airway collapse most often results from a combination of anatomic factors that predispose the airway to collapse during inspiration plus neuromuscular compensation that is insufficient during sleep to maintain airway patency. The relative contribution of anatomic versus neuromuscular factors is likely to vary greatly among individuals and may vary considerably among groups defined on the basis of age, sex, body habitus, race, and ethnicity, although there are no data from epidemiological studies with which to address this. In contrast to OSA, central sleep apnea is characterized by repeated episodes of apnea or hypopnea resulting from decreased neural output to respiratory motoneurons, without airflow obstruction. There may be some overlap in the pathophysiology underlying these conditions, as reducing neural output to both the diaphragm and pharyngeal dilator muscles may lead to central apnea or hypopnea if the upper airway is not anatomically prone to collapse, but to obstructive apnea or hypopnea if the airway is more collapsible (22). In general, however, the pathophysiology, epidemiology, and clinical characteristics of central and obstructive sleep apnea are distinct, and this review addresses only the latter. It is important to note that the methods employed in most epidemiological studies, and indeed in many sleep laboratories, although good at distinguishing central from obstructive apneas, cannot reliably discriminate between central and obstructive hypopneas. Hypopneas, which are in general much more common than apneas, are typically included in the AHI and assumed to reflect obstructive respiratory events, although the validity of this assumption, particularly in subgroups of the population such as the elderly, remains to be demonstrated. Epidemiology studies have used a variety of methods to measure OSA, including in-laboratory polysomnography; unattended in-home polysomnography; unattended polygraphic or other recording of a few physiological parameters; and self-report data on markers of OSA, such as habitual snoring. Studies using objective measures of apnea and hypopnea have employed variable respiratory event definitions, with differing requirements for fractional decrease in airflow, oxyhemoglobin desaturation, or associated evidence of cortical or autonomic arousal. Moreover, there has been no standardization in the methodology used to quantify airflow, with methods such as thermistry, inductance plethysmography, and nasal cannula/pressure transducer systems providing different sensitivities to changes in airflow. Like other conditions based on a severity continuum, the definition of the units of the continuum and the ultimate thresholds used to designate the presence of OSA will affect the magnitude of prevalence and estimates of associations with risk factors and outcomes. The use of more restrictive definitions of apnea and hypopnea, higher AHI cutpoints, or an additional requirement for symptoms of sleepiness will obviously lower prevalence estimates and affect values expressing associations, such as odds ratios (23–25). The traditional use of the AHI, while having substantial face validity as a summary parameter of OSA, does not reflect a large body of empiric data demonstrating that the AHI fully and succinctly captures those physiologic aspects of OSA that are most germane to the pathogenesis of adverse consequences of OSA. Certainly it is possible that the AHI and other parameters, such as duration of hypoxemia or degree of sleep fragmentation, are of varying relative importance among the many sequelae of OSA, such as sleepiness, cognitive impairment, and hypertension. Although there has been interest in developing alternatives to the AHI, event frequency was used as the basis for describing OSA by a task force with the aim of standardizing measurement techniques and syndrome definitions for sleep-related breathing disorders (26). Recommended diagnostic criteria for OSA syndrome include an AHI of 5 or more, determined by overnight monitoring, and evidence of disturbed or unrefreshing sleep, daytime sleepiness, or other daytime symptoms. The task force suggested AHI cutpoints of 5, 15, and 30 events/hour to indicate mild, moderate, and severe levels of OSA. These recommendations were acknowledged to be an expert consensus statement based on a paucity of objective data, and intended to stimulate further research to identify the optimal approach to quantifying sleep-related breathing disorders. Precision is often neglected when comparing findings of various studies, and a false sense of disagreement among studies can occur when only the point estimates are considered. Measurement error and sample size affect the precision of prevalence and other point estimates and, consequently, confidence intervals are necessary to interpret findings. Less importance can be placed on a point estimate when confidence intervals are wide or unreported. Finally, findings from both case-based and population studies can be seriously flawed by inappropriate sample sources, participation bias, and loss to follow-up. In particular, a spurious estimate of prevalence or the association of interest can result if selection, participation, or dropout is related to the study factors. For example, in a case–control study, choosing a sample of patients with OSA from sleep clinics where referral from cardiovascular clinics is high would lead to an overestimation of the association of OSA and cardiovascular disease (CVD). A similar bias would result in a population-based study if participation rates were higher for snorers with hypertension, compared with that of snorers without hypertension. To assess and correct possible bias due to sample construction, a probability sample drawn from a sampling frame that provides some information about participants as well as nonparticipants is needed.
Prevalence Understanding disease prevalence, that is, the proportion of a population with the condition, is critical to anticipating health care needs and allocating appropriate resources. In addition, comparisons of prevalence by demographic factors may yield etiological clues and identify subgroups at particularly high risk for targeted case finding. Prevalence studies conducted over the past decade provide considerable data from diverse populations to estimate the health burden of OSA and to explore interesting aspects of its occurrence. However, prevalence estimates are extremely vulnerable to the methodological issues discussed above. Previous reviews of OSA prevalence have taken some of these issues into account by roughly adjusting for differences in definitions or by comparing results from studies with similar study designs. Davies and Stradling (27) analyzed 12 studies of OSA prevalence in Western populations and, using conservative approaches to account for methodological differences, estimated that 1 to 5% of adult men have OSA syndrome (i.e., frequent apnea and hypopnea episodes and daytime sleepiness). Lindberg and Gislason (2) considered prevalence estimates for undiagnosed OSA syndrome from nine studies that all used two-stage sampling procedures in which sleep studies were conducted on subsets of participants drawn from large-sample surveys. A strength of this type of study design is that subsets are drawn from a defined probability sample, permitting some evaluation of participation bias. Furthermore, the two-stage design, with oversampling of important subgroups and weighting of results to the survey sample, increases study efficiency and allows extrapolation of the subgroup-specific prevalence estimates to populations of different composition. Prevalence of undiagnosed OSA syndrome in these studies ranged from 0.3 to 5%; samples from countries with lower mean body mass index (BMI) tended to yield lower prevalence estimates. However, some of these estimates are based on the extremely conservative assumption that all survey participants who did not report snoring and sleepiness were free of OSA. This assumption is almost certainly false and could lead to a serious underestimation of prevalence. Thus, up to 5% of adults in Western countries are likely to have undiagnosed OSA syndrome, and hence be candidates for treatment. Consequently, in addressing the burden of OSA, the resources to identify and treat up to 5% of middle-aged adults with OSA constitute a minimum need. Not counted in this estimate is the large proportion of adults with OSA, defined by frequent episodes of apnea and hypopnea, who do not report sleepiness. At present, the clinical significance of OSA without overt daytime symptoms is controversial and the public health significance remains to be determined. Many studies, however, have shown adverse health outcomes to be associated with OSA regardless of the presence of sleepiness (discussed in the following section) and, consequently, the prevalence of OSA as determined solely by abnormal breathing during sleep is extremely important in understanding the potential OSA burden in the population.
Prevalence estimates from studies with probability samples range, for OSA of at least mild severity (defined by AHI
Ethnicity. OSA prevalence has been established in few populations other than those of Western nations, and therefore the worldwide importance of OSA, as well as potentially important racial or ethnic prevalence patterns, are poorly understood. Epidemiologists have traditionally investigated geographical distributions of disease occurrence to find etiologic clues, but it is often hard to disentangle environmental risk factors, including cultural differences in diet and lifestyle, from genetic factors. When there are large differences in disease prevalence between countries, studies of whether the rates change in migrant populations and their succeeding generations have often been enlightening. At present, data from studies of groups other than white subjects are too sparse even to determine with confidence if prevalence differs worldwide. Population-based studies suggest that OSA prevalence is as high or higher in African-Americans compared with Caucasians. Ancoli-Israel and coworkers (29) studied community dwelling adults, age 65 years or greater, by in-home monitoring, and found that the odds of having an AHI of 30 or higher was 2.5 times greater in African-Americans relative to Caucasians, controlling for BMI and other confounding factors. In the Cleveland Family Study, a racially heterogeneous sample of families with one index OSA case and neighborhood control subjects, Redline (30) found that in participants less than 25 years of age, the prevalence of OSA (adjusted for BMI and other potentially confounding factors) was higher in AfricanAmericans than in Caucasians. On the basis of in-home polysomnography on more than 6,000 participants in the multicenter Sleep Heart Health Study, however, the prevalence of OSA was not higher in African-Americans compared with Caucasians after adjustment for age, sex, and BMI (31). Unfortunately, there are no data about OSA prevalence in African countries for comparison. Ip and colleagues (32) reported the first estimates of OSA prevalence in an Asian population, using two-stage sampling methodology and in-laboratory polysomnography. From a survey sample of 784 Hong Kong men, 30 to 60 years of age, 153 completed polysomnography studies; of these, 25% had AHI of 15 or more events per hour, but the authors found evidence of self-selection of men most likely to have OSA. To account for this bias, the authors made a conservative assumption that there were no cases of OSA among the nonparticipants, and estimated the prevalence of OSA defined as an AHI of 15 or more to be 5%, and of OSA syndrome defined as an AHI of 5 or more plus excessive daytime sleepiness to be 4%. Preliminary data from a similar study of Chinese women in Hong Kong indicated a conservative estimate of OSA syndrome prevalence of 2% (33). The similarity of the estimates from the Hong Kong studies with those reported from Western nations are provocative because obesity, a strong risk factor for OSA, is prevalent in white populations, but is relatively uncommon in Asian countries. Ip and coworkers (32) noted that whereas BMI and other measures of body habitus were associated with OSA in the Hong Kong study, the correlations were weaker than those seen in studies of white subjects. In view of the higher than expected prevalence, these investigators hypothesized that other strong OSA risk factors that are more prevalent in Chinese relative to Western populations, such as craniofacial features that compromise the upper airway, must exist. Some, but not all, clinical observations of Asian patients with OSA support this hypothesis (34–36). Other studies suggest that correlates of OSA may differ by race. Redline and coworkers (37) found that the association of BMI with OSA in younger participants was stronger in Caucasians than in African-Americans in the Cleveland Family Study. Furthermore, associations of head form and craniofacial measures with OSA in this sample were stronger in Caucasians (37, 38). Race-specific risk factors were also reported from a comparison of patients with OSA in New Zealand (39). Again, BMI was a stronger predictor of OSA severity in Caucasians compared with Polynesians, and the specific craniofacial risk factors associated with OSA severity differed by race. More information is clearly needed about the distribution of OSA in countries in Africa, Asia, and the South Pacific to understand the worldwide burden of OSA. Furthermore, because environmental and genetic risk factors for OSA vary considerably among these and Western countries, studies of OSA risk factors in resident populations, as well as in migrants and their offspring born in Western nations, would be illuminating.
Sex. Pregnancy may be a period of particular risk for OSA in women. In a survey completed by 350 women at two U.S. Army hospitals during the second or third trimester of pregnancy and by 110 nonpregnant women, 14% of the pregnant women reported snoring often or always, compared with only 4% of the nonpregnant women (48). Both frequency and loudness of snoring, and episodes of awakening with a choking sensation, appear to increase during pregnancy, with half of the women in one study reporting snoring and 14% reporting choking awakenings at 35 to 38 weeks of gestation, versus 37 and 4%, respectively, at 8 to 12 weeks of gestation (49). A questionnaire administered to 502 Swedish women at the time of delivery found that 23% reported snoring often or always during the week before delivery, whereas only 4% reported snoring before pregnancy. Most of the increase in snoring occurred in the third trimester (50). There is evidence that the impact of pregnancy on snoring resolves within several months after delivery (51). The high prevalence of snoring and choking awakenings during pregnancy suggests that pregnancy may be associated with OSA; however, there are few data regarding the prevalence of OSA during pregnancy. In the largest reported study, polysomnography was performed in 11 snoring women early in the third trimester. All had an AHI less than 5, although all had evidence of increased upper airway resistance characterized by either crescendo respiratory effort or abnormal sustained increases in respiratory effort, occurring more commonly than in nonsnoring control subjects (51). The mechanisms underlying the increase in snoring during pregnancy are uncertain, but may include excess weight gain (50, 52), diffuse pharyngeal edema of pregnancy (53), or the effect of sleep deprivation on pharyngeal dilator muscle activity.
Age. Several studies have found OSA to be highly prevalent in people older than age 65 years. In the first large population-based study of older people, Ancoli-Israel and coworkers (55) conducted in-home polygraphy on a probability sample of 427 men and women 65 to 95 years of age. OSA defined as an AHI of 10 or more occurred in 70% of the men and 56% of the women, approximately 3-fold higher than the prevalence estimates for OSA in middle age given in Table 1.
Four subsequent cohort studies have samples with wide age ranges, allowing internal comparisons of prevalence in broad categories of older age and middle age without the problem of interstudy methodological differences, such as measurement and definition of OSA. Bixler and coworkers found that in both men and women, those 65 to 100 years of age had a prevalence of OSA that was approximately twice as high as the upper bounds of the 95% confidence interval (95% CI) for OSA prevalence in middle age, shown in Table 1 (15, 16). The largest prevalence difference between middle and older age was found in the Vitoria-Gasteiz, Spain Cohort (17, 56). For ages 71–100 years, the prevalence of AHI These observations raise two critical questions: does aging over the middle to older years per se have an etiological role in OSA, and what is the significance of the high prevalence of OSA in older age? A higher prevalence of OSA in older versus middle age does not necessarily mean that physiological changes associated with later life cause an increase in the OSA incidence rate: The prevalence of an unremitting, nonfatal disease would be expected to increase with age as cases accumulate from a constant or even declining incidence rate. Differences in age-specific prevalence beyond what would be expected by accumulation of cases would indicate that changes in OSA mortality rate, incidence rate, or both occur with aging. Because we do not know the incidence or mortality rate of OSA at any age, we can only speculate as to why the prevalence of OSA is high in older people. If aging actually leads to an increase in OSA incidence, we would expect prevalence to continue to rise over the older age range; however, the studies by Ancoli-Israel and coworkers (55), Bixler and coworkers (15, 16), and the Sleep Heart Health Study (31) all suggest that most of the age-related prevalence increase occurs before age 65 (Figure 1) . A plateau in OSA prevalence at some point after age 65 years necessitates a new perspective on the occurrence of OSA in older people. Unless incidence drops with older age, either the mortality rate of persons with OSA relative to those without OSA must increase, or else OSA must remit. Certainly the mortality rates of many chronic diseases increase with age, but at present there is no solid evidence that OSA causes death. Similarly, there is no evidence for remission of OSA with aging.
Some data suggest that OSA in older age may be a condition distinct from that of middle age. Several studies of OSA in older populations report little or no association of OSA with sleepiness, hypertension, or decrements in cognitive function (57–61), all common correlates of OSA in middle age. Eighteen-year follow-up data from the San Diego sample of older adults indicated that changes in BMI were weakly associated with change in AHI (62) and that the association of obesity with AHI was weaker in older compared with middle-aged participants in the Sleep Heart Health Study (31). Furthermore, despite the high prevalence of OSA, the prevalence of self-reported snoring, a strong marker for OSA, clearly decreases past middle age (15, 63). One possible explanation for this paradox is that bed partners may no longer be alive or may be unable to report validly on snoring, owing to age-related conditions such as hearing loss. Alternatively, older compared with younger adults may be more likely to have central sleep apnea or a prominent central neuromuscular component to their OSA. Central apnea and hypopnea are not associated with snoring and may be less likely to elicit daytime symptoms. Data to evaluate this theory are sparse, but findings from the cohort of Bixler and colleagues (15) offer some support. Central sleep apnea, defined by 20 or more central apnea events/hour of sleep, was nonexistent before age 65 years, but occurred in 5% of the sample over age 65 years. In contrast, despite the higher prevalence of OSA in those over age 65 years, the prevalence of OSA syndrome, defined as an AHI of 10 or more plus symptoms (including daytime sleepiness and hypertension), was actually lower in those over 65 years (1.7%) than in those age 45 to 64 years (4.7%), although the difference was not significant. In summary, the occurrence of OSA in older people is more complex than previously appreciated. The scant data available suggest that instead of a continual rise in prevalence with age due to accumulating cases, prevalence tends to level off after age 65 years. This trend, if correct, implies either a relative increase in the mortality rate from OSA or a remission of OSA with aging. However, it is possible that biases, including poor measurement of OSA in older people or birth cohort effects, explain some or all of the age-related prevalence trends seen in cross-sectional studies. More importantly, a better understanding of sleep-related breathing disorders in older age and how they differ, if at all, from the typical OSA of middle age is crucial for proper clinical management of older patients. Prospective data to investigate sleep-related breathing disorders and aging, as well as the role of OSA in increased mortality, are clearly needed.
Incidence and Progression Only preliminary findings of OSA progression are currently available from population studies. Data from baseline and 8-year follow-up studies of 282 participants in the Wisconsin Sleep Cohort show a significant increase in OSA severity over this interval (Table 2): The overall mean AHI increased by 2.6 events/hour, from 2.5 at baseline to 5.1 at follow-up. There were significant increases in mean AHI in all strata of sex, BMI, age, and snoring. Whereas the change in mean AHI was not significantly greater in men compared with women, progression was significantly greater in obese compared with non-obese, older compared with younger, and habitually snoring compared with nonhabitually snoring subjects. In Figure 2 , AHI at 8-year follow-up is plotted against baseline AHI. Although there is a considerable amount of scatter in the plot, two patterns are evident. First, consistent with the elevation in mean AHI, there are more persons whose AHI increased than decreased. Second, many people who had an AHI of 1 or less at baseline increased to an AHI greater than 1 at follow-up, whereas few individuals with an AHI over 1 at baseline decreased to an AHI of 1 or less at follow-up.
Preliminary data from the Cleveland Family Study demonstrate trends similar to the Wisconsin data. Redline and coworkers (64) monitored 232 participants from the Cleveland Family Study with AHI less than 5 at baseline and reported a 5-year increase from a baseline mean (SD) AHI of 2.0 (1.4) to a follow-up AHI of 6.2 (7.9). Significant predictors of higher AHI at follow-up were excess body weight, central obesity, cardiovascular disease, and diabetes. Because of the lack of older participants, neither the Wisconsin Sleep Cohort nor the Cleveland Family provides insight into disease progression beyond age 65 years, but 18-year follow-up data from the San Diego Older Adult cohort showed little change in AHI with aging (62). In follow-up studies with small sample sizes (n = 11 to 55) of community-dwelling persons monitored for 3 to 8 years, most (65–67), but not all (68), found evidence of progressive increase in AHI over time. In two studies of patients with mild to moderate OSA who had refused treatment and were later restudied, significant progression was found (69, 70); however, in a study of patients with severe OSA (mean AHI = 52), no net progression, and some individual regression, was seen after 5 years (71). The validity of these studies may have been compromised by incomplete follow-up and, in the clinic-based studies, by regression to the mean resulting from the selection of participants with evidence of substantial OSA. However, the available data from both population and clinic studies suggest that change in severity, mostly toward progression, does occur in individuals with mild or moderate OSA. The available epidemiological evidence suggests that substantial progression of OSA can occur over relatively short time periods. An efficient method of recognizing individuals likely to develop severe OSA would allow interventions to reduce or reverse OSA progression before the development of significant morbidity. It appears that habitual snoring and obesity may be useful markers of risk for OSA progression, although data remain limited. One particularly difficult methodologic issue in characterizing OSA incidence and progression is the worsening epidemic of obesity in the United States (72). Any temporal nonuniformity among studies or differences among sex, race, and age subgroups in secular trends in obesity will likely be reflected in progression trends of OSA. Thus, researchers attempting to characterize underlying age-related natural history patterns of OSA face the difficult task of separating these patterns from contemporaneous secular trends in obesity prevalence. In any event, as the overall prevalence of obesity increases, it is reasonable to expect an increase in prevalence and severity of OSA beyond the high levels already seen.
Obstructive sleep apnea is associated with conditions that account for the leading causes of mortality in adults: hypertension, cardiovascular, and cerebrovascular diseases. In addition, several neurobehavioral morbidities that are of potentially great public health and economic importance are linked with OSA, including daytime sleepiness and impaired cognitive function that may, in turn, contribute to motor vehicle crashes and job-related accidents (73).
Hypertension Most earlier epidemiological studies assessing the OSA–blood pressure association reflected a trade-off between state-of-the-art OSA assessment in inherently unrepresentative clinic samples (80–88) and methodologically more rigorous population-based studies that employed OSA assessment instruments with poor or unknown validity (e.g., self-report of OSA symptoms or at-home nocturnal oximetry) (89–95). Collectively, these studies shed little light on the connection between OSA and hypertension, with findings ranging from no association whatsoever to strong associations. Reviewers of this earlier research concluded that because of the varying results and the potential biases due to the use of clinic samples, crude assessment of OSA, or inadequate control of important confounding factors, an independent association had not been established (12, 77, 96, 97). Since those earlier reviews, several epidemiologic studies have consistently found positive associations between OSA and hypertension. These studies, described below, have used a variety of designs, but most have had large samples and all have attempted to rigorously account for important confounding factors such as obesity, age, and sex. The uniformity of these positive results has lead some researchers to conclude that OSA should be considered a cause of secondary hypertension (98) and that the controversy over the presence of a causal association is passé (78). These most recent findings and the related question of whether treating OSA can lower blood pressure are the focus of this section.
Four large cross-sectional population-based studies and one prospective population-based study have estimated associations between polysomnographically assessed AHI and daytime hypertension (use of antihypertensive agents or systolic blood pressure Although these four studies have had compatible results, because of their cross-sectional design, none has been able to demonstrate that OSA predated hypertension. This issue was addressed in a prospective analysis from the Wisconsin Sleep Cohort (102, 103). Even minimally elevated AHI at baseline (0 < AHI < 5 events per hour) was associated with 42% (95% CI, 13 to 78%) increased odds of developing hypertension over a 4-year follow-up period. A dose–response relationship was observed for more severe categories of AHI, with an odds ratio of 2.9 (95% CI, 1.5 to 5.6) for an AHI of 15 or greater versus an AHI of zero events per hour. However, although the dose–response trend was significant, there appeared to be a plateau in the hypertension response at high levels of OSA severity. Such a plateau was also suggested by the cross-sectional analyses from Spain (17) and from the Sleep Heart Health Study (99). The results of these studies are also consistent with those seen in other studies of OSA with different study designs. In a longitudinal analysis of the large Nurses' Health Study (n = 73,231), Hu and coworkers (104) found that self-reported snoring at baseline moderately increased the risk of subsequent development of hypertension over an 8-year follow-up period. Grote and coworkers (105) and Lavie and coworkers (106) found significant associations between polysomnography-assessed OSA and increased blood pressure or the occurrence of hypertension in large samples of sleep clinic patients. Davies and coworkers (107) assessed 24-hour ambulatory blood pressure in 45 persons with OSA and 45 persons without OSA, matched on several factors including age, BMI, alcohol and cigarette use, and heart disease. The patients with OSA had higher daytime and nighttime blood pressures and demonstrated an attenuated nighttime reduction in blood pressure compared with the matched participants without OSA. Finally, in a large community-sample British study, Stradling and colleagues (108) found that the degree of overnight reduction in blood pressure (evening minus morning blood pressure) was stunted in proportion to the severity of oxygen desaturation and excess respiratory effort during sleep. This finding is consistent with an earlier study examining the evening-to-morning blood pressure difference in patients with varying degrees of OSA (84) and findings of attenuated nighttime blood pressure "dipping" in patients with OSA (86, 107, 109). Although the associations found in observational studies suggest a causal, albeit not strong, relationship between OSA and elevated blood pressure, the potential for remediating hypertension by treating OSA is unclear. This is a particularly important issue given the finding that patients with severe OSA may be relatively likely to have hypertension that responds poorly to pharmacotherapy (110). The effectiveness of reducing blood pressure by treating OSA with continuous positive airway pressure (CPAP) has been addressed in numerous intervention studies (21, 81, 87, 98, 111–121). These studies showed mixed results. However, few of them used appropriate placebo-controlled comparison groups, an important consideration given the demonstration of a placebo response of blood pressure to sham CPAP (115). Among studies using placebo control groups, results have also been inconsistent. Barbé and coworkers (114) found no effect after 6 weeks of CPAP on 24-hour blood pressure measures in a study of 29 treatment and 25 sham CPAP-control patients with OSA. Three other placebo-controlled studies found small to moderate effects of CPAP on ambulatory blood pressure that varied in degree depending on time of day (i.e., sleeping or awake) or among subsets of treated patients (116, 117). Thus, it remains uncertain to what degree CPAP use can lower daytime blood pressure in most patients with OSA. This may reflect methodologic issues such as inadequate study power, given the fairly modest magnitude of the effect of OSA on blood pressure. A more important consideration, however, is the possibility that chronically elevated blood pressure due to OSA leads to vascular damage, which causes hypertension to persist even when the OSA is treated. Whether long-term CPAP therapy might result in reduced blood pressure in such patients is unknown. There is a growing consensus that OSA is an important risk factor for hypertension independent of excess weight and other potentially confounding factors. An association appears to be present even at the mild end of the OSA severity spectrum. Despite the generally modest magnitude of the association, the high prevalence of OSA implies that it may be responsible for a substantial portion of the population burden of hypertension. However, the potential for remediating hypertension by treating OSA remains unclear.
Cardiovascular Morbidity and Mortality Several case–control studies of patients assessed for OSA after myocardial infarction (MI) support an association between the two conditions, with odds ratios generally in the range of 4.1 to 4.5 in both men and women (130–132). The odds ratio for MI associated with OSA was as high as 23 in one study (133); however, the confidence interval was wide (95% CI, 4 to 140). Although these case–control studies all demonstrated strong associations between OSA and MI, they also share an important limitation: OSA status in cases was assessed after an MI, and therefore may be a poor surrogate for OSA severity during the relevant etiologic time frame before the occurrence of MI. This is especially true if the MI itself could affect OSA severity, by virtue of changes in cardiac function, medication use, or peri-infarction sleep deprivation. Cross-sectional epidemiologic studies of objectively measured OSA or self-reported snoring and cardiovascular disease (CVD) have found a positive association, although of considerably smaller magnitude than that observed in case–control studies. Schmidt-Nowara and coworkers (94), in a population sample of 1,222 Hispanic Americans, found an elevated odds of self-reported CVD in snorers that was of borderline significance (odds ratio, 1.8; 95% CI, 0.9 to 3.6). Olson and coworkers (134) assessed OSA by in-home monitoring and found a nonsignificant elevated odds of self-reported CVD in an Australian sample (n = 441). The adjusted odds ratio for coronary artery disease in persons classified as having OSA versus nonsnorers without OSA was 1.4 (95% CI, 0.4 to 4.5). In both these studies, lack of significance could have been a result of insufficient study power. Shahar and coworkers (135) found a significant cross-sectional association of OSA with prevalent CVD in persons undergoing in-home polysomnography in the Sleep Heart Health Study. Among the 6,424 participants, those in the upper quartile of AHI (11.0 events or more per hour) had a 42% (95% CI, 13 to 78%) greater odds of prevalent CVD (including coronary heart disease, stroke, and congestive heart failure) than participants in the lowest quartile (AHI less than 1.3 events per hour), after adjusting for multiple potential confounders. Additional analyses examining the association of OSA and CVD along the entire spectrum of OSA severity suggested that most of the elevation in risk of CVD occurs as the AHI rises from zero to 10 events per hour. The analysis included adjustment for hypertension, suggesting that hypertension is not the only mechanism by which the risk of cardiovascular sequelae is heightened in persons with OSA. If confirmed, this would seem to imply that pharmacologic treatment of hypertension would not fully insulate these patients from heightened cardiovascular risk. The only prospective data on OSA and CVD come from three large population-based studies of snoring and incident CVD. In two of these studies, the magnitude of the increased risk of CVD in regular snorers was similar to that observed in the cross-sectional studies. In a report from the Nurses' Health Study, Hu and coworkers (136) found significant associations between self-reported snoring and CVD among nearly 72,000 women monitored for up to 8 years. Adjusting for several possible confounding factors, regular snorers had a 33% (95% CI, 6 to 67%) elevation in risk of incident CVD relative to nonsnorers. Koskenvuo and coworkers (137) surveyed 3,847 male participants, 40 to 69 years of age, on snoring status and then ascertained CVD status with hospital discharge data and mortality records 3 years later. The odds ratio (95% CI) for new ischemic heart disease was 1.4 (1.2 to 1.7), for regular versus infrequent snorers, independent of BMI, age, smoking, alcohol, and hypertension. Jennum and coworkers (138) conducted a similar large prospective study (n = 2937) with participants aged 54 to 74 years surveyed on snoring and then monitored for CVD outcomes through hospital and mortality records for up to 6 years. In this study, snoring was not related to CVD (adjusted relative risk, 1.0; 95% CI, 0.6 to 1.6). The disparity in results from these two Scandinavian studies with similar study methods and large, well-constructed samples is puzzling. Although this could simply be due to sampling variability (the confidence intervals overlap substantially), it is also consistent with the interpretation that the association of OSA and CVD is present only in younger men, below the age range of the Jennum and coworkers study (138), but not below the age range of the Koskenvuo and coworkers study (137) (i.e., less than 54 years old). Stroke has been linked to OSA in cross-sectional and case–control studies. In the Sleep Heart Health Study, Shahar and coworkers (135) found a stronger association between stroke and OSA than between total CVD and OSA; the odds ratio of prevalent stroke in persons in the upper OSA quartile compared with those in the lowest quartile, adjusted for several possible confounding factors, was 1.58 (95% CI, 1.02 to 2.46). In case–control studies of stroke patients and hospital control subjects (139, 140) or control subjects selected from cases' friends or family (141), snorers were found to have substantially elevated odds of stroke (significant odds ratio ranged from 2 to 10). More definitive assessment of the causal role of OSA in stroke awaits data from prospective designs in which it can be determined that OSA precedes stroke, and that associations are not influenced by recall bias, choice of control groups, or stroke-related breathing disturbance. Several studies of mortality in sleep clinic patients suggest that OSA results in increased CVD mortality. He and coworkers (142) attempted to ascertain the vital status of 706 patients with sleep apnea. Mortality in treated versus untreated patients was compared in the 385 patients (only 55% of the original sample) who were successfully tracked. Conservatively treated patients (e.g., weight loss was advised) had a significantly higher death rate than did patients treated by tracheostomy. Using a similar study design, Partinen and colleagues conducted a 5-year mortality follow-up (143) and 7-year morbidity (144) analysis of 198 patients with sleep apnea. Extensive tracking determined that conservatively treated patients, compared with those treated by tracheostomy, had more than two times the risk of new vascular disease and nearly five times the risk of cardiovascular or stroke-related death. Also, in separate 5-year prospective Swedish studies of patients with coronary artery disease, OSA was found to increase the risk of mortality in one study (145) and increase the composite risk of occurrence of death, MI, or cerebrovascular event (stroke or transient ischemic attack) in the other study (146).
The association between snoring or AHI and all-cause or cardiovascular mortality has also been examined in several population-based studies. In a study from Sweden, data on snoring, excessive daytime sleepiness, and other factors were obtained by mailed questionnaire from 3,100 men (ages 30 to 69 years). Lindberg and colleagues (147) tracked mortality outcomes of the complete sample over a 10-year period. An overall association between snoring and mortality was not observed. However, in the subset of men less than 60 years of age, those with snoring and excessive daytime sleepiness were approximately twice as likely to die over the study period as men without those symptoms (relative risk, 2.2; 95% CI, 1.3 to 3.8), after adjustment for several possible confounding factors. In a community sample, Ancoli-Israel and coworkers (148) conducted an 8- to 10-year follow-up of 426 older persons by in-home nocturnal polygraphy-assessed breathing disturbance. In unadjusted analyses, severe respiratory disturbance (AHI These few data preclude firm conclusions about the magnitude of the associations between OSA and mortality. The clinic-based studies indicate that people with untreated OSA are at greater risk for early mortality. However, without randomization to treatment groups, it is possible that observed differences at follow-up merely reflect differences in baseline health. It might be expected that the patients with the most severe OSA would be treated aggressively, and indeed comparisons of AHI, weight, and other factors in the studies by He and coworkers (142) and Partinen and coworkers (143) support this. As this should bias the studies toward a null result, the observed associations do provide evidence consistent with a role for OSA in excess mortality. Because these studies were conducted with patients experiencing severe OSA, however, the findings may not be applicable to mild or moderate OSA. In contrast to the clinic-based studies, population studies have not demonstrated strong associations between OSA and mortality. This may, in part, reflect an association of OSA with mortality only in younger to middle-aged adults, as the data from Lindberg and coworkers suggest. It is possible that OSA in older persons represents a less noxious disease than OSA in younger populations, as previously discussed, or that older persons with OSA are constitutionally resistant to its adverse consequences, having survived a selection process that claimed their less resistant contemporaries. In summary, although it appears that OSA is likely to increase moderately the risk of cardiovascular morbidity and mortality, strong empirical evidence of that conclusion and precise estimates of the magnitude of the association will have to await incidence data from several ongoing population-based cohort studies of objectively assessed OSA. It remains to be demonstrated whether increased CVD risk is truly independent of the effects of OSA on blood pressure and whether treatment of OSA (e.g., with CPAP) can reduce cardiovascular risk. This important issue will be more difficult to address than the relationship of OSA treatment and elevated blood pressure because of the large samples and lengthy follow-up periods required to assess those outcomes.
Sleepiness Notwithstanding the strong association of AHI with self-reported sleepiness, the majority of subjects with an AHI of 5 or greater in each of these studies did not report excessive sleepiness. Indeed, the mean ESS score in Sleep Heart Health Study subjects with "severe" OSA, defined as an AHI 30 or greater, was lower than the mean ESS score previously reported for clinically identified cases of "mild" OSA, defined as an AHI from 5 to 15. Although self-report measures may underestimate the severity of sleepiness in the setting of chronic hypersomnolence, it is likely that many, if not most, individuals with polysomnographic evidence of OSA have minimal daytime sleepiness. This implies that there is considerable inter-individual variation in susceptibility to sleepiness resulting from OSA and indicates the potential magnitude of the bias inherent in attempting to extrapolate to the general population from studies of clinical cases. As reviewed elsewhere, sleep fragmentation due to repeated arousals from apneas and hypopneas is thought to be the cause of excessive sleepiness in patients with OSA (153). In the Sleep Heart Health Study, however, differences in the frequency of arousals, defined by American Sleep Disorders Association Atlas Task Force criteria (154) did not explain the observed variation in resultant sleepiness (152). More detailed study of subjects with OSA with and without excessive sleepiness, drawn from the same population, is needed to explain the factors underlying individual differences in susceptibility to daytime sleepiness. A number of epidemiological studies have evaluated the relationship between snoring and daytime sleepiness and almost all have found a significant association. As snoring is a strong marker of the presence of OSA, the association of snoring with sleepiness might be due to their joint association with OSA; however, several studies suggest that snoring is independently associated with excessive sleepiness. Stradling and coworkers (155) found that the report of snoring "often" was associated with 5-fold increased odds of subjects reporting that they fall asleep during the day against their will after adjusting for the severity of OSA as measured by the frequency of 4% dips in blood oxygen saturation during the night. On each of the three questions quantifying sleepiness in the Wisconsin Sleep Cohort Study, subjects with an AHI less than 5 who reported habitual snoring (three or more nights per week) had a prevalence of daytime sleepiness approximately midway between those of subjects with an AHI less than 5 who did not report habitual snoring and subjects with an AHI 5 or more (11). Among 5,777 subjects participating in the Sleep Heart Health Study, there was a progressive increase in sleepiness as measured by the Epworth Sleepiness Scale across five categories of snoring frequency, from a mean of 6.4 in current nonsnorers to 9.3 in subjects who snored 6 to 7 nights per week (156). The prevalence of excessive daytime sleepiness, defined as an ESS score 11 or more, increased from 15% in never-snorers to 39% in those who snore 6 to 7 nights per week. The relationship of snoring to sleepiness was seen at all levels of AHI, with no significant change in the relationship of snoring to ESS score after adjustment for AHI in multivariate models. These studies suggest that snoring without frank apnea and hypopnea episodes is associated with daytime sleepiness independent of AHI. If so, the very high prevalence of snoring in the adult population suggests that public health burden of snoring-related sleepiness might well exceed that of overt OSA. One must interpret the available data with caution, however, as the mechanism underlying the association of snoring with sleepiness is unclear. While snoring-related arousal due to increased upper airway resistance or snoring noise is possible, an increased arousal frequency, measured using American Sleep Disorders Association criteria, did not explain the observed association in the Sleep Heart Health Study (156). Similarly, although Stradling and coworkers (157) found an association between snoring and sleepiness, as measured by the Epworth Sleepiness Scale, this was not explained by an increase in either arousals or increased inspiratory effort as measured using pulse–transit time. As there is well-documented night-to-night variability in the measurement of AHI, it is possible that among subjects without elevated AHI on a single night of monitoring, habitual snoring is an indicator of a higher "usual" AHI. Confounding by the effects of voluntary sleep restriction, which is a cause of both snoring (158) and sleepiness, could contribute to the observed association, although in one study the relationship of snoring to sleepiness was independent of self-reported sleep restriction (156). A true association of snoring with sleepiness is suggested, however, by the observation that habitually snoring subjects with an AHI less than 5 in the Wisconsin Sleep Cohort Study had 3-fold increased odds of experiencing multiple motor vehicle accidents during a 5-year period compared with subjects without habitual snoring (159).
Cognitive Function
Health-related Quality of Life
Motor Vehicle Crashes and Occupational Accidents
Although the associations of OSA and motor vehicle crashes demonstrated in sleep clinic samples are alarming, there is great potential for overestimation of risk in clinic patient samples due to selection bias. However, two population studies of undiagnosed OSA and objectively measured motor vehicle crashes also suggest that the association is strong. Young and coworkers (159) investigated 5 years of state records of reported motor vehicle accidents and OSA in the Wisconsin Sleep Cohort. Men, but not women, with an AHI Most recently, similar findings of an association of OSA with motor vehicle crash history, independent of sleepiness, were reported from a novel study conducted in Spain (177). From a survey of a random population sample (n = 4,000), subjects reporting that they often felt so sleepy while driving that they feared falling asleep were identified. Laboratory polysomnography, including esophageal pressure recording to identify episodes of upper airway resistance, was conducted on a subsample. Of these sleepy drivers, those with self-reported motor vehicle crashes in the previous 5 years, compared with those without crashes, were twice as likely to have OSA defined as an AHI of 5 or more, but the association was not significant. When arousals due to upper airway resistance were taken into account in the definition of OSA, associations were stronger and significant: the odds ratio (95% CI) for 15 or more respiratory events per hour, compared with fewer than 15 events per hour, was 8.5 (1.3 to 62). Although the population studies do appear to support for a role for undiagnosed OSA in vehicle crashes, it is important to stress the wide confidence intervals for the odds ratios reported. The lack of finding sleepiness as an explanatory factor in the OSA–motor vehicle crash association is disconcerting because it may indicate that drivers with OSA do not perceive performance impairment and thus may not be likely to take extra precautions when driving. A more precise estimate of the magnitude of crash risk associated with OSA based on prospective data is critically needed to determine the risk of motor vehicle crashes attributable to OSA at different severity levels and identify vulnerable subgroups. No large population-based study of OSA measured by polysomnography and occupational accidents has been conducted; however, Lindberg and colleagues found support for this association based on self-reported snoring and sleepiness as an indicator of OSA. In a cohort of men in Uppsala (n = 2,724), the authors found baseline snoring and sleepiness was significantly related to occupational injuries as recorded in 10 years of government records (odds ratio, 2.2; 95% CI, 1.3 to 3.8). In another study in Sweden, using the same source of occupational injury data, clinic patients with OSA syndrome or heavy snoring were 2- to 3-fold as likely to have had an occupational injury in the past 10 years than were employed control subjects from the general population. The clinic sample, however, is likely to reflect selection bias of men with OSA who are most impaired, and so the association may be overestimated.
Impact on Pregnancy There are few epidemiologic data concerning the association of OSA with pregnancy outcomes. Loube and coworkers (48) found no significant differences in birth weight, Apgar score, or frequency of perinatal complications between infants born to women with or without frequent snoring during the second or third trimester of pregnancy. In contrast, Franklin and coworkers (50) found that women who reported snoring often or always during the week before delivery were more than twice as likely to have pregnancy-induced hypertension (14 versus 6%) and pre-eclampsia (10 versus 4%) as were women without frequent snoring, and were more than twice as likely to give birth to an infant small for gestational age (7.1 versus 2.6%) or with an Apgar score less than 7 at both 1 and 5 minutes. After adjusting for maternal age, weight, and smoking habits, odds ratios for associations with frequent snoring were 2.0 for pregnancy-induced hypertension, 2.2 for pre-eclampsia, and 3.5 for intrauterine growth retardation.
Several factors have been hypothesized to have a role in the development and progression of OSA. Potential causes of OSA most relevant to the questions central to this review are those that are prevalent and have the potential to be ameliorated by relatively simple, noninvasive, and preferably population-based interventions: excess body weight, smoking, alcohol consumption, nasal congestion, and hormonal changes during menopause.
Excess Body Weight Excess body weight has been hypothesized to affect breathing in numerous ways, including alterations in upper airway structure (e.g., altered geometry) or function (e.g., increased collapsibility), disturbance of the relationship between respiratory drive and load compensation, and by exacerbating OSA events via obesity-related reductions in functional residual capacity and increased whole-body oxygen demand (193, 194). These putativ |
TOP
ABSTRACT
CONTENTS
5), from 3 to 28%; for OSA of at least moderate severity (defined by AHI 
