Type 2 diabetes is a chronic illness that is increasing in epidemic proportions worldwide. It affects 25.6 million adults in the U.S., with an estimated 1.9 million new diagnoses per year [Center for Disease Control (CDC), 2011]. Major factors contributing to the development of type 2 diabetes include obesity and poor lifestyle habits (e.g., excess dietary intake and limited physical activity). The current standard criteria for the diagnosis of type 2 diabetes are as follows: (1) Hemoglobin A1c (A1c) ≥6.5%, or (2) fasting plasma glucose ≥126 mg/dl, or (3) 2 h plasma glucose ≥200 mg/dl during an oral glucose tolerance test (OGTT), or (4) random plasma glucose ≥200 mg/dl in a patient with classic symptoms of hyperglycemia [American Diabetes Association (ADA), 2010].

High glucose levels not sufficient to meet the diagnostic criteria for type 2 diabetes are considered as prediabetic states, namely impaired fasting glucose (IFG) and impaired glucose tolerance [IGT; American Diabetes Association (ADA), 2010]. According to the CDC, 35% of U.S. adults aged 20 years or older may have prediabetes, which suggests that prediabetic states exist in an estimated 79 million Americans [Center for Disease Control (CDC), 2011]. Fasting plasma glucose levels that are in the range of 100–125 mg/dl define an IFG state. A 2-h plasma glucose level during the 75-g OGTT that is in the range of 140–199 mg/dl defines an IGT state. In prediabetic states, the A1c ranges between 5.7 and 6.4% [American Diabetes Association (ADA), 2010]. The hallmark of prediabetic states is insulin resistance. Transition from the early states of impaired glucose metabolism to frank diabetes may take several years, and it is estimated that ∼70% prediabetic individuals eventually develop type 2 diabetes (Nathan et al., 2007). The most commonly used clinical and research methods to assess glucose tolerance, insulin sensitivity, and secretion are summarized in Table 1.

In normal individuals, insulin resistance is compensated by an up-regulation of insulin secretion to maintain normal glucose tolerance. According to the “hyperbolic law of glucose tolerance,” under normal conditions the product of insulin sensitivity and secretion (i.e., the disposition index; DI) is constant (Bergman, 2005). In individuals who are at risk for developing type 2 diabetes, the DI may be reduced due to inadequate compensatory insulin secretion for a given degree of insulin resistance. Thus, type 2 diabetes is a two-step process characterized by both insulin resistance and impaired insulin secretion. Initially, insulin resistance occurs resulting in progression from a normal glucose tolerant state to IGT. At this stage, insulin levels are elevated (compensatory hyperinsulinemia) and beta-cell dysfunction is already present. Subsequently, progressive decline in beta-cell function occurs, which leads to development of type 2 diabetes (DeFronzo and Abdul-Ghani, 2011).

Despite the proven efficacy of lifestyle interventions and the use of multiple pharmacological agents, the economic and public health burden of type 2 diabetes remains substantial. Cardiovascular disease is a major cause of mortality and morbidity in both type 2 diabetes and prediabetes. In addition to cardiovascular disease, type 2 diabetes is associated with longterm complications affecting the eyes, kidneys, and nervous system. (Kashyap and Defronzo, 2007; Nathan et al., 2007; Mazzone et al., 2008). It is well documented that both lifestyle modification and metformin use are effective in reducing the incidence of diabetes (Tuomilehto et al., 2001; Knowler et al., 2002, 2009). However, even with substantial efforts to achieve lifestyle modification, data suggest that the effectiveness of this intervention diminishes over time. Thus, the development of additional strategies to prevent or delay the onset of diabetes and its cardiovascular complications remains a critical public heath challenge.

The overall prevalence of OSA diagnosed by full polysomnography in type 2 diabetic patients is approximately 71% based on the average of data from five studies including a total number of nearly 1200 type 2 diabetic patients (Resnick et al., 2003; Einhorn et al., 2007; Foster et al., 2009; Laaban et al., 2009; Aronsohn et al., 2010; Table 2). The highest estimate of 86% was reported in obese diabetic patients enrolled in the multi-center Sleep AHEAD study (Foster et al., 2009). The lowest estimate of 58% was found in the Sleep Heart Health Study (Resnick et al., 2003), which is a multi-center cohort that included older individuals (with more than half aged over 65 years) and used self-reported diabetes and a desaturation criteria of 4% for the definition of hypopneas. In another study by Aronsohn et al., OSA prevalence was found to be 77% using a less stringent cut-off of 3% for oxygen desaturations. Of note, when the dataset was re-analyzed using more strict 4% desaturation criteria, this estimate was reduced to 58% (Aronsohn et al., 2010). These variable prevalence estimates may be due to differences in study populations as well as different scoring criteria used to capture OSA (Redline et al., 2000). Nevertheless, considering that the prevalence of OSA by full-night polysomnography in patients with type 2 diabetes averages 71%, this would then suggest that approximately 19 million diabetics may have undiagnosed and untreated OSA. Indeed, a recent retrospective analysis of a total of 16,066 diabetic patients with one or more primary care office visits in 27 primary care ambulatory practices has revealed that only 18% of the study population (23% of obese patients) carried an OSA diagnosis (Heffner et al., 2012). These findings suggest that in the primary care setting, OSA is largely unrecognized in patients with diabetes and thus millions of diabetics may currently suffer from this co-morbidity. In sum, these staggering statistics emphasize the need for a better understanding of the links between OSA and type 2 diabetes (Shaw et al., 2008).

Several population and clinic-based studies have assessed the prevalence and incidence of type 2 diabetes in OSA (Meslier et al., 2003; Reichmuth et al., 2005; Seicean et al., 2008; Tamura et al., 2008; Botros et al., 2009; Mahmood et al., 2009; Marshall et al., 2009; Ronksley et al., 2009; Celen et al., 2010; Fredheim et al., 2011). The majority of these studies have used validated diabetes definitions based on fasting glucose or 2 h post-challenge glucose levels, while a few studies relied on self-reported diabetes and/or antidiabetic medication use. Most, but not all studies (Marshall et al., 2009; Ronksley et al., 2009; Celen et al., 2010), used full polysomnography to diagnose OSA. The majority of studies used statistical adjustments for shared risk factors such as age, sex, body mass index (BMI) and various other parameters (e.g., neck and waist circumference). Overall, the cross-sectional analyses have shown a significantly higher prevalence of diabetes in patients with OSA as compared to those without OSA. The prevalence estimates range from 15 to 30%, depending on the study population, the definition of OSA severity, and the methods used to diagnose type 2 diabetes (Meslier et al., 2003; Reichmuth et al., 2005; Mahmood et al., 2009; Ronksley et al., 2009; Fredheim et al., 2011). A few studies have also reported a significant relationship between increasing severity of OSA and the prevalence of diabetes (Reichmuth et al., 2005; Tamura et al., 2008; Ronksley et al., 2009; Samson et al., 2012). In two studies (Mahmood et al., 2009; Marshall et al., 2009), the higher odds ratio for diabetes in patients with OSA was no longer significant after controlling for confounders. In one study (Ronksley et al., 2009), a higher adjusted odds of self-reported diabetes was found only in the severe OSA group who had excessive sleepiness. In this regard, more research is needed to elucidate the role of sleepiness in adverse metabolic outcomes and to better understand the links between OSA and diabetes in sleepy versus non-sleepy patients.

A total of five prospective studies have examined the incidence of type 2 diabetes in patients with OSA (Table 3). In the longitudinal follow-up of about 1000 individuals from the Wisconsin Sleep Cohort (Reichmuth et al., 2005), OSA was found to be a risk factor for incident diabetes over a 4-year period, but this association was no longer significant after adjusting for age, sex, and body habitus. The Busselton Health Study (Marshall et al., 2009) reported a significant independent association between moderate to severe OSA and incident diabetes over a 4-year follow-up period, but the sample size was small with only a few incident cases of diabetes. In another study by Botros et al. (2009), an independent association between OSA and incident diabetes was found after adjusting for age, sex, race, BMI, baseline fasting glucose and weight change over a mean follow-up period of 2.7 years. Interestingly, one clinic-based study of 168 middle-aged patients (Celen et al., 2010), reported a significantly higher incidence of diabetes in patients with OSA compared to those without OSA in women, but not in men, after a follow-up period of 16 years. More recently, in a population-based cohort of 156 men from Sweden (Lindberg et al., 2012), the increasing severity of oxygen desaturations in OSA was found to be a predictor of developing diabetes after adjustments for various confounders including years of treatment for OSA. In another prospective study of about 4000 middle-aged individuals, nocturnal intermittent hypoxia was associated with an increased risk of developing type 2 diabetes after a 3-year median follow-up (Muraki et al., 2010).

Thus, the current evidence from cross-sectional analyses suggest that type 2 diabetes is more prevalent among patients with OSA compared to those without OSA, independent of shared risk factors. However, it is still unclear whether OSA may lead to the development of diabetes over time. More data from large-scale longitudinal studies with rigorous assessments of diabetes and OSA are needed.

An earlier study including 279 diabetic patients found no significant associations between OSA and glycemic control, as assessed by A1c (Einhorn et al., 2007). In this negative study, only about 22% of the study population was assessed by full polysomnography and the duration of sleep recording was reported to be as low as 4 h. In a more recent study (Aronsohn et al., 2010), 60 patients with physician-diagnosed diabetes underwent full laboratory polysomnography (with a minimum laboratory recording time of 7 h) to assess the presence and severity of OSA. Increasing severity of untreated OSA was associated with poorer glucose control after controlling for age, sex, race, BMI, number of diabetes medications, level of exercise, years of diabetes, and total sleep time. Compared to patients without OSA (adjusted mean A1c: 5.7%), the adjusted mean A1c was 7.3% in mild OSA, 7.7% in moderate OSA, and 9.7% in severe OSA. Importantly, the magnitude of these associations is comparable, if not exceeding to those of widely used hypoglycemic medications. Of note, when only the first 4 h of recording was analyzed, the robust relationship between OSA severity and A1c was no longer apparent, perhaps due to insufficient recording time for REM sleep to occur. This would suggest that sleep recordings longer than the commonly used minimum of 4 h might be needed for better assessment of the links between OSA and type 2 diabetes. Another cross-sectional study (Pillai et al., 2011) involving 52 consecutive patients recruited from a diabetes-obesity clinic found that after adjusting for age, gender, BMI, duration of diabetes, and insulin dose, increased severity of OSA was associated with increased A1c levels. The adjusted mean values of A1c in each OSA category were 8.62% for no OSA, 9.36% for mild, 10.61% for moderate, and 9.91% for severe OSA. In a smaller study (Fendri et al., 2011) of 26 overweight and obese individuals with type 2 diabetes, nocturnal glycemia (as assessed by a continuous glucose monitoring sensor) was found to be 38% higher in those who had OSA, compared to those without OSA, independent of BMI.

Overall, there is evidence to suggest that the presence and severity of untreated OSA may be associated with poor glucose control in type 2 diabetic patients. These findings may have important clinical implications in the management and care of type 2 diabetes as they support the hypothesis that reducing the severity of OSA may be an adjunct therapeutic strategy to optimize glucose control.

Over the past few decades, an increasing number of studies from population and clinic-based cohorts have consistently found a robust independent association between the presence and severity of OSA and prediabetic states such as insulin resistance and glucose intolerance. A comprehensive review of all these studies is beyond the scope of this article, but has been the topic of several review articles (Punjabi and Polotsky, 2005; Tasali and Ip, 2008; Jun and Polotsky, 2009; Levy et al., 2009; Punjabi, 2009; Pamidi et al., 2010). In the majority of these studies the estimation of insulin sensitivity was based on the homeostatic model assessment (HOMA) index, which is a normalized product of fasting glucose and insulin levels. Several studies have also used the OGTT to assess glucose and insulin response after glucose ingestion. Interestingly, many of these studies were conducted in men, so the data on possible gender effects in the associations between OSA, insulin resistance, and glucose intolerance is lacking. In the majority of studies, commonly used indices of OSA severity were the apnea-hypopnea index (AHI) and the frequency as well as the degree of intermittent hypoxia.

Overall, the prevalence of prediabetes (i.e., the presence of either IFG and/or IGT) was found to be significantly higher in patients with OSA than those without OSA, but the estimates have been variable, ranging between 20 and 67% (Meslier et al., 2003; Kono et al., 2007; Seicean et al., 2008; Tamura et al., 2008; Fredheim et al., 2011). In a large cross-sectional analysis of a subset of data from the Sleep Heart Health Study, involving over 2500 non-diabetic individuals (Seicean et al., 2008), the presence of OSA was associated with a significantly higher odds of IFG and IGT after controlling for age, sex, race, BMI, and waist circumference. Importantly, the magnitude of these associations was similar in non-overweight and overweight individuals (Seicean et al., 2008).

In the earliest studies from large population-based samples, OSA was found to be independently associated with insulin resistance (Ip et al., 2002; Punjabi et al., 2002) and glucose intolerance (Punjabi et al., 2002). In a clinic-based sample of 595 men, the increasing severity of OSA was associated with worsening glucose tolerance and insulin resistance, independent of age and BMI (Meslier et al., 2003). In an earlier report from the Sleep Heart Health Study, the increasing severity of OSA was independently associated with both fasting and 2 h post-load glucose levels during an OGTT (Punjabi et al., 2004). In a population-based sample of 400 women, those with severe OSA had significantly lower insulin sensitivity as compared to those without OSA, and the severity of OSA was associated with increasing fasting and 2-h post-challenge insulin levels during an OGTT, after adjusting for age, waist-hip ratio, and other potential confounders (Theorell-Haglow et al., 2006). In an analysis of a subset of participants in the Wisconsin Sleep Cohort Study (n = 546), OSA was significantly associated with insulin resistance as measured by the HOMA index (Nieto et al., 2009). More recently, Punjabi and Beamer (2009) estimated insulin sensitivity in 118 non-diabetic adults using a frequently sampled intravenous glucose tolerance test (ivGTT). The authors found that compared to normal individuals, those with mild, moderate, and severe OSA showed a 26.7, 36.5, and 43.7% decrease in insulin sensitivity, respectively, after controlling for age, gender, race, and percent body fat. Despite this progressive decline in insulin sensitivity, the acute insulin response to glucose was unchanged, suggesting an insufficient up-regulation of insulin secretion and thus impaired beta-cell function. A recent cross-sectional study of 1,599 non-diabetic patients, where participants with an A1c of ≥6.5%, self-reported diabetes or use of diabetic medication were excluded, a dose–response relationship between OSA severity and the percentage of patients with A1c >6% was observed, suggesting a gradual increase in prediabetic state with increasing severity of OSA (Priou et al., 2012). Although some of these studies have controlled for adiposity using several surrogate markers (e.g., waist circumference, percent body fat), central adiposity, particularly visceral fat is an important confounding factor in the observed links between OSA and prediabetic states. In this regard, there is some evidence to suggest that in the absence of confounding effects of visceral adiposity, OSA may still be associated with insulin resistance and higher fasting glucose levels (Kono et al., 2007).

A recent study by Pamidi et al. (2012), the researchers assessed whether the presence of OSA affects glucose metabolism in young, lean individuals who are healthy and free of cardiometabolic disease. In a prospective design, 52 healthy men (age range 18–30 years and BMI range 18–25 kg/m²) underwent a laboratory polysomnogram followed by a morning OGTT to assess glucose metabolism. All subjects were stratified according to the presence or absence of ethnicity-based diabetes risk and family history of diabetes. The authors then used a frequency matching approach and randomly selected individuals without OSA, yielding a total of 20 control men without OSA and 12 men with OSA. Men with OSA and controls were similar in terms of age, BMI, ethnicity-based diabetes risk, family history of diabetes, and level of exercise. Both groups had normal systolic and diastolic blood pressure and fasting lipid levels. Following ingestion of a 75-g glucose load, men with OSA had 27% lower insulin sensitivity (estimated by the Matsuda index during the OGTT) and 37% higher total insulin secretion than the controls, despite comparable glucose levels. This study is the first to demonstrate that in young lean men (mean age <24 years, mean BMI <23 kg/m²) who are otherwise healthy and free of cardiometabolic disease, the presence of OSA is associated with insulin resistance and compensatory hyperinsulinemia to maintain normal glucose homeostasis. These findings provide evidence to support the hypothesis that OSA may be associated with early changes in the natural history of type 2 diabetes even in the absence of obesity, cardiovascular disease, and other known risk factors for type 2 diabetes.

In summary, these observational studies strongly support an association between OSA and insulin resistance and glucose intolerance, even after statistical adjustments for shared risk factors. However, central adiposity, particularly visceral fat, needs to be considered as an important confounder when interpreting these studies. Finally, studies with longitudinal and interventional designs are needed to better address the potential role of OSA in the development of prediabetic states.

The two major characteristics of OSA, namely intermittent hypoxia and sleep fragmentation can lead to derangements in glucose metabolism by several mechanisms including activations of the sympathetic nervous system and hypothalamic-pituitary axis, and changes in the inflammatory pathways (Jun and Polotsky, 2009). Other putative mechanisms include hypoxic injury to pancreas and potential alterations in central (e.g., hypothalamic) pathways for glucose control. Animal models mimicking OSA have been used to elucidate the potential mechanisms that may be responsible for metabolic dysregulation in OSA. Intermittent hypoxia stimuli of varying frequency and severity were used in rodents when they are most likely to be asleep. In one study, electroencephalogram tracings were recorded and showed that intermittent hypoxia stimuli were accompanied by arousals, suggesting that sleep fragmentation is also an intrinsic feature of these hypoxia models (Polotsky et al., 2006). In these models, intermittent hypoxia led to sympathetic activation and hypertension (Fletcher et al., 1992; Fletcher, 2003), insulin resistance in both lean (Iiyori et al., 2007) and obese (Polotsky et al., 2003) rodents, impaired glucose homeostasis (Yokoe et al., 2008), and pancreatic beta-cell proliferation (Yokoe et al., 2008; Xu et al., 2009) as well as apoptosis of beta-cells (Yokoe et al., 2008; Xu et al., 2009). In a study by Gharib et al. (2012) mice that were exposed to 6 h of sleep fragmentation by the movement of an automated bar showed insulin resistance and glucose intolerance as well as altered transcriptional profile of visceral adipocytes. In healthy humans, one study that used intermittent hypoxia selectively during sleep found that 2–4 weeks of exposure to intermittent hypoxia resulted in increased morning blood pressure (Tamisier et al., 2009), but glucose metabolism was not assessed. Another study exposed healthy subjects to 5 h of intermittent hypoxia or normoxia during wakefulness and reported a decreased insulin sensitivity estimated by ivGTT (Louis and Punjabi, 2009). Two studies exposed healthy subjects to sleep fragmentation using acoustic stimuli and examined the effects on glucose metabolism by ivGTT. In the first study by Tasali et al. (2008a), all night selective suppression of slow wave sleep for three consecutive nights resulted in an approximate 25% decrease in insulin sensitivity without adequate compensatory rise in insulin secretion, leading to reduced glucose tolerance and increased diabetes risk (i.e., decreased disposition index). In the second study, non-selective sleep fragmentation for two nights was associated with a decrease in insulin sensitivity and non-insulin dependent glucose disposal (Stamatakis and Punjabi, 2009). In contrast to the study by Tasali et al. the second study noted an increase in the acute insulin response to glucose, suggesting that the beta-cell function was preserved. Notably, the second study was also associated with marked reductions in slow wave sleep, whereas other sleep stages were only minimally affected. Both of these sleep fragmentation studies were associated with elevated cardiac sympathetic output (Tasali et al., 2008a; Stamatakis and Punjabi, 2009). Overall, data from animal and human models that mimic OSA support an adverse effect of OSA on insulin resistance, glucose intolerance, and increased diabetes risk. The relative contributions of intermittent hypoxia versus sleep fragmentation on cardiometabolic outcomes remain to be determined.

In summary, current evidence strongly supports an independent association between OSA and insulin resistance and glucose intolerance, but a causal link remains to be determined. A substantial proportion of patients with type 2 diabetes suffer from unrecognized OSA. Conversely, type 2 diabetes is more prevalent among patients with OSA compared to those without OSA. Thus, the role of OSA in the management of type 2 diabetes is in urgent need of further rigorous assessment. The question of whether OSA represents an independent risk for the development of prediabetes and type 2 diabetes over time remains to be investigated by large prospective studies. Notably, weight loss, if successfully achieved can improve both OSA and glucose control, and thus should be recommended systematically to all patients with OSA. There is still debate whether CPAP treatment of OSA improves glucose metabolism. Also, the sufficient amount and duration of CPAP that is required to improve metabolic outcomes remains to be determined. Large-scale randomized-controlled trials of CPAP treatment of OSA with well-validated assessments of insulin sensitivity and glucose tolerance are needed. Identification of phenotypes of OSA patients who show better metabolic response to CPAP therapy would be important.

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.