Is Sleep Apnea an Independent Risk Factor for Prevalent and Incident Diabetes in the Busselton Health Study?

Abstract

Background:

Cross-sectional analyses of North American population-based cohorts and one nonsignificant longitudinal analysis have suggested that obstructive sleep apnea (OSA) is a risk factor for diabetes mellitus. However, this observation has yet to be replicated outside the USA or be observed lonigitudinally.

Methods:

Residents of the Western Australian town of Busselton had their OSA quantified by the respiratory disturbance index (RDI) overnight in their own homes (MESAM IV device). Diabetes was defined as either a fasting blood glucose ≥ 7 mmol/L or physician diagnosed diabetes.

Results:

Of 399 participants at baseline, 295 had complete data and did not have diabetes at baseline; 9 incident cases were observed within 4 years. At baseline moderate-severe OSA was associated with a univariate, but not multivariate, increased risk of diabetes (odds ratio = 4.37, 95% CL = 1.12, 17.12). Longitudinally, moderate-severe OSA was a significant univariate and independent risk factor for incident diabetes (fully adjusted OR = 13.45, 95% CL = 1.59, 114.11).

Conclusions:

Moderate-severe sleep apnea was a significant risk factor for incident diabetes in this Australian population-based cohort. However, the confidence intervals were wide and meta-analyses or studies with greater power will be required to verify the relationship between sleep apnea and the incidence of diabetes in community-based populations.

Citation:

Marshall NS; Wong KKH; Phillips CL; Liu PY; Knuiman MW; Grunstein RR. Is sleep apnea an independent risk factor for prevalent and incident diabetes in the busselton health study? J Clin Sleep Med 2009;5(1):15–20.

Obstructive sleep apnea (OSA) is characterized by repetitive upper airway closure during sleep resulting in repeated reversible blood oxygen desaturation and fragmented sleep. There is clear evidence that OSA promotes the development of hypertension and thus probably increases the risk for stroke, myocardial infarction and death from cardiovascular disease.¹ However, hypertension appears to be only part of the mechanism by which OSA is likely to be causing cardiovascular disease/mortality.^2,3 It has been proposed that one of the additional ways that OSA could be causing morbidity/mortality is through type 2 diabetes. There are several biologically plausible pathways to explain this, including the effects of slow wave sleep deprivation and intermittent hypoxia.^4,5

Unsurprisingly for 2 diseases with obesity as their major modifiable risk factor, patients seen clinically for either diabetes or OSA frequently also exhibit the other condition.^4,6 In addition, both clinical and population-based studies frequently, but not always, find OSA (or surrogate markers) are independently associated with diabetes (or markers).^4,6,7 More generally, OSA has also been proposed as a cause of metabolic syndrome with some supporting observations.^8–12 However, almost all population-based data supporting these hypotheses are from cross-sectional analyses and thus fail to provide evidence of the proposed direction of causality.¹³ The exception to this is the Wisconsin Sleep Cohort, which has measured sleep apnea by in-laboratory polysomnography and has found a non-significant but suggestive association between moderate-severe sleep apnea and incident physician-diagnosed diabetes after 4 years (OR = 1.62, 95% CI 0.67–3.65).¹⁴ This observation, however, requires independent verification from another source. Observations from countries other than the United States may also aid generalizability to other societies and health systems.

The community of Busselton in Western Australia has been the subject of cross-sectional and follow-up health surveys since 1966.^15–17 We aimed to investigate whether sleep apnea is an independent risk factor for the baseline prevalence and 4-year incidence of diabetes in a cohort within the Busselton Health Study who had objective assessment of sleep disordered breathing.¹⁸

METHODS

Target Population

The Busselton Sleep Study was conducted in November-December 1990 and involved a community-based sample of men and women aged 40 to 65 years who had participated in previous cross-sectional health surveys of the Busselton population.¹⁹ Many of these sleep study participants also participated in the 1994/95 Busselton Follow-up Survey of all previous cross-sectional survey participants. This study is based on the 400 people in the sleep study^3,18 who attended the 1994/95 follow-up survey and provided the required data at both surveys.

Sample Construction

The 2-stage sampling process has already been described elsewhere.¹⁸ Briefly, 758 men and 810 women 40 to 65 years of age were sent the initial sleep questionnaire; 486 men and 537 women responded. The 486 men were randomly telephoned until a sample of 311 had been recruited for an overnight sleep study (all available study appointments were filled). The women were stratified according to snoring (never, sometimes, and almost always/always) and were randomly telephoned until a sample of 38 had been recruited in each snoring group (total of 114). Five men and 6 women who were invited to participate in the overnight study declined the test. Of sleep recordings performed, 294 studies in men and 106 in women were of adequate quality for scoring; these participants also had matched follow-up data in 1994/95. Fewer women than men were sampled for financial and logistic reasons.

Sleep Disordered Breathing

Sleep disordered breathing was assessed in 1990 by a single night of recording in the homes of participants, using the MESAM IV device. The MESAM IV (Madaus Medizen-Elektronik, Freiberg, Germany), is a 4-channel portable home-monitoring device used to quantify sleep disordered breathing via the measurement of snoring, heart rate, oxygen saturation, and body position. The full methods for implementation and manual scoring of this technique are described in the original prevalence paper.¹⁸

Sleep disordered breathing was quantified by the respiratory disturbance index (RDI), which is calculated by summing the total number of respiratory disturbances and dividing by the estimated hours of sleep to give an event rate per hour. Respiratory disturbances were defined as oxygen desaturations ≥ 3% from the preceding baseline level accompanied by either (a) an increased heart rate ≥ 10 beats/min (i.e., a physiological response to apnea) and/or (b) a burst of snoring associated with commencement and termination of the desaturation event (i.e., an audible apnea). Hours of sleep were estimated using lights off and lights on time—a method that overestimates sleep time and therefore causes a lower estimation of RDI than traditional polysomnography (PSG). The MESAM IV has also been validated against PSG by other investigators.^20,21

Outcomes: Diabetes and Metabolic Syndrome

Fasting blood samples were taken the morning after the overnight sleep study in 1990 for fasting blood glucose (FBG), total and HDL cholesterol and analyzed using standard assay methods at the Western Australian State Health laboratories. During the 1994/95 follow-up survey, fasting blood samples were taken in the morning and analyzed using a hexokinase assay at PathWest in Perth, Western Australia.²²

Diabetes was classified when there was a report of a physician diagnosis or treatment of diabetes by the participant or when fasting blood glucose was ≥ 7.0 mmol/L (126 mg/dL).¹⁴

Other Exposure and Confounding Variables

Anthropometric variables were measured by study investigators on the evening before the sleep study and included height, weight, abdominal height while supine (sagittal diameter), as well as neck, waist, and hip circumference. Blood pressure was then measured twice with an electronic sphygmomanometer (Spacelabs, Redmond, WA) with measurements ≥ 5 min apart and after the participant had been lying quietly for ≥ 10 min. The mean of these measurements was then calculated.

Medically diagnosed angina was measured using a self-report questionnaire. If a participant did not answer these medical history questions, the participant was excluded from relevant analyses. Smoking status was ascertained on questionnaire by asking whether the patient was a current, former, or never smoker. Former and current smokers were queried about smoking duration and use in order to calculate pack-years. Alcohol consumption patterns were queried to calculate approximate grams of alcohol consumption per week.

Data Handling and Statistical Analyses

Analyses were undertaken using SAS (v 9.1 SAS Institute, Cary, NC, USA). We categorized sleep apnea using standard clinical cut-points for severity.²³ No or sub-clinical sleep apnea served as the reference category (i.e., 0 to 4 respiratory disturbances per hour), and the 2 sleep apnea categories were mild (5 to 14 RDI) and moderate-to-severe OSA ( ≥ 15 RDI). We were unable to examine the association of severe OSA with outcomes because only 3 participants in the cohort had an RDI ≥ 30 at baseline.

Those with missing glucose or diabetes data at baseline were excluded from prevalent analyses. Participants with diabetes at baseline or with missing fasting glucose or diabetes data at follow-up or at baseline were excluded from the incidence analyses.

Mean arterial pressure was defined as (2/3*Systolic + 1/3*Diastolic). Body habitus was investigated using BMI as a continuous variable or classified into normal/overweight/obese (< 25, 25 to 30, or ≥ 30 kg/m²). Body habitus was also investigated using waist circumference and waist/hip ratio. Smoking was classified categorically (Never or Ex-smoker versus Current) and as a continuous variable using pack-years.

Risk factors for diabetes and sleep apnea were investigated using χ², Fisher exact, ANOVA, or Kruskal-Wallis tests, where appropriate.

To examine whether sleep apnea was a risk factor for incident diabetes, we built univariate and multivariate models using logistic regression (PROC LOGISTIC). Regardless of univariate association with diabetes, the following risk factors were forced into the fully adjusted model because of known associations with either OSA, diabetes, or both: age, gender, and obesity (continuous BMI and waist circumference). Other risk factors were examined when they exhibited some evidence of a univariate association with either diabetes or with sleep apnea (p < 0.1).

RESULTS

The sleep apnea cohort consisted of 399 participants (294 males). Sixty-seven had missing fasting glucose data from either baseline (n = 18) or follow-up and were excluded from the analyses of incident or prevalent diabetes, as relevant. Table 1 lists the association of baseline risk factors with sleep apnea severity, and Table 2 lists the univariate associations between these risk factors and incident diabetes. Table 3 shows the raw numbers of prevalent cases of diabetes across sleep apnea severity categories and the univariate and multivariate modelling. Table 4 shows the raw numbers of cases of incident diabetes in each of the 3 sleep apnea groups and that sleep apnea is a significant independent risk factor for incident diabetes before and after adjustment for other identified risk factors and potential confounders. The partially adjusted model presented is based on that used in the Wisconsin Sleep cohort¹⁴ and controlled for age, gender, and waist circumference.

Table 1.

Association of Sleep Apnea With Risk Factors in People Without Diabetes at Baseline

Baseline Confounder Mean (SD)	No OSA (RDI < 5) n = 230	Mild OSA RDI (5 to < 15) n = 63	Moderate-Severe OSA (RDI ≥ 15) n = 10	p Value for Difference
Females % (n, cases)	29.1 (67)	25.4 (16)	30.0 (3)	0.83
Age (years)	52.5 (7.6)	55.1 (7.0)	54.8 (10.0)	0.0497
Body mass index (kg/m2)	26.0 (3.5)	27.7 (3.9)	33.8 (8.8)	< 0.001
Current smokers % (n, cases)	12.2 (28)	20.6 (13)	0.0 (0)	0.21
Pack-years	12.9 (22.8)	16.7 (21.5)	9.8 (12.5)	0.15†
Waist hip ratio	0.89 (0.08)	0.91 (0.09)	0.94 (0.05)	0.047
Waist circumference (cm)	91.0 (10.7)	95.6 (12.0)	107.2 (7.1)	<0.001
Alcohol consumption (g/week)	86.3 (97.2)	103.7 (113.2)	124.8 (120.9)	0.33†
Mean arterial pressure (mm Hg)	112.2 (12.1)	116.4 (13.0)	114.7 (13.5)	0.049
High density lipoprotein cholesterol (mmol/L)	1.28 (0.33)	1.20 (0.35)	1.07 (0.22)	0.045
Total cholesterol (mmol/L)	5.93 (1.02)	6.03 (1.14)	5.96 (0.88)	0.81
Doctor diagnosed angina % (n, cases)	8.3 (19)	4.8 (3)	10 (1)	0.62

Figures represent means and SD, unless otherwise stated for categorical variables. No participants with missing baseline or follow-up fasting glucose values or diabetes status are listed in this table. Categorical variables tested with chi square and continuous variables with ANOVA or with a Kruskal-Wallis test as appropriate (marked with ^†). The association between sleep apnea and other risk factors in the full cohort has already been published.³ The standard SI unit mmol/L can be converted into approximate North American units mg/dL by multiplying by 18.18 for glucose and 38.67 for the cholesterol variables.

Table 2.

Univariate Associations Between Baseline Risk Factors and Incident Diabetes

Baseline Risk Factor	Unit of Measure	Odds Ratio (95% CL)	p Value
Gender	Female	REF
	Male	3.19 (0.39, 25.9)	0.28
Angina	No	REF
	Yes	1.74 (0.21, 14.62)	0.61
Age	Per decade	1.29 (0.53, 3.13)	0.57
Mean arterial pressure	Per 10 mm Hg	1.78 (1.10, 2.89)	0.020
High density lipoprotein cholesterol	mmol/L	0.094 (0.007, 1.25)	0.07
Total cholesterol	mmol/L	1.26 (0.70, 2.23)	0.45
Smoking	Per pack-year	1.012 (0.99, 1.03)	0.28
Smoker categories	Never	REF
	Past	0.36 (0.71, 1.81)	0.21
	Current	0.58 (0.07, 4.9)	0.61
Body mass index	Per BMI unit	1.05 (0.91, 1.22)	0.50
Body mass categories	Normal < 25	REF
	Overweight 25 to < 30	0.75 (0.18, 3.06)	0.68
	Obese ≥ 30	0.60 (0.07, 5.51)	0.65
Waist circumference	Per cm	1.049 (0.987, 1.114)	0.12

Figures are derived from univariate logistic models. REF = Reference category, BMI = body mass index, CL = confidence limits

Table 3.

Univariate and Multivariate Association of Sleep Apnea with Diabetes at Baseline

	N	Prevalent Cases of Diabetes n (%)	Unadjusted Odds Ratio	Partially Adjusted Odds Ratio†	Fully Adjusted Odds Ratio‡
No sleep apnea	278	13 (4.7%)	REF	REF	REF
Mild sleep apnea	77	4 (5.2%)	1.12 (0.35, 3.53)	0.83 (0.26, 2.71)	0.84 (0.25, 2.81)
Moderate-to-severe sleep apnea	17	3 (17.7%)	4.37 (1.12, 17.12)	2.13 (0.48, 9.50)	1.98 (0.41, 9.55)

^†Adjusted for age, gender, waist circumference based on the model presented from the Wisconsin sleep cohort¹⁴; n = 367.

^‡Additionally adjusted for BMI, mean arterial pressure, and HDL cholesterol n = 367.

Table 4.

Univariate and Multivariate Association of Sleep Apnea with Incident Diabetes

	N	Incident Cases of Diabetes n (%)	Unadjusted Odds Ratio	Partially Adjusted Odds Ratio†	Fully Adjusted Odds Ratio ‡
No sleep apnea	226	5 (2.2%)	REF	REF	REF
Mild sleep apnea	59	2 (3.4%)	1.47 (0.28, 7.76)	1.31 (0.23, 7.35)	1.51 (0.25, 9.12)
Moderate-to-severe sleep apnea	10	2 (20%)	11.20 (1.88, 66.75)	8.62 (1.14, 65.20)	13.45 (1.59, 114.11)

^†Adjusted for age, gender, waist circumference based on the model presented from the Wisconsin sleep cohort¹⁴; n = 294.

^·Additionally adjusted for BMI, mean arterial pressure, and HDL cholesterol (n = 291).

DISCUSSION

In the Busselton cohort, moderate-to-severe sleep apnea was a univariate risk factor for prevalent diabetes and an independent risk factor for the 4-year incidence of the condition. The longitudinal model remained significant after adjustment for age, gender, body mass index, waist circumference, HDL cholesterol, and mean arterial pressure. This is the first report of a significant independent association between obstructive sleep apnea (OSA) and incident diabetes^4,14 and confirms observations from clinic-based studies and from cross-sectional population-based studies that the conditions are linked.^14,24–26 However, given the small number of participants, the magnitude and significance of these findings will require replication in other larger studies or the quantitative combination of a number of such studies in a meta-analysis. These data are also the first from an established population-based cohort from outside of the USA to confirm these observations.

Previous studies include cross-sectional analyses from 3 American cohorts^14,24–26 showing that sleep apnea is associated with diabetes or its early markers. However these cross-sectional findings failed to rule out the possibility of reverse-causation—that the diabetic process causes sleep apnea. Reichmuth and colleagues also undertook longitudinal analyses of their data from the Wisconsin Sleep Cohort. They used 2 methods for defining diabetes—physician diagnosis alone or a composite of diagnosis and/or a fasting glucose ≥ 7 mmol/L (equivalent to 126 mg/dL).¹⁴

Reichmuth et al.'s longitudinal analyses (n = 978 people) are similar to the Busselton sleep apnea cohort, with similar duration of follow-up (approximately 4 years) and similar age of the participants at baseline (W = 30-60 versus B = 40-65). The Wisconsin investigators found that, although sleep disordered breathing was a risk factor for the incidence of physician diagnosed diabetes at the univariate level (apnea hypopnea index ≥ 15/hr vs. AHI < 5/hr OR = 4.06, 95% CL 1.86–8.85), this association was no longer significant after adjustment for age, sex, and waist circumference (OR = 1.62, 95% CL 0.67–3.65). When they used exactly the composite definition of diabetes and adjusted for age, gender, and waist circumference, they found no evidence that moderate-to-severe SDB was a risk factor for incident diabetes (OR = 0.91, 0.36–2.33). In contrast, our findings, using this same composite definition of diabetes, are consistent with a large increased risk for diabetes.

What are the potential salient differences between the cohorts that might explain the divergent results? The Wisconsin report does not include a cross-tabulation of the exact numbers of people who were classified as diabetic (via the composite method), but it appears that the Busselton cohort has a greater incidence of diabetes defined this way, is slightly older, leaner, less apneic, and, more importantly, does not have the same risk factor structure with regard to diabetes as Wisconsin. Differences between the standard of healthcare provided to Wisconsin state employees and residents of Busselton, Western Australia might also have caused differences in the likelihood that a person would be diagnosed as diabetic. Differences in genetic inheritance could also play a role. More information from other community-based cohorts is clearly needed, and meta-analyses of these cohorts may be needed to give more precise risk estimates than we are able to provide. We have aimed to facilitate this process by reporting 4-year data with comparable data analyses to the existing report.¹⁴

If well-designed clinical trials of sleep apnea treatment show improvement in diabetic indices, this would offer evidence that sleep apnea causes diabetes. However, despite a very promising finding from a non-randomized study,²⁷ randomized controlled trials of continuous positive airway pressure for sleep apnea have not yet shown positive effects.^28,29 These negative findings do not rule out the possibility that sleep apnea causes diabetes, but they have so far failed to confirm the hypothesis that diabetes might be reversed via the treatment of OSA. They also do not rule out the possibility that CPAP might effectively prevent diabetes in people with sleep apnea, however this must be demonstrated in studies of sufficient duration and power.

In neither the Busselton nor Wisconsin¹⁴ cohorts was mild OSA significantly associated with greater risk of incident diabetes (Busselton odds ratio =1.51, 95% CL = 0.25, 9.12) compared to no OSA. However, the increased odds ratio is consistent with increasing risk for diabetes as the severity of OSA rises. This observation needs to be confirmed by cohorts with greater power or by cohort combination (i.e., meta-analysis). Confirming the association with mild OSA might be more important from a public health perspective because mild OSA is far more common than moderate-severe OSA.^30,31 Alternatively, confirmation that this mild subgroup are not at risk would allow health care resources to be more appropriately distributed to those who might benefit.

Sleep apnea was measured using the MESAM IV device rather than polysomnography. This device had very close agreement with standard PSG (ICC = 0.98)^3,18 in the original prevalence study, as well as being validated by other research groups.^18,20,21 Moreover, the prevalence estimate using this method in Busselton was in very close agreement (26% versus 24%) with the PSG measured prevalence estimate from the widely accepted prevalence data from the Wisconsin Sleep Cohort.³² The use of a 3% desaturation criterion and the limited number of physiological channels employed by MESAM IV is not a method currently recommended by the most recent American Academy of Sleep Medicine guidelines for the measurement of sleep apnea. However, the ability of the MESAM IV device to predict mortality associated with sleep apnea³—and in these analyses to predict diabetes—also provides evidence that this may be a valid OSA measurement technique in community-based epidemiological investigations. The optimal way to score and categorize AHI to communicate prognostic information and to guide treatment decisions has yet to be determined.

We did not have information about, nor did we attempt to control for, any effects of sleep apnea treatment on the incidence of diabetes or metabolic syndrome. Treatment effects are unlikely to explain the observed associations as any effective OSA treatment/s should reduce risk, not increase it. Furthermore, only around 2% of the Australian population have accessed diagnostic and treatment services; access is lower in the state of Western Australia^33,34 and probably much lower in rural communities such as Busselton.

Visceral obesity is an important risk factor for the development of diabetes. While we have attempted to control for visceral obesity by the inclusion of both BMI and waist circumference in our models, the measurement of both is only a proxy measure that does not fully capture the variability in visceral obesity. Thus it is likely that some residual confounding remains. It has also been proposed that habitual sleep durations might play a role in the development of diabetes.^5,35–37 At present, there are no cohort studies in which sleep has been measured objectively that have found an association with incident diabetes. Unfortunately, we do not have an objective measure of sleep duration in this cohort, and thus sleep duration remains a potential uncontrolled confounder.

Assuming that OSA is independently confirmed as a risk factor for incident diabetes, it might be added to the list of standard diabetes risk factors. Should this occur, the evidence base supports the need for high-quality randomized controlled trials of treatment/s for sleep apnea that are powered to detect primary disease prevention.

DISCLOSURE STATEMENT

This was not an industry supported study. The authors have indicated no financial conflicts of interest.