Association between obstructive sleep apnea severity and endothelial dysfunction in an increased background of cardiovascular burden

Authors


Correspondence

Reena Mehra, MD, MS, Case Western Reserve University, 11100 Euclid Avenue, Cleveland, OH 44106-6003, USA.

Tel.: 216-844-5218;

fax: 216-844-8708;

e-mail: reena.mehra@case.edu

Summary

The objective of this study is to examine whether increasing obstructive sleep apnea (OSA) severity is associated with worsening endothelial function. The design is a cross-sectional examination of the baseline assessment of a multi-centre randomized controlled clinical trial examining the effects of oxygen, continuous positive airway pressure (CPAP) therapy or lifestyle modifications on cardiovascular biomarkers. Participants were recruited from cardiology clinics at four sites. Participants with an apnea–hypopnea index (AHI) of 15–50 and known cardio/cerebrovascular disease (CVD) or CVD risk factors were included. OSA severity indices [oxygen desaturation index (ODI), AHI and percentage of sleep time below 90% oxygen saturation (total sleep time <90)] and a measure of endothelium-mediated vasodilatation [Framingham reactive hyperaemia index (F-RHI) derived from peripheral arterial tonometry (PAT)] were assessed. The sample included 267 individuals with a mean AHI of 25.0 ± 8.5 SD and mean F-RHI 0.44 ± 0.38. In adjusted models, the slope of the relationship between ODI and F-RHI differed above and below an ODI of 24.6 (= 0.04), such that above an ODI of 24.6 there was a marginally significant decline in the geometric mean of the PAT ratio by 3% [95% confidence interval (CI): 0%, 5%; = 0.05], while below this point, there was a marginally significant incline in the geometric mean of the PAT ratio by 13% (95% CI: 0%, 27%; = 0.05) per 5-unit increase in ODI. A similar pattern was observed between AHI and F-RHI. No relation was noted with total sleep time <90 and F-RHI. There was evidence of a graded decline in endothelial function in association with higher levels of intermittent hypoxaemia.

Introduction

Obstructive sleep apnea (OSA) is a prevalent disorder characterized by repetitive complete or partial upper airway collapse leading to adverse physiological consequences. Epidemiological data provide strong evidence implicating OSA as an independent risk factor for cardiovascular morbidity and mortality (Peker et al., 2002; Peppard et al., 2000). Several physiological effects of OSA have been proposed to explain the pathogenesis of cardiovascular morbidity. Endothelial dysfunction is one mechanism that may result from OSA-related intermittent hypoxaemia, oxidative stress, enhanced sympathetic nervous system activation and increased blood pressure (Dean and Wilcox, 1993). Endothelial dysfunction is characterized by alteration of normal endothelial physiology consisting of a reduction in the bioavailability of vasodilators such as nitric oxide leading to impaired endothelium-dependent vasodilation. Endothelial dysfunction is considered to represent the integrated functional expression of cardiovascular risk factor burden and a reflection of atherogenic vascular milieu (Corretti et al., 2002). The clinical relevance of endothelial dysfunction is supported by the consistency of its associations with incident cardiovascular disease events (Brunner et al., 2005). In fact, the persistent impairment of endothelial function in individuals with established coronary artery disease despite optimal medical therapy has been observed to be a strong independent predictor of adverse cardiovascular events (Corretti et al., 2002; Poredos and Jezovnik, 2012).

Clinic-based studies of individuals mostly without overt cardio/cerebrovascular disease (CVD) have suggested that OSA is associated with impaired brachial artery flow-mediated dilation (FMD), a surrogate of endothelial dysfunction (Kato et al., 2000). The association of OSA with the primarily nitric oxide-dependent (Nohria et al., 2006) endothelial-mediated vasodilation impairment, as assessed by brachial ultrasound FMD or finger plethysmography, has been demonstrated in several studies (Ip et al., 2004; Itzhaki et al., 2005). However, other studies have failed to demonstrate these relationships (Chami et al., 2009; Nieto et al., 2004). The two largest epidemiological studies (Chami et al., 2009; Nieto et al., 2004) showed no association between OSA defined by the apnea–hypopnea index (AHI) and endothelial dysfunction measured by FMD after adjusting for obesity. Also, disparate findings have been noted with sleep-related hypoxia and endothelial dysfunction (Chami et al., 2009; Nieto et al., 2004). Although endothelial function has been reported to improve after OSA treatment, these studies were non-randomized (Bayram et al., 2009) or involved a small sample size (Ip et al., 2000). Furthermore, many studies have been limited by single-centre geographic distributions (Bayram et al., 2009; Itzhaki et al., 2005), and involved evaluation of endothelial dysfunction with techniques such as brachial artery ultrasound that may be prone to intra- and interoperator variability (Ip et al., 2004; Itzhaki et al., 2005).

In this analysis, we examined the relationship of several common metrics of OSA severity with reactive hyperaemia peripheral arterial tonometry (PAT), a measure shown to assess endothelial dysfunction accurately and associated with fewer technical difficulties and operator dependency than arterial ultrasound (Corretti et al., 2002) in a sample of individuals with moderate to severe OSA and CVD risk factors participating in the baseline examination of a multi-centre trial. We carefully explored the thresholds and ‘dose–response’ relationships among OSA metrics and reactive hyperaemia. We postulate that increasing severity of OSA defined by AHI, and alternatively by measures of hypoxaemia, will be associated linearly with impaired endothelial function, even after adjustment for confounders such as obesity and standard cardiovascular risk factors.

Methods

Study samples

Patients with moderate to severe OSA were recruited from outpatient cardiology clinics at four sites (Brigham and Women's Hospital, Case Medical Center, Johns Hopkins University and Veteran's Affairs Boston Healthcare System) as part of a randomized controlled trial (Heart Biomarker Evaluation in Apnea Treatment—HeartBEAT) aimed at comparing conservative medical therapy, supplemental nocturnal oxygen therapy and positive airway pressure therapy on cardiovascular biomarkers in OSA (www.clinicaltrials.gov Trial Registration number: NCT01086800).

Study protocol

Screening sleep questionnaires were administered either through mailings to targeted participants receiving care at collaborating clinics, or by direct administration at the time of routine clinic appointments to determine potential eligibility. These screening questionnaires included the Epworth Sleepiness Scale (ESS;) (Johns, 1991), which quantifies the likelihood of falling asleep in a number of common situations, and the Berlin Questionnaire (Netzer et al., 1999), a simple 10-item questionnaire that categorizes OSA risk in three domains: snoring/nocturnal breathing disruption, sleepiness/fatigue and obesity or hypertension. Those who scored ≥16 on the ESS or had drowsy driving were excluded from participation. Subjects who had a positive score (greater than two of three domains) on the Berlin questionnaire indicating a high likelihood of OSA underwent more detailed eligibility assessment. Inclusion criteria were: age 45–75 years and patients at high risk for cardiovascular disorders defined as: (i) established stable coronary artery disease (documented prior myocardial infarction or coronary revascularization >3 months prior to entry or angiographically documented ≥50% stenosis in a major coronary artery); or (ii) ≥3 cardiovascular risk factors characterized by: (a) hypertension (HTN), defined by physician-reported hypertension or antihypertensive medication use [including angiotensin-converting enzyme (ACE) inhibitor, angiotensin receptor blocker, beta adrenergic blocker, alpha adrenergic blocker, diuretic and calcium channel blocker usage] or systolic blood pressure >140 mmHg or diastolic blood pressure >90 mmHg; (b) diabetes mellitus treated by a physician; (c) body mass index (BMI) ≥30 kg m2; or (d) dyslipidaemia defined by a total cholesterol >240 mg dl−1, low-density lipoprotein (LDL) cholesterol >160 mg dl−1 or high-density lipoprotein (HDL) cholesterol <45 mg dl−1; or physician-diagnosed dyslipidaemia treated by medication). Exclusion criteria were: central sleep apnea (central apnea index >5); nocturnal oxygen saturation <85% for >10% of the record; heart failure [ejection fraction <35% or New York Heart Association (NYHA) ≥2]; poorly controlled hypertension (HTN; >170 mmHg/>110 mmHg); prior stroke with functional impairment; severe uncontrolled medical problems; severe chronic insomnia or circadian rhythm disorder with <4 h of sleep per night; resting wake oxygen saturation <90%; smoking in the location of sleep; current use of either supplemental oxygen or positive airway pressure; and <3 months since myocardial infarction, stroke or any revascularization procedure. Subjects who met these eligibility criteria then underwent unattended type 3 sleep studies; those with an AHI between 15 and 50 events h−1 were considered eligible and scheduled for a research visit. Institutional Review Board approval was obtained from all sites. Full written informed consent was obtained.

Data collection

Sleep apnea assessment

At the screening visit, subjects were instructed in the use of the sleep monitor (Embletta; Embla, Broomfield, CO, USA). The studies were scored by a trained, registered polysomnologist following the American Academy of Sleep Medicine guidelines for alternative hypopnea definitions with modification, such that arousal was not considered in the identification of hypopneas (Ruehland et al., 2009). An apnea was defined as a complete cessation of airflow, measured using nasal pressure, for ≥10 s. Hypopnea was defined as 50% reduction in breathing amplitude lasting ≥10 s associated with ≥3% oxygen desaturation. The following parameters were obtained: oxygen desaturation index (ODI) defined as the number of oxygen desaturations ≥3% per hour of analysed recording time, AHI defined as the number of apneas and hypopneas per hour of analysed recording time and total sleep time (TST) below 90% oxygen saturation (TST <90). In the determination of the total recording time or analysed time, sleep onset and offset were marked by the scorer by taking into consideration data self-reported in the sleep log.

Measure of endothelial function

Peripheral arterial tonometry (PAT) was measured using the Endo-PAT2000 device (Itamar Medical Ltd, Caesarea, Israel). The test was approximately 20 min in duration and performed in the morning in a quiet environment with the participants in a supine position after a 12-h fasting period that included refraining from smoking and drinking caffeinated beverages. A blood pressure cuff was placed on the non-dominant arm; the other arm was used as a control and measurements made according to published guidelines (Brunner et al., 2005). The study consisted of three phases: (i) a 5-min period of baseline recording; (ii) a 5-min period of occlusion of the brachial artery where the blood pressure cuff is inflated to 60 mmHg above systolic blood pressure (and to at least 200 mmHg); and (iii) a release period where the cuff is deflated rapidly (Corretti et al., 2002; Poredos and Jezovnik, 2012). Endothelium-mediated vasodilatation was assessed by measuring pulse wave amplitude in the finger before and after 5 min of brachial artery occlusion. The Framingham reactive hyperaemia index (F-RHI), which has been associated with multiple CVD risk factors (Hamburg et al., 2008) and identified to have superior reproducibility relative to the standard reactive hyperaemia index (RHI; Selamet Tierney et al., 2009), was calculated as the natural logarithm of the PAT ratio, given by the ratio of the average pulse amplitude in the post-hyperaemic phase (during the 90–120-s post-deflation interval) divided by the average baseline amplitude (Hamburg et al., 2008), normalized by the ratio of pulse amplitudes obtained from corresponding measurements in the non-occluded arm. A lower F-RHI is consistent with poorer endothelial function, as it is reflective of a smaller increase in the post-hyperaemic pulse amplitude relative to baseline. All individual PAT raw data tracings were reviewed manually (F.S.) to assess signal quality and to assign quality grades. Those observations with poor-quality tracings were excluded from analysis (= 41 of 318 baseline examinations, 12.9%). The reliability of this quality grade assignment was ascertained by a random rescoring of data quality grades (= 50 studies) and was consistent with excellent intra-observer reliability with an overall intraclass correlation coefficient (ICC) using Kendall's coefficient of concordance of 0.88 and an ICC using Cohen's kappa in distinguishing poor-quality studies (grade 4) from those that were included in the current study (good- and adequate-quality studies, grades 1–3) of 0.86. Details of the quality grading protocol are provided in an online supplement.

Statistical methods

Standard descriptive statistics were used to describe the study sample. Continuous data were presented as mean ± SD and categorical data in percentages. The OSA exposure metrics evaluated included: ODI, AHI and TST <90. To identify possible inflection points in the association between OSA metrics and endothelial dysfunction, the linear model as well as three piecewise linear models (defined using one knot at the first, second and third quartile for each OSA metric) were compared using a 0.632 bias-corrected mean squared error (MSE) that was obtained from a leave-one-out bootstrap cross-validation procedure based on 5000 bootstrap samples. The final model was selected using the model that minimized the MSE. Cross-validation procedures of this type are used commonly to select the best-fitting model among competing models.

To examine the effect of potential confounders on these associations, three pre-specified multivariable models were considered: model 1 (adjusted for the OSA exposure and site only), model 2 (adjusted for model 1 covariates as well as age, sex, race and BMI) and model 3 (adjusted for model 2 covariates as well as HTN, diabetes (diagnosed or taking oral hypoglycaemic medications or insulin), dyslipidaemia (diagnosed or on statin medication), smoking (pack years) and established CVD. Established CVD was defined by the presence of coronary artery disease (CAD; per above, outlined under inclusion criteria) or >3 months since stroke occurred without functional impairment. We also explored if either CVD or diabetes mellitus was an effect modifier of the association between each exposure and F-RHI by including an interaction term between either CVD or diabetes mellitus and each exposure in the model. In order to facilitate interpretation of the linear and piecewise linear models, the parameter estimates were exponentiated and these transformed estimates were interpreted as the geometric mean ratio of the PAT ratio. All tests were performed assuming a significance level of 0.05 and using sas version 9.2 (SAS Institute Inc., Cary, NC, USA) for analyses and r version 2.9.2 for graphs.

Results

Subject characteristics

Of the 318 participants with baseline examinations, five had missing PAT data, 41 had poor-quality pulse wave amplitude tracings and five had missing covariate data, resulting in 267 participants in the analytical sample. There were no statistically significant differences in subject characteristics between the analytical sample and those who were excluded (age 62.9 ± 7.2 versus 63.3 ± 7.9 years, male gender 71.5% versus 84.3% and BMI 34.4 ± 6.6 versus 34.1 ± 6.2 kg m2, respectively). As expected in a cohort at high CVD risk with OSA, the majority of the analytical sample included older, obese participants who were predominantly men. By design, the sample was not excessively sleepy (ESS = 8.9 ± 3.7). More than half the cohort had documented coronary artery disease and there was a high prevalence of hypertension (Table 1).

Table 1. Baseline subject characteristics
Baseline characteristicsAnalytical sample (= 267)
Mean (SD)/frequency (percentage)
  1. Continuous data were presented as mean ± SD and categorical data as frequencies and percentages.

  2. ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blocker; CMC, Case Medical Center; BWH, Brigham and Women's Hospital; JHU, Johns Hopkins University; BVA, Boston Veteran's Administration Hospital; F-RHI, Framingham reactive hyperaemia index; PAT, peripheral arterial tonometry.

  3. a

    The geometric mean and geometric SD.

Age (years)62.9 (7.2)
Male191 (71.5%)
Caucasian212 (79.4%)
Body mass index (kg m2)34.4 (6.6)
Dyslipidaemia257 (96.3%)
History of smoking166 (62.2%)
Smoking (pack years)20.1 (27.8)
Coronary artery disease138 (51.7%)
Stroke13 (4.9%)
Cardio/cerebrovascular disease140 (52.4%)
Diabetes mellitus123 (46.1%)
Hypertension + blood pressure medication use236 (88.4%)
Anti-hypertensive medications257 (96.3%)
ACE inhibitor or ARB189 (70.8%)
Beta adrenergic blocker177 (66.3%)
Alpha adrenergic blocker6 (2.2%)
Calcium channel blocker79 (29.6%)
Diuretics102 (38.2%)
Epworth Sleepiness Score8.9 (3.7)
SiteCMC: 85 (31.8%)
BWH: 44 (16.5%)
JHU: 73 (27.3%)
BVA: 65 (24.3%)
Apnea–hypopnea index (AHI)25.0 (8.5)
Oxygen desaturation index (ODI)32.3 (10.1)
Minimum oxygen saturation79.3 (5.6)
Percentage sleep time less than 90% oxygen saturation9.8 (13.9)
F-RHI0.44 (0.38)
PAT ratio1.55 (1.46)a

ODI and F-RHI

The piecewise linear regression model that resulted in the minimum MSE for ODI was based on an inflection point at the first quartile of ODI (24.6; MSE = 0.1234 versus 0.1236–0.1249 for all other models). This model was then used to estimate the geometric mean ratio in the PAT ratio per 5-unit increase in ODI above and below an ODI of 24.6. As shown in Table 2 and Fig. 1, we found that the relationship between ODI and F-RHI differed above and below an ODI of 24.6 (= 0.035). Furthermore, there was a marginally significant 13% increase in the geometric mean of the PAT ratio when ODI was less than 24.6 [95% confidence interval (CI): 0%, 27%; P-value = 0.05] and a marginally significant 3% decrease in the geometric mean of the PAT ratio after this inflection point (95% CI: 0%, 5%; = 0.05) for every 5-unit increase in ODI.

Table 2. Geometric mean ratio of pulse arterial tonometry ratio (PAT) for every 5-unit increase in oxygen desaturation index
ModelODIGeometric mean ratio of PAT ratio (95% CI); P-valueaTest if geometric mean ratio of PAT ratio differs before and after 24.6b
  1. Model 1: adjusted for site.

  2. Model 2: adjusted for site, age, gender, race and body mass index.

  3. Model 3: adjusted for site, age, gender, race, body mass index, hypertension + high blood pressure medication use, diabetes, dyslipidaemia, smoking pack per years and cardiovascular disease.

  4. a

    Geometric mean ratio of the PAT ratio for every 5-unit increase in the oxygen desaturation index.

  5. b

    Test if geometric mean ratio of the PAT ratio for every 5-unit increase in the oxygen desaturation index differs before and after 24.6.

  6. CI, confidence interval; ODI, oxygen desaturation index.

Model 1<24.6 (= 66)1.14 (1.01, 1.29); = 0.040.02
≥24.6 (= 201)0.97 (0.95, 1.00); = 0.04
Model 2<24.6 (= 66)1.14 (1.01, 1.28); = 0.030.03
≥24.6 (= 201)0.98 (0.95, 1.00); = 0.07
Model 3<24.6 (= 66)1.13 (1.00, 1.27); = 0.050.04
≥24.6 (= 201)0.97 (0.95, 1.00); = 0.05
Figure 1.

Plot of oxygen desaturation index versus pulse arterial tonometry (PAT) ratio. Each point represents the oxygen desaturation index (ODI) and corresponding PAT ratios for each given subject. The axis opposite to the PAT ratio provides a histogram for the PAT ratio, while the axis opposite to ODI provides a histogram for ODI. The solid line indicates the model estimates for the adjusted geometric mean of the PAT ratio across the range of ODI for a hypothetical 63-year-old Caucasian male studied at Case Medical Center with a body mass index (BMI) of 33.3, 6.5 smoking pack years, hypertension, dyslipidaemia and coronary vascular disease, but without diabetes. Dotted lines indicate the associated 95% confidence interval.

AHI and F-RHI

The piecewise linear regression model that resulted in the minimum MSE for AHI was based on an inflection point at the first quartile of AHI (18.4; MSE = 0.1244 versus 0.1251–0.1259 for all other models). As seen in Table 3 and Fig. 2, there was evidence that the fully adjusted association between AHI and F-RHI differed when AHI <18.4 versus AHI ≥18.4 (= 0.04). Additionally, while there was a statistically significant 26% increase in the geometric mean of the PAT ratio per 5-unit increase in AHI when AHI was less than 18.4 (95% CI: 11%, 58%; P-value = 0.04), there was no statistically significant association between AHI and F-RHI above this inflection point.

Table 3. Geometric mean ratio of the pulse arterial tonometry ratio (PAT) for every 5-unit increase in apnea–hypopnea index (AHI)
ModelAHIGeometric mean ratio of PAT ratio (95% CI); P-valueaTest if geometric mean ratio of PAT ratio differs before and after a threshold of 18.4b
  1. Model 1: adjusted for site.

  2. Model 2: adjusted for site, age, gender, race and body mass index.

  3. Model 3: adjusted for site, age, gender, race, body mass index, hypertension + high blood pressure medication use, diabetes, dyslipidaemia, smoking pack per years and cardiovascular disease.

  4. a

    Geometric mean ratio of PAT ratio for every 5-unit increase in the apnea–hypopnea index.

  5. b

    Test if geometric mean ratio of PAT ratio for every 5-unit increase in the apnea–hypopnea index differs before and after 18.4.

Model 1<18.4 (= 66)1.25 (0.99, 1.56); = 0.060.05
≥18.4 (= 201)0.98 (0.95, 1.01); = 0.16
Model 2<18.4 (= 66)1.26 (1.01, 1.58); = 0.040.04
≥18.4 (= 201)0.99 (0.96, 1.01); = 0.32
Model 3<18.4 (= 66)1.26 (1.01, 1.58); = 0.040.04
≥18.4 (= 201)0.98 (0.95, 1.01); = 0.26
Figure 2.

Plot of the apnea–hypopnea index (AHI) versus pulse arterial tonometry ratio. Each point represents the AHI and corresponding pulse arterial tonometry (PAT) ratio for each given subject. The axis opposite to the PAT ratio provides a histogram for the PAT ratio, while the axis opposite to AHI provides a histogram for AHI. The solid line indicates the model estimate for the adjusted geometric mean of the PAT ratio across the range of AHI for a hypothetical 63-year-old Caucasian male studied at Case Medical Center with a body mass index of 33.3, 6.5 smoking pack years, hypertension, dyslipidaemia and coronary vascular disease but without diabetes. Dotted lines indicate the associated 95% confidence interval.

Time at oxygen saturation <90% and F-RHI

The linear regression model for TST <90 resulted in the minimum MSE (MSE = 0.1251 versus 0.1251–0.1259 for piecewise linear regression models). However, there was no statistically significant association between TST <90 and F-RHI; i.e. for every 5-unit increase in TST <90, the adjusted geometric mean of the PAT ratio decreased by 1% (95% CI: 0%, 3%; P-value = 0.12).

Finally, to explore whether the association between any of the three OSA metrics and F-RHI was modified by CVD or diabetes mellitus, separate interaction terms were included in all models. There was no evidence that either CVD or diabetes mellitus was an effect modifier.

Discussion

Although numerous studies have demonstrated that OSA is associated with CVD as well as with abnormalities in intermediate endpoints, such as endothelial function, blood pressure and insulin resistance, there is uncertainty regarding which levels of OSA severity confer greatest risk for CVD. In this secondary analysis of a sample of patients with an AHI range of 15–50, a series of rigorous statistical analyses found some evidence of declining endothelial function in individuals with a moderate to severely elevated ODI (≥24.6). In contrast, there was a positive, albeit marginally significant, trend between increasing ODI and endothelial function at levels of ODI between 13.9 and 24.6. A similar pattern was observed when modelling the association between AHI and F-RHI; i.e. a positive slope was observed below an AHI threshold of 18.4 while a negative slope was seen after this threshold. The finding that impaired endothelial function is most evident at higher levels of hypoxaemia is consistent with prior work from the Sleep Heart Health Study, which has demonstrated that rates of mortality (Punjabi et al., 2009), stroke (Redline et al., 2010) and coronary artery disease and heart failure (Gottlieb et al., 2010) increase most at moderately elevated AHI levels (20–30). In contrast, increased prothrombotic potential as measured by plasminogen activator inhibitor-1 and fibrinogen has been described at low levels of AHI exposure with a plateau effect at moderate levels of OSA (Mehra et al., 2010). Further research is needed to address differences in the sensitivity of various intermediate measures of cardiovascular disease to different degrees of OSA-related stress.

The results of this study suggest that endothelial dysfunction may manifest at a certain critical level of OSA severity, where protective mechanisms are lost. The finding suggesting a trend for improved endothelial function at an ODI of 13.9–24.6 in our study is consistent with a potential protective influence of mild to moderate levels of intermittent hypoxia (Pohlman and Harlan, 2000). This inference is limited by the restricted AHI range represented in this study sample and lack of data on individuals with no or milder levels of OSA. However, the mean F-RHI in those with an ODI <24.6 in this study is similar to that noted in a non-OSA sample (0.44 ± 0.38 versus a normal range of 0.5–0.6; Selamet Tierney et al., 2009).

Although ODI and AHI were correlated strongly (= 0.85) and the relationship of each with F-RHI was similar, associations appeared to be somewhat stronger for ODI. The ODI captures oxygen desaturation events not associated with scoreable reductions in flow, and incorporates a running average to identify baseline levels probably accounting for the higher ODI versus AHI values. This may reflect less measurement error in an index that is derived automatically compared to one that requires manual annotation, or because ODI may measure more directly the relevant exposure (intermittent hypoxia). Abundant research suggests that chronic intermittent hypoxia may affect endothelial function adversely through a myriad of pathways, including up-regulation of nuclear factor (NF) kappa B (Jurado-Gamez et al., 2012), increased reactive oxygen species (Lavie, 2003) and reduced availability of endothelial nitric oxide (Ip et al., 2000).

Endothelial function was measured using PAT, which is a non-invasive technique that enables plethysmographic recording of pulse wave amplitude, a measure of changes in the arterial pulsatile volume of the distal phalanx of the finger, before and during reactive hyperaemia (Corretti et al., 2002; Poredos and Jezovnik, 2012). PAT measures flow response hyperaemia, which is related to the endothelial function of small arteries and to the endothelial function of the microcirculation, therefore providing information on the functional capability of the microcirculation. The baseline pulse wave analysis is ascertained by plethysmographic finger cuffs placed simultaneously on the index fingers of both hands for 5 min. An inflatable cuff is used to induce hyperaemia by occluding blood flow through the brachial artery for 5 min. The reactive hyperaemia index is then calculated as the ratio of the average pulse wave amplitude between the post- and pre-occlusion values (Poredos and Jezovnik, 2012).

Due to the superior widespread applicability, lesser degree of operator dependence, lower technical difficulty, ability to provide information regarding the control of arterial tone at rest and comparable accuracy of PAT compared to brachial artery ultrasound, PAT was used as a measure of endothelial function in the current study (Corretti et al., 2002). We analysed the F-RHI, which is considered to represent the most clinically relevant portion of the hyperaemia response and has a stronger relationship to known cardiovascular risk factors than does the traditional RHI (Hamburg et al., 2008). Digital vascular function measured by PAT and conduit vascular function measured by brachial artery ultrasound may correlate with different risk factors in subjects with low cardiovascular disease burden (Hamburg et al., 2011). However, in samples with increased cardiovascular disease prevalence (~50%) similar to our study sample, digital vascular function measured by PAT is related significantly to vasodilator function in conduit vessels (Dhindsa et al., 2008).

The relationship between hypoxia and impaired endothelial function has been studied in epidemiological (Nieto et al., 2004) as well as clinic-based studies (Bayram et al., 2009; Kraiczi et al., 2001). Similar to our findings, data from the Framingham Heart Study site of the Sleep Heart Health Study did not report a significant association between AHI and hypoxia (TST <90) with endothelial dysfunction measured by brachial artery percentage FMD (Chami et al., 2009). In a different Sleep Heart Health Study subgroup representing older participants from the Cardiovascular Health Study, a significant association was noted with increased baseline brachial artery diameter and overall hypoxia (TST <90; Nieto et al., 2004). However, FMD and baseline diameter were not associated statistically with AHI after adjusting for BMI. In contrast, in a clinic-based sample, FMD was correlated significantly and inversely with TST <90; however, the effect of confounders such as obesity was not assessed (Kraiczi et al., 2001). Similar to our findings, clinic-based studies have identified associations between ODI with FMD and reactive hyperaemia (Jurado-Gamez et al., 2012), but not TST <90, after consideration of obesity (Chung et al., 2007). Unlike the current investigation, the vast majority of these studies used brachial artery ultrasound to define endothelial dysfunction, which may be prone to intra- and interobserver measurement variability. None of these studies evaluated inflection points carefully in the OSA–endothelial function relationship.

The strengths of the current study include the use of a technique to evaluate endothelial function that is reliable, reproducible, not operator-dependent and able to characterize endothelial function accurately (Corretti et al., 2002) in a multi-centre sample with a high background of cardiovascular risk. We employed a quality grading system involving careful review of the endothelial arterial tonometry raw data with high intrascorer reliability. Standardized methods were used for the collection of sleep and vascular measures, which were scored by certified technicians using centralized reading and subject to quality control procedures. We also accounted carefully for various confounding factors. To identify the possibility of threshold effects when modelling the association between OSA severity metrics and endothelial dysfunction, we used a commonly used statistical method based on cross-validation to select the final model.

The main limitation in our study is the restricted range of AHI, which precludes our ability to make inferences on levels of AHI below 15 or above 50. However, despite this limited range of OSA severity, we found evidence of a non-linear association with endothelial function across this OSA severity rate, with evidence of an inflection point near the first quartile of ODI (24.6) and the first quartile of AHI (18.4). Also, our study is limited by its cross-sectional nature precluding the inference of temporal relationships. Given the increased participant burden, we did not investigate vascular reactivity after administration of nitroglycerin, an endothelium-independent donor of nitric oxide.

In summary, the findings of this study provide evidence that moderate to severe intermittent hypoxia (defined by ODI) is associated with decrements in endothelial function among individuals with high cardiovascular risk or with established cardiovascular disease. Future studies should consider potential non-linear effects of intermittent hypoxia.

Acknowledgements

This study was supported by NIH National Heart Lung Blood Institute RC2 HL101417 and K23 HL079114, NIH M01 RR00080, American Heart Association National Scientist Development Award 0530188N, Central Society of Clinical Research, NHLBI K08 HL081385 and NCI 1U54CA116867. The project described was also supported by UL1 RR024989 from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH) and its contents are solely the responsibility of the authors and do not necessarily represent the official view of NCRR or NIH.

Disclosure Statement

Dr Sanjay Patel has served as a consultant for Sleep Health Centers and Apnex and has received research support from Philips Respironics. Dr Deepak L. Bhatt discloses the following relationships: Advisory Board, Medscape Cardiology; Board of Directors, Boston VA Research Institute, Society of Chest Pain Centers; Chair, American Heart Association Get With The Guidelines Science Subcommittee; honoraria from American College of Cardiology (Editor, Clinical Trials, Cardiosource), Duke Clinical Research Institute (clinical trial steering committees), Slack Publications (Chief Medical Editor, Cardiology Today Intervention), WebMD (CME steering committees); research grants from Amarin, AstraZeneca, Bristol-Myers Squibb, Eisai, Ethicon, Medtronic, Sanofi Aventis, The Medicines Company; and unfunded Research: FlowCo, PLxPharma and Takeda. Dr Eldrin F. Lewis discloses the following: ResMed, research grant support (minor); Novartis, Inc., research grant support (major); Amgen, Inc., research grant support and consulting (major); Theracos, research grant support (minor); Sunovian, research grant support (minor); and Sanofi Aventis, research grant support (minor). Dr Naresh M. Punjabi received research grant support paid to Johns Hopkins University for a multi-centre study on CPAP therapy in patients with obstructive sleep apnea. Dr Susan Redline discloses the following: receipt of a grant paid to Brigham and Women's Hospital from ResMed Foundation, and Brigham and Women's Hospital has received equipment for use in NIH studies from ResMed Inc and Philips-Respironics. Dr Reena Mehra serves on the Medical Advisory board for CareCore and has given presentations for the American Academy of Sleep Medicine. University Hospitals Case Medical Center has received positive airway pressure machines and equipment from Respironics for research for which Dr Mehra is the Principle Investigator. Dr Fadi Seif, Dr Harneet Walia, Michael Rueschman, Dr Daniel J. Gottlieb, Dr Susheel Patil and Dr Denise Babineau have no disclosures.

Ancillary