Sleep Apnea and Diabetes Predict Progression of Cardiac Diastolic Dysfunction in Pre-heart Failure Patients: a Prospective Cohort HSCAA Study

[Background] Sleep apnea, a common co-morbid condition of diabetes, has been shown to be associated with established heart failure. However, its role in association of the presence of diabetes in the progression of cardiac diastolic function in pre-heart failure phase is not known. This prospective study is to longitudinally examine the predictive value of sleep apnea and diabetes on progression of left ventricular (LV) diastolic dysfunction in patients without heart disease. [Methods] Among 976 patients registered in HSCAA prospective cohort study, 517 without heart disease (175 type 2 diabetes, 342 non-diabetes) were followed with repeated echocardiography in every 1-3 years for a mean 34.7 months. LV diastolic dysfunction was determined by transmitral early inow velocity/early diastolic tissue velocity (E/e´) >14. Annual change in E/e’ was calculated by using the slope of the linear regression line calculated from at least 3 echocardiographic measurements. Sleep apnea was determined by an apnomonitor device in conjunction with percutaneous oxygen saturation, and apnea hypopnea index (AHI) was calculated. [Results] Kaplan-Meier analyses revealed that subjects with diabetes or sleep apnea had a signicantly (p<0.01) greater risk for LV diastolic dysfunction, with hazards ratio of 2.21 (1.41-3.47) and 2.23 (1.40-3.55), respectively. Subjects with sleep apnea exhibited signicantly higher risk for LV diastolic dysfunction in non-diabetic subjects (p<0.01), while showed tendency of higher risk in diabetic subjects (p=0.10). The annual change of E/e’ was signicantly and independently associated with both diabetes (β=0.278, p<0.01) and AHI (β=0.138, p<0.01). Finally, ROC analyses revealed that addition of AHI and diabetes to classical risk factors best predicted individuals with fast progression of diastolic dysfunction (annual change of E/e’ >1.0) with an AUC of 0.81. [Conclusions] In is an important predictor for progression of LV diastolic dysfunction. Its association is partly confounded, but still independent of the presence of diabetes. lipoprotein; glomerular ltration


Introduction
Diabetic cardiomyopathy is de ned by abnormal myocardial structure and performance in the absence of coronary artery disease, hypertensive cardiomyopathy, and valvular heart disease, in individuals with diabetes mellitus [1]. In early stages, diabetic cardiomyopathy exhibits subclinical structural and functional abnormalities, including left ventricular (LV) hypertrophy and brosis, and these pathophysiological changes are associated with subclinical diastolic dysfunction that often progresses to heart failure (HF) with preserved ejection fraction (HFpEF) [2,3].
In addition to hyperglycemic burden, the risk of HFpEF has been shown to sharply increase in the presence of other risk factors [4]. Obstructive sleep apnea, a frequent comorbid condition noted in diabetic patients [5], is commonly observed in association with heart failure and may be attributable to worsening of the disease [6]. Sleep problems including sleep apnea have also been shown to be closely associated with increased incidence of HFpEF [7]. Thus, mutual interactions of diabetes and sleep apnea may contribute to progression of diabetes cardiomyopathy, particularly diastolic dysfunction. However, integrated predictors including these speci c factors have yet to be investigated, which is important for demonstrating progression of left ventricular diastolic dysfunction in pre-heart failure patients.
In this prospective cohort study, we examined the impact of sleep apnea and diabetes as predictors of progression of cardiac diastolic dysfunction in pre-heart failure patients registered in the Hyogo Sleep Cardio-Autonomic Atherosclerosis (HSCAA) study.

DESIGN AND STUDY PARTICIPANTS
The HSCAA study is an ongoing single center cohort performed for a review of cardio-metabolic risk factors, including quantitatively measured sleep parameters, for elucidation of the clinical implications of cardiovascular and metabolic diseases [8]. From October 2010 to December 2018, 976 patients with at least 1 cardiovascular risk factor, including obesity, hypertension, dyslipidemia, diabetes mellitus, and chronic kidney diseases , were enrolled in that cohort study. Patients (n=240) with ischemic heart disease, moderate to severe valvular heart disease, hypertrophic cardiomyopathy, atrial brillation, HF, or cardiac diastolic dysfunction [transmitral early in ow velocity/early diastolic tissue velocity (E/e' >14)], assessed according to the recommendations of the American Society of Echocardiography and European Association of Cardiovascular Imaging [9] were excluded. After excluding another 131 with missing baseline data for cardiac ultrasonography and 88 with a follow-up period <12 months, 517 of those patients were enrolled in the present study ( Figure 1). A large majority of the participants (n=462) underwent apnomonitor monitoring for a review of sleep apnea. The HSCAA study was approved by an appropriate institutional ethical committee (approval No. 2351) and informed written consent was obtained from each participant.

ASSESSMENT OF CLASSICAL CARDIOVASCULAR RISK FACTORS
Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters (kg/m 2 ). Smoking status was based on self-reported cigarette smoking habit. Blood pressure and pulse rate were measured twice in a sitting position after a deep breath, with the mean values used for analysis. Type 2 diabetes was diagnosed based on results showing fasting plasma glucose ≥126 mg/dl, causal plasma glucose ≥200 mg/dl, or 2-hour plasma glucose ≥200 mg/dl during a 75-g oral glucose tolerance test, or previous therapy for diabetes [10]. Hypertension was de ned as systolic BP ≥140 mmHg, diastolic BP ≥90 mmHg, or treatment for hypertension. Dyslipidemia was de ned based on results showing low density lipoprotein (LDL) cholesterol (≥140 mg/dl), high density lipoprotein (HDL) cholesterol (≤40 mg/dl), elevated triglyceride level (≥150 mg/dl), or treatment for dyslipidemia.
Among these parameters, E/e' >14 was evaluated as LV diastolic dysfunction as recommended [9]. Annual change in E/e' was estimated by using the slope (b) of the linear regression line calculated from at least 3 echocardiographic measurements with the following equation: E/e' = a + b x months.

ASSESSMENT OF SLEEP APNEA
To examine the presence of sleep apnea, we used an apnomonitor device (SAS-2100 ® , Teijin, Tokyo, Japan) to determine apnea hypopnea index (AHI), as previously described [8]. Percutaneous oxygen saturation was recorded using a pulse oximeter. Apnea was de ned as complete cessation of air ow lasting ≥10 seconds, hypopnea as a ≥50% reduction in air ow lasting ≥10 seconds associated with a 4% decrease in oxygen saturation, and AHI as the average number of apnea and hypopnea episodes per hour.

PLASMA BIOCHEMICAL PARAMETERS
Blood samples were obtained in the morning after an overnight fast and then quickly centrifuged to obtain plasma. Whole blood was used for hemoglobin A1c, EDTA-plasma for glucose, and lipids, and serum for other biochemical assays. Glucose was measured using a glucose oxidase method. Insulin was determined by chemiluminesce enzyme immunoassay results and serum creatinine concentration with an enzymatic method. Estimated glomerular ltration rate (eGFR) in each subject was calculated using an equation for Japanese subjects, as follows: eGFR (ml/min/1.73 m 2 ) = 194×age(years) -0.287 ×S-creatinine -1.094 (if female, ×0.739).

STATISTICAL ANALYSES
To compare mean and median values among the groups, Student's t-test and a Mann-Whitney test were used. A chi-square test was used for comparisons for dichotomous variables. To analyze associations between factors, parameters with skewed distribution were natural logarithm-transformed (ln) to normalize skewed distribution. For this study, the primary outcome was LV diastolic dysfunction, which as de ned as E/e' >14, as previously recommended [9]. Outcome rates for the groups were compared using Kaplan-Meier analysis and a log-rank test. Prognostic variables for primary outcome were examined using a Cox proportional hazards regression model. Predictors for annual changes of E/e' were examined using Pearson's correlation coe cient and logistic regression analysis. To compare each logistic model in aggregate, we constructed receiver operating characteristic (ROC) curves and determined the area under the curve (AUC). All statistical analyses were performed using the Statistical Package for Social Sciences software package (PASW Statistics version 18.0). All reported p values are 2tailed and were considered statistically signi cant at <0.05.

Subjects
Among 976 patients registered in the HSCAA cohort study, 517 (342 non-diabetic, 175 diabetic) were prospectively followed, including echocardiographic evaluations repeated every 1-3 years ( Figure 1). The mean follow-up period was 34.7 ± 18.0 (standard deviation) months. Table 1 summarizes basal clinical characteristics of the enrolled subjects who were categorized into the diabetic and non-diabetic groups. As compared with the non-diabetic subjects, the diabetic group exhibited higher age, male prevalence, higher body mass index, more current smokers, higher systolic BP, lower LDL-and HDL-cholesterol, higher fasting plasma glucose and HbA1c, and a higher number with comorbidity of dyslipidemia, more likely to use statin and renin-angiotensin system inhibitors. AHI and prevalence of a score >10 were signi cantly higher in the diabetic group ( Table 1).

Predictors of events of cardiac diastolic dysfunction (E/e'>14)
As shown in Table 1, the majority of parameters examined for LV diastolic function (A wave, E/A, e', E/e') were signi cantly worse in the diabetic group. LVEF was identical between the groups, while LVMI tended to be higher in the diabetic than non-diabetic subjects. During the follow-up period, cardiac diastolic dysfunction (E/e'>14) events were more frequently observed in the diabetic subjects (Table 1) and Kaplan-Meier analysis results revealed that the event (E/e' >14)-free survival rate was also signi cantly lower in that group (Figure 2A) with hazards ratio (HR) of 2.21 (1.41-3.47 95% con dence interval). In all subjects, those with sleep apnea (AHI >10) (p<0.01) with HR 2.23 (1.40-3.55) exhibited higher events of cardiac diastolic dysfunction as compared to subjects without that condition (Figure 2A). In the nondiabetic group, subjects with sleep apnea had a signi cantly greater number of diastolic dysfunction events ( Figure 2B), while in the diabetic group, sleep apnea tended to be associated with higher rate of diastolic dysfunction events ( Figure 2C).

Predictors of annual change in E/e'
Baseline E/e' markedly affected events of cardiac diastolic dysfunction (HR 1.36, 95%CI 1.25-1.48), thus we analyzed clinical factors associated with annual change in E/e' based on results of multiple (≥3 times) echocardiographic measurements. In accordance with the results shown in Figure 2, diabetes and AHI were signi cantly and positively associated with annual change in E/e' in all subjects (Table 2A).
Baseline E/e' was strongly and inversely associated with annual E/e' change, suggesting blunted E/e' progression in patients with already high E/e'. Multiple linear regression analyses in all subjects showed that both diabetes and AHI were signi cantly independent and positive predictors of annual progression of E/e' (Table 2B).
To further examine the impact of diabetes and AHI on prediction of diastolic dysfunction progression, we performed multiple logistic regression analyses to predict fast progression of diastolic dysfunction, which was determined by annual change in E/e' >1.0. As shown in Figure 3, 14.4% of all subjects exhibited an annual change in E/e' >1.0, while the percentage was 22.7% in the diabetic and 9.7% in the non-diabetic group. As seen in the upper panel of Figure 4, ROC analysis revealed that addition of diabetes to basal models including age, gender, BMI, hypertension, dyslipidemia and baseline E/e' resulted in a signi cant (p=0.04) increase in AUC from 0.73 to 0.77, whereas addition of AHI (0.75) instead of diabetes to the basal model (0.73) did not signi cantly change AUC. When AHI was further added to basal + diabetes, AUC was improved up to the highest value of 0.81, signi cantly higher than that for the basal model (p<0.01) but not reaching statistically signi cant than that for basal + diabetes (p=0.28).

Discussion
This is the rst study to examine the impact of sleep apnea as well as diabetes on progression of cardiac diastolic dysfunction in patients without heart failure. Patients with asymptomatic HF based on cardiac ultrasonography ndings together with the index of E/e' >14 were carefully excluded from analyses. We found that in addition to the presence of diabetes, sleep apnea was signi cantly associated with worsening of LV diastolic function. Moreover, addition of sleep apnea index to diabetes resulted in the best prediction of subjects with rapid progression of diastolic dysfunction.
Sleep apnea as predictor of progression of cardiac diastolic dysfunction Although sleep problems including obstructive sleep apnea have been demonstrated to be closely associated with increased incidence of HFpEF [7], reports showing an association of sleep apnea with LV diastolic function, such as E/e', in patients without HF are limited. In 206 patients with AHI ≥ 5/h and without cardiac disease, hypertension, or diabetes, AHI was shown to be associated with higher E/e' [12]. Also, in 78 patients with obstructive sleep apnea without any comorbidities, disease severity tended to be associated with both LV diastolic dysfunction and abnormalities in regional systolic function with myocardial deformation changes, in spite of normal LV ejection fraction [13]. In a recent report of 1506 adults in the Hispanic Community Health Study/Study of Latinos ECHO-SOL Ancillary Study, greater AHI was associated cross-sectionally with subclinical markers of LV diastolic dysfunction (E/e') and LVMI [14].
To the best of our knowledge, no longitudinal studies that examined the predictive value of sleep apnea on progression of cardiac diastolic dysfunction are available. In our study, sleep apnea was conclusively shown to be a signi cant predictor of progression of LV diastolic dysfunction based on multiple results; (1) patients with AHI >10 exhibited signi cantly greater numbers of diastolic dysfunction events (E/e' >14), (2) AHI value was signi cantly and independently associated with annual progression of E/e', and (3) addition of AHI to classical risk factors including diabetes in ROC analyses best predicted an individual with annual worsening of E/e' >1.0. Interestingly, AHI was not signi cantly associated with annual changes in LVMI in all, non-diabetic and diabetic subjects (data not shown), implicating discrete underlying mechanisms involved in the progression of LV hypertrophy and diastolic dysfunction. In addition to diabetes, we consider that sleep apnea is a promising potential predictor for progression of LV diastolic dysfunction.
Diabetes and cardiac diastolic dysfunction LV diastolic dysfunction is an important cause of HFpEF in individuals with diabetes and epidemiologic studies have demonstrated the presence of diastolic dysfunction in type 2 diabetes patients [15,16]. In our recent study, diabetic patients exhibited higher E/e' as compared to those with either normal or impaired glucose tolerance [11]. Among the available parameters of LV diastolic function, E/e′ determined with a non-invasive approach has long been used to assess LV lling pressure in various patient populations [17]. In type 2 diabetes patients, E/e' has been shown to provide independent and incremental prognostic information for predicting cardiovascular morbidity and mortality [18]. Thus, elevated E/e' can be a useful parameter of early diastolic impairment in diabetic patients, potentially foretelling diabetic cardiomyopathy. Longitudinal studies of diabetic patients are needed to better understand and predict this subtle slow onset of diabetic cardiomyopathy, though such evaluations of changes in continuous parameters of LV diastolic function over time in patients without clinically diagnosed HF are scant. Previously, the presence of diabetes was shown to be associated with changes in several cardiac parameters over an 8-year period, including LVEF by -2.98%, LV mass index by +5.86 g, and E/e' by +1.64 [19]. In that study, LV systolic function and LV mass deteriorated similarly along with E/e' during the 8-year follow-up period. In the present study, frequent measurements of cardiac function showed that diabetic subjects exhibited signi cantly greater annual changes in E/e' than non-diabetic subjects, whereas LVEF did not show obvious changes. Also, baseline LVMI was not signi cantly different between the diabetes and non-diabetes groups. On the other hand, annual changes in LVMI were signi cantly greater in our diabetic subjects, although these changes were not as marked as the changes in E/e' (data not shown). Moreover, baseline LVMI was not signi cantly associated with annual change in E/e' in all subjects (r=0.20, P=0.73), suggesting that LV hypertrophy may not be essential for the progression of E/e'. Future kinetics analyses to compare the progression of E/e' and LVMI in diabetic patients are warranted to clarify the pathological sequence of diabetic cardiomyopathy.
Potential mediators of the association between diabetes and cardiac diastolic dysfunction have yet to be well clari ed. A previous study showed that even in a population with normal glucose tolerance, a parameter of diastolic dysfunction was inversely associated with HbA1c [20]. Although variations of HbA1c may be serve as predictors of changes in E/e' [21], in the present cohort, HbA1c was not a major predictor of annual change in E/e'. In a prior cross-sectional study, glycemic variability, determined by continuous glucose monitoring, was also shown to be associated with left ventricular diastolic function in type 2 diabetes mellitus subjects [22]. Population-based cohort study found that the presence of hyperinsulinemia at the baseline predicted worsening of LV diastolic dysfunction [23]. In the present study, plasma insulin was not associated with annual change in E/e' in all subjects (r=0040, p=0.53), or after dividing them into diabetic (r=0.184, p=0.11) and non-diabetic (r=0.001, p=0.99) groups. As previously shown by our groups [24,25], visceral fat adiposity, or an adipocytokine such as leptin may be attributed to diastolic dysfunction in diabetes. Pericardial fat accumulation may also contribute to cardiac diastolic dysfunction in diabetes [26]. Another potential mediator includes presence of diabetic microangiopathy, E/e' shown to be worsened in groups classi ed based on albuminuria severity [27]. However, in our cohort, the presence of diabetic microangiopathy regardless of type (retinopathy, neuropathy, nephropathy) was not signi cantly associated with annual changes in E/e' (data not shown). Anti-diabetic agents such as the sodium glucose cotransporter type 2 (SGLT2) inhibitors may potentially affect the progression of LV diastolic dysfunction [28], which were recently shown to reduce worsening heart failure or death from cardiovascular causes in patients with heart failure [29]. Less than 5% of the present cohort was treated with SGLT2 inhibitors during registration period. Future surveys regarding antidiabetic reagents given during follow-up may unveil their potential associations with progression of LV diastolic dysfunction.

Study Limitations
This study has several limitations. First, the results of the sub-group analyses of diabetes and nondiabetes must be interpreted carefully, due to the number of subjects, potentially resulting in statistical type 1 and 2 errors. Second, not all items used as diagnostic criteria for cardiac diastolic dysfunction could be evaluated, particularly atrial volume. In several studies, E/e′ has been found to lack accuracy for re ecting LV lling pressure in patients with preserved ejection fraction. Nevertheless, we consider that our results provide important initial information to unveil the pathophysiological mechanisms related to HFpEF and diabetic cardiomyopathy.

Conclusions
In patients without heart disease, sleep apnea is an important predictor for progression of LV diastolic dysfunction. Its association is partly confounded, but still independent of the presence of diabetes.     Figure 1 Study participants.  Distribution of subjects (all, non-diabetic, and diabetic subjects) according to annual changes in E/e'.