Cardiac Autonomic Dysfunction and Risk of Silent Myocardial Infarction Among Adults With Type 2 Diabetes

Study Design

This report followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline for observational studies. ¹⁵ We conducted a prospective cohort analysis of the ACCORD study. The design and methods of the ACCORD study have been reported elsewhere. ¹⁶ The ACCORD study was a randomized 2‐by‐2 factorial clinical trial that enrolled 10 251 adults with type 2 diabetes in the United States and Canada between January 2001 and October 2005. These participants were randomly assigned to either an intensive glucose‐lowering arm aiming for of a glycated hemoglobin (HbA_1C) <6% or a standard treatment arm with an HbA_1C goal of 7.0% to 7.9%, as well as specific blood pressure (BP) and lipid intervention arms. For the current analysis, we excluded participants who at baseline had established atherosclerotic cardiovascular disease (defined as prior myocardial infarction [MI], angina, stroke, history of coronary revascularization, carotid or peripheral revascularization; n=3609), artificial pacemaker (n=23), an atrioventricular conduction defect (n=232), atrial fibrillation/flutter (n=73), premature beats and other arrhythmias (n=456), missing ECG (n=650), or poor‐quality ECG (n=236). We further excluded participants with clinically recognized MI during follow‐up (n=130). After the relevant exclusions, our final simple included 4842 participants. The process of selecting study participants, including the various reasons for exclusion, is shown in Figure S1. A comparison of the baseline characteristics of ACCORD participants who were included to those excluded from the final sample is shown Table S1. The study was approved by an institutional review committee. Informed consent was obtained from all participants.

Assessment of CAN

We assessed CAN at baseline using heart rate variability (HRV) metrics derived from a digitalized 12‐lead ECG. The digital ECG acquisition and procession, including among other signal processing, filtering, sampling frequency, and management of ectopic beats, were performed using standardized procedures. ¹⁷

ECGs were recorded over 10 consecutive seconds at 10 mm/mV calibration and a speed of 25 mm/s (GE MAC 1200 electrocardiograph system; GE, Milwaukee, WI) among fasting participants (who also abstained from smoking and drinking alcohol) lying flat. ⁷ ECGs were electronically transmitted to the reading center where they were analyzed and reviewed for technical quality before they were automatically processed using GE 12‐SL Marquette Version 2001 (GE). ECG reading was performed centrally at the Epidemiological Cardiology Research Center, Wake Forest School of Medicine, Winston‐Salem, North Carolina. The ECG recordings were used to derive 2 HRV time‐domain indices: the standard deviation of all normal‐to‐normal R‐R intervals (SDNN) and the root mean square of successive differences between normal‐to‐normal R‐R intervals (rMSSD).

We defined low HRV using cutoff values derived from healthy US adult populations: low SDNN defined as SDNN <8.2 ms; low rMSSD defined as rMSSD <8.0 ms. ¹⁸ We used a composite measure of CAN, which was derived as SDNN and rMSSD both being below the fifth percentile of the general population distribution (SDNN <8.2 ms and rMSSD <8.0 ms). ¹⁸ , ¹⁹

Ascertainment of Silent MI

The participants were prospectively followed from the baseline visit until the occurrence of SMI, death, or study end (in June 2009). Baseline and follow‐up resting ECGs were obtained in all ACCORD sites. ¹⁶ Incident SMI cases were ascertained from 12‐lead ECGs obtained at the biennial follow‐up visits. We defined incident SMI as the presence of a major Q‐wave abnormality or minor Q/QS waves in the setting of major ST‐T abnormalities in the absence of history of clinical cardiovascular disease, which is consistent with prior studies. ¹³ We excluded individuals with clinically recognizable MI from the analyses. The events were adjudicated by an expert committee in ACCORD. ¹⁶ , ²⁰

Covariates

The covariates, selected a priori based on their relation with CAN or SMI, included age, sex, race and ethnicity, treatment arm, history of retinopathy, duration of diabetes, cigarette smoking, alcohol consumption, body mass index, BP, use of BP‐lowering medications, use of medications that can affect HRV (calcium channel blockers, β‐blockers, digitalis, and other antiarrhythmics), total cholesterol, high‐density lipoprotein cholesterol, low‐density lipoprotein cholesterol, HbA_1C, and serum creatinine, ¹⁶ and estimated glomerular filtration rate was calculated based on the Modification of Diet in Renal Disease formula. ²¹ Cholesterol fractions were measured at the ACCORD central laboratory using standard biochemical methods. ¹⁶ Serum creatinine was measured via enzymatic methods on a Roche Double Modular P Analytics automated analyzer.

Statistical Analysis

We compared the baseline characteristics of study participants by CAN status using the t test or Kruskal‐Wallis test for continuous variables depending on their distribution and the χ² test for categorical variables.

We calculated incidence rates and their associated 95% CIs by dividing the number of SMIs by the total at‐risk person‐years, with the person‐years estimated from baseline through the earliest of SMI, date of death, or study termination (June 2009). We used multivariable Cox proportional hazards regression models to generate hazard ratio (HR) and 95% CI for SMI. Model 1 adjusted for age, sex, race and ethnicity, and treatment arm. Model 2 included variables in Model 1 plus duration of diabetes, glycated hemoglobin, cigarette smoking, alcohol intake, body mass index, systolic BP, use of BP‐lowering medications, estimated glomerular filtration rate, and total/high‐density lipoprotein cholesterol ratio. Model 3 included the Model 2 variables with further adjustment for use of medications affecting HRV (β‐blockers, calcium channel blockers, digitalis, and antiarrhythmics). Model 4 included variables in Model 3 plus history of retinopathy at baseline.

In supplementary analyses, we examined the association of CAN status and each HRV index with the incidence of clinically recognized (overt) MI and overall MI (including both clinically recognized and silent MI events).

We calculated the sensitivity, specificity, and positive and negative predictive values of CAN for the detection of SMI.

All analyses were performed using Stata 14.2 (StataCorp, College Station, TX). A 2‐sided P value of <0.05 was deemed statistically significant.

Baseline Characteristics of the Study Population

A total of 4842 participants were included in our investigation (mean age, 62.5 [SD, 5.7] years; 46.6% women; 60.2% White). The prevalence of CAN at baseline in the analytical sample was 18.6% (903/4842). Table 1 displays the characteristics of included participants according to their CAN status at baseline. The participants with low HRV had higher body mass index, HbA_1C, duration of diabetes, lower high‐density lipoprotein cholesterol or estimated glomerular filtration rate, and were more likely to be current smokers, insulin users, or to have a history of retinopathy (Table 1).

CAN and Incidence of SMI

Over a median follow‐up of 4.9 years, 73 participants experienced an SMI (incidence rate 3.1/1000 person‐years [95% CI, 2.5–3.9]). In terms of absolute risk, the crude incidence rate of SMI was >1.5‐fold higher among those with CAN as compared with those without CAN (Table 2).

After multivariable adjustment, CAN was associated with an increased risk of SMI (HR, 1.91 [95% CI, 1.14–3.18], Model 2, Table 2). The magnitude and significance of this association remained unchanged after further adjustments for medications affecting HRV (HR, 1.92 [95% CI, 1.15–3.20], Model 3; Table 2) and history of retinopathy at baseline (HR, 1.91 [95% CI, 1.14–3.20], Model 4; Table 2).

Examination of each of the HRV indices in isolation (Table 3) showed that low HRV was significantly associated with a higher risk of SMI. A 1‐SD decrease in SDNN was associated with a 1.29‐fold greater risk of SMI (HR, 1.29 [95% CI, 1.02–1.65]). Participants with low SDNN had a 1.67‐fold higher risk of SMI compared with those with normal SDNN (HR, 1.67 [95% CI, 1.02–2.72]). Low rMSSD was associated with a higher, but nonstatistically significant, risk of SMI (HR, 1.56 [95% CI, 0.94–2.58]; Table 3). The sensitivity, specificity, and positive and negative predictive values of CAN for the detection of SMI were 30.1%, 81.5%, 2.4%, and 98.7%, respectively.

Supplementary Analyses: Association of CAN and HRV Indices With Clinically Recognized MI and Overall MI

The individuals who experienced SMI did not substantially differ from those who experienced clinically recognized MI (Table S2). After full adjustment, CAN was associated with increased risks of clinically recognized MI (HR, 1.60 [95% CI, 1.08–2.36], Model 4; Table S3) and of overall MI (including both overt and silent MI events; HR, 1.70 [95% CI, 1.24–2.31], Model 4; Table S4). Similar positive associations were observed for relation of each HRV index (SDNN and rMSSD) with clinically recognized MI on one hand (Table S5) and overall MI on the other hand (Table S6).