We performed a retrospective cohort study using electronic medical records (EMRs). Eligible subjects were patients with type 2 diabetes mellitus who had undergone cardiovascular autonomic reflex tests (CARTs) in the outpatient clinic of Seoul St. Mary's Hospital between October 2008 and September 2011. Patients were included if they were between 20 and 75 years of age. Patients with prediabetes, gestational diabetes, type 1 diabetes mellitus, latent autoimmune diabetes in adults, history of diabetic ketoacidosis, or a history of admission for poor glycemic status within the 6 months after the CARTs were performed were excluded. Subjects with an additional health condition that could influence their glycemic variability were also excluded, such as the presence of a malignancy except for early stage papillary thyroid cancer; abnormal liver function (aspartate aminotransferase or alanine aminotransferase higher than three times the upper limit of normal) or liver cirrhosis; estimated glomerular filtration rate (eGFR) calculated by the Modification of Diet in Renal Disease (MDRD) formula of under 45 mL/min/1.73 m²; arrhythmia; autoimmune disease; hematologic disease; and use of steroid therapy except for the stable maintenance of physiologic dose (equivalent dose of prednisolone ≤7.5 mg/day). Patients with proliferative diabetic retinopathy and a history of myocardial infarction or stroke within 6 months could not undergo CARTs due to the Valsalva maneuver. To assess long-term glucose variability, those patients with at least 4 years of follow-up after the CARTs and who had at least 6 HbA1c data points were selected among the eligible subjects for inclusion in this study. Finally, a total of 681 subjects were included in the analysis. This study was approved by the Institutional Review Board (IRB) of Seoul St. Mary's Hospital (KC17RESI0131). The written informed consent from the participants was waived by the IRB as only de-identified data were accessed and analyzed.

Anthropometric data such as height, weight, waist circumference (WC), hip circumference (HC), and blood pressure (BP) were measured by an experienced nurse while wearing light clothing without shoes. Body mass index (BMI) was calculated using the participant's height and weight (kg/m²). The waist-hip ratio (WHR) was calculated as WC divided by HC. BP was measured by the oscillometric method using an appropriate cuff. Diabetic retinopathy was evaluated using fundus photography or ophthalmologic outpatient clinic records. Other clinical data (age, sex, duration of diabetes, insulin therapy, use of lipid-lowering agents and anti-hypertensive agents, smoking history, and laboratory data) were retrieved from the EMR. All of the HbA1c values after the date of the CARTs were collected. HbA1c was measured by high-performance liquid chromatography using DCCT-aligned methods (Tosoh-G8; Tosoh, Tokyo, Japan).

The CARTs were performed at the diabetes complication test clinic by one skilled nurse examiner. The confounding factors that could influence the CARTs were avoided [3]. The patients were advised to fast for 8 hours before the CARTs and to avoid caffeine, tobacco, alcohol, insulin, β-blockers, antihistamines, and antidepressants for 12 hours before the CARTs. The patients were also requested to avoid strenuous physical exercise within 24 hours before the CARTs.

CAN was assessed by five standard CARTs using the Ewing method [19]. The three tests were for parasympathetic function: heart rate responses to deep breathing (maximum–minimum heart rate), postural change from lying to standing (30:15 ratio), and the Valsalva maneuver. These heart rate responses were assessed automatically by the continuous electrocardiogram recoding using the DICAN (Medicore Co. LTD., Seoul, Korea). The two tests for sympathetic function were BP responses to standing up and sustained handgrip. Based on the reference value and scoring system for the classification of CAN described by Bellavere et al. [20] and Boulton et al. [21], each test was scored as normal=0, borderline=1, or abnormal=2 (Supplementary Table 1). Patients' severity of CAN was assessed by the total CARTs score, which was the sum of scores for each of the five CARTs (minimum 0, maximum 10). The scores 0 and 1 were classified as normal autonomic function, scores from 2 to 4 as early CAN, and 5 and over as the severe CAN group. As a sensitivity analysis, we also analyzed the results only using parasympathetic function tests for the diagnosis and staging of CAN [3]. Using this criteria, CAN was defined as the presence of at least one abnormal result in the parasympathetic tests. The CAN staging was defined as follows: (1) one abnormal parasympathetic test result defined as early CAN or (2) at least two abnormal parasympathetic function tests defined as definite CAN.

HbA1c variability was measured as the standard deviation (SD) of serial HbA1c measurements (HbA1c-SD), the coefficient of variation of HbA1c (HbA1c-CV) to correct for the mean of the serial HbA1c measurements, and the adjusted SD of the serial HbA1c measurements (adj-HbA1c-SD) to adjust for the number of HbA1c assessments [9]. Indices of HbA1c variability were calculated as follows:

x=mean of serially measured HbA1c,

n=number of HbA1c measurements

Since there is no standard cutoff value for HbA1c variability indices, we categorized the subjects into two groups using the median value of each index: a low HbA1c variability group (HbA1c-SD <0.475%, HbA1c-CV <6.913%, adj-HbA1c-SD <0.463%) and a high HbA1c variability group (HbA1c-SD ≥0.475%, HbA1c-CV ≥6.913%, adj-HbA1c-SD ≥0.463%).

Continuous variables are presented as the mean±SD, and categorical variables are presented as number (%). Differences between groups of continuous variables were evaluated with a t-test or the Mann-Whitney U test. Differences between three groups were evaluated with a one-way analysis of variance and post hoc analysis, and P for trend was evaluated with a Jonckheere-Terpstra test. Categorical variables were analyzed with a chi-square test or Fisher's exact test. Multivariable logistic regression analysis was performed to test the association between the CAN and HbA1c variability indices and to compute the odds ratio (OR) (95% confidence interval [CI]) after adjusting for potential confounders. Model 1 was adjusted for age, sex, and BMI. Model 2 was further adjusted for duration of diabetes and mean of serial HbA1c. Model 3 was further adjusted for heart rate, eGFR, pre-existing CAD (imaging evidence with multi-detector coronary computed tomography or coronary angiography: ≥50% stenosis of at least one of the coronary artery, or history of myocardial infarction), presence of diabetic retinopathy, insulin treatment, and use of hypertension medication. The analyses were repeated entering HbA1c-CV or adj-HbA1c-SD instead of HbA1c-SD as a measure of HbA1c variability. Subgroup analysis was performed after categorizing the subjects according to the mean serial HbA1c of 7% and the duration of diabetes of 10 years. The statistical analysis was performed using SPSS version 21.0 (IBM Co., Armonk, NY, USA). All P values were two-sided, and a P value <0.05 was considered significant.

The median (interquartile range [IQR]) follow-up period of the study subjects was 5.5 years (IQR, 4.9 to 6.0 years). Of the 681 total subjects, 294 (43.2%) showed normal CARTs, 318 (46.7%) early CAN, and 69 (10.1%) showed severe CAN at baseline. Table 1 shows the baseline characteristics of the subjects according to CAN severity. As the severity of CAN increased from normal to severe, the subjects were older, more likely to be female, to have a longer duration of diabetes, a higher systolic BP, a lower eGFR, a higher high density lipoprotein cholesterol level, a higher prevalence of diabetic retinopathy, and they used insulin and antiplatelet agents more frequently. There were no differences in the follow-up period and the number of HbA1c measurements among the three groups. The mean of the serial HbA1c values was higher in proportion to the severity of the baseline CAN.

The comparison of baseline characteristics between lower and higher HbA1c-SD groups is shown in Supplementary Table 2. The subjects found to have higher HbA1c variability (SD) during the follow-up period were younger; had higher BMI, WC, WHR, systolic and diastolic BP; had faster heart rates; had higher fasting glucose values, baseline HbA1c, and triglyceride levels; had a higher prevalence of diabetic retinopathy, CAN and CAD; and more frequently used insulin at baseline compared with those subjects with lower HbA1c-SD. Although there was no difference in the follow-up period between the two groups, the higher HbA1c-SD group had more HbA1c measurements during the follow-up. In addition, subjects with higher HbA1c-SD had higher mean values of serial HbA1c, HbA1c-CV, and adj-HbA1c-SD compared with those subjects with lower HbA1c-SD. These findings were similar when the subjects were categorized according to HbA1c-CV or adj-HbA1c-SD (data not shown).

Fig. 1 shows three HbA1c variability indices during the follow-up period according to baseline CAN severity. All the HbA1c variability indices showed positive correlations with the severity of baseline CAN (Fig. 1A); (1) HbA1c-SD: normal, 0.514%±0.287%; early CAN, 0.582%±0.350%; severe CAN, 0.687%±0.467%; P for trend=0.001; (2) HbA1c-CV: normal, 7.228%±3.642%; early CAN, 8.077%±4.533%; severe CAN, 9.053%±5.044%; P for trend=0.001; and (3) adj-HbA1c-SD: normal, 0.499%±0.280%; early CAN, 0.565%±0.339%; severe CAN 0.668%±0.456%; P for trend <0.001. Furthermore, the percentage of the higher HbA1c variability group increased according to the baseline CAN severity in a dose-dependent manner (Fig. 1B).

To identify risk factors associated with future HbA1c variability, multivariable logistic regression analysis was performed. Compared to subjects in the normal group, the subjects with early and severe CAN were significantly associated with developing higher HbA1c variability (SD) during the follow-up period (Table 2). The ORs (95% CI) of early CAN (OR, 1.65; 95% CI, 1.12 to 2.43) and severe CAN (OR, 2.86; 95% CI, 1.47 to 5.56) were significantly increased, even after adjusting for multiple covariates. In addition to baseline CAN severity, younger age, higher BMI, shorter duration of diabetes, higher mean value of serial HbA1c, faster heart rate, and the presence of diabetic retinopathy and pre-existing CAD were significantly associated with higher HbA1c variability. These results were largely consistent when HbA1c-CV or adj-HbA1c-SD were used instead of HbA1c-SD (Supplementary Table 3). However, in the analysis using HbA1c-CV, BMI, duration of diabetes, and early CAN lost statistical significance in models 2 and 3.

We also performed a sensitivity analysis using only three parasympathetic CARTs for the diagnosis and staging of CAN. When CAN was defined as the presence of at least one abnormal result of the parasympathetic tests (Supplementary Table 4), 378 subjects (55.5%) had CAN. The OR (95% CI) for developing higher HbA1c-SD was significantly increased, even after adjusting for multiple covariates (OR, 1.52; 95% CI, 1.05 to 2.18). When the CAN was staged according to the numbers of abnormal parasympathetic tests (Supplementary Table 5), 303 (44.5%) showed normal CARTs, 240 (35.2%) early CAN, and 138 (20.3%) showed definite CAN at baseline. The subjects with definite CAN showed a significantly increased OR (95% CI) for developing a higher HbA1c-SD, even after adjusting for multiple covariates (OR, 1.76; 95% CI, 1.07 to 2.88). The results were similar when using adj-HbA1c-SD, although significance was lost with analysis using HbA1c-CV. The other risk factors that were significantly associated with higher HbA1c variability were similar with the original analysis.

Table 3 shows the subgroup analysis according to the mean value of serial HbA1c during the observation period (<7% vs. ≥7%) and the duration of diabetes (≤10 years vs. >10 years) at baseline. In the analysis of subjects with mean HbA1c <7%, a stepwise increase in the OR (95% CI) for higher HbA1c-SD was noted according to the severity of CAN (model 3: early CAN [OR, 1.72; 95% CI, 1.02 to 2.90]; severe CAN [OR, 3.34; 95% CI, 1.41 to 7.91]). However, this association was insignificant in the subject of mean HbA1c ≥7%. In the analysis of subjects with diabetes duration >10 years, severe CAN showed a significant association with higher HbA1c-SD (OR, 4.75; 95% CI, 1.71 to 13.20). In subjects with diabetes duration ≤10 years, only early CAN was significantly associated with higher HbA1c-SD. These results were similar when using HbA1c-CV or adj-HbA1c-SD instead of HbA1c-SD (Supplementary Table 6), except that the association between early CAN and higher HbA1c-CV was insignificant in the subgroups of mean HbA1c <7% and diabetes duration ≤10 years.