Visit-to-visit HbA1c variability is associated with in-stent restenosis in patients with type 2 diabetes after percutaneous coronary intervention

Background Patients with type 2 diabetes are under substantially higher risk of in-stent restenosis (ISR) after coronary stent implantation. We sought to investigate whether visit-to-visit HbA1c variability is a potential predictor of ISR in diabetic patients after stent implantation. Methods We consecutively enrolled type 2 diabetic patients who underwent successful elective percutaneous coronary intervention and performed follow-up coronary angiography after around 12 months. The incidence of ISR and its relationship with visit-to-visit HbA1c variability, expressed as coefficient of variation (CV), standard deviation (SD) and variability independent of the mean (VIM), were studied. Multivariable Cox proportional hazards models were constructed to analyze the predictive value of HbA1c variability for ISR. Results From September 2014 to July 2018 in Ruijin Hospital, a total of 420 diabetic patients (688 lesions) after stent implantation were included in the final analysis. During a mean follow-up of 12.8 ± 1.3 months, the incidence of ISR was 8.6%, which was significantly increased in patients with higher CV of HbA1c (P = 0.001). The mean diameter stenosis (DS), net luminal loss and net luminal gain were 22.9 ± 16.8%, 0.42 ± 0.88 mm and 1.66 ± 0.83 mm, respectively. Greater DS was observed in subjects with higher tertiles of CV of HbA1c (P < 0.001), and this trend was more prominent in patients with optimal glycemic control (HbA1c ≤ 7%) in the baseline. In multivariate analysis, HbA1c variability was independently associated with incidence of ISR after adjustment for traditional risk factors and mean HbA1c (HR: 3.00 [95% CI 1.14–7.92] for highest vs. lowest tertile). Inclusion of CV of HbA1c led to a better risk stratification accuracy. Assessing HbA1c variability by SD or VIM yielded similar findings. Conclusions This study suggests that visit-to-visit HbA1c variability is an independent predictor of incidence of ISR in patients with type 2 diabetes after stent implantation. Trial registration NCT02089360: NCT

Background Patients with type 2 diabetes are under substantially increased risk of rapid-progressive and diffuse atherosclerosis [1,2], myocardial infarction [3] and poor coronary collateralization [4]. After percutaneous coronary intervention (PCI) and deployment of stents, diabetic patients are predisposed to exaggerated neointimal hyperplasia and the development of in-stent restenosis (ISR) [5,6]. In the era of drug-eluting stents (DES), although restenosis rate has significantly declined, diabetic patients still suffer from higher risk of ISR than non-diabetic patients [7,8]. The prognosis of diabetic patients after DES implantation is also more dismal than that of non-diabetic patients, with increased rates of cardiac death, myocardial infarction, target lesion failure and target vessel revascularization [9].
Hyperglycemia is a critical contributory factor to the development of restenosis [10], partly attributed to endothelial dysfunction [11], excessive production of reactive oxygen species [12] and formation of advanced glycation end-production [13]. Pre-procedural optimal glycemic control was shown to be associated with lower rate of stent failure in comparison with suboptimal control patients [10]. A retrospective study analyzing glycemic control based on sequential HbA 1c measurements from preprocedural to 6-month follow-up also suggested that sustained glycemic control is associated with better clinical outcomes in diabetic patients after PCI [14].
On the other hand, emerging evidence suggests that glycemic variability confers an additional risk to diabetic complications, which is predicted by mean glucose levels alone and may, to some extent, underlie the pathogenesis of micro-and macro-vascular diabetic complications. In the short-term, glycemic variability assessed by continuous glucose monitoring or serial glucose levels during hospitalization is associated with poor prognosis in patients with coronary artery disease (CAD) [15][16][17][18]. In the long-term, a retrospective study analyzing data from Diabetes Control and Complications Trial (DCCT) demonstrated that HbA 1c variability adds to mean HbA 1c in predicting the development of retinopathy and nephropathy in type 1 diabetes [19]. A prospective study of cohort of type 2 diabetes from Renal Insufficiency and Cardiovascular Events (RIACE) revealed that HbA 1c variability affects chronic kidney disease more than average HbA 1c [20]. Recent studies further showed that long-term glycemic variability, either estimated by serial measurements of fasting plasma glucose or by HbA 1c , is a strong predictor of all-cause mortality and cardiovascular events [21][22][23]. However, the relationship between glycemic variability and ISR is still unclear. Therefore, in the present study, we sought to investigate whether visit-tovisit HbA 1c variability is a potential predictor of ISR in patients with type 2 diabetes after DES implantation.

Study population
A total of 920 consecutive patients with type 2 diabetes and CAD were screened, who received follow-up coronary angiography ~ 12 months after DES-based PCI of de novo lesions in native coronary arteries between September 2014 and July 2018 from the database of Advanced Glycation Endproducts and Development of CAD Program (AGENDA) in Ruijin Hospital, Shanghai. Patients were referred to coronary angiography for the evaluation of established or suspected CAD due to typical chest pain, positive exercise stress test, or positive myocardial perfusion scan. ISR was defined as recurrence of luminal diameter stenosis (DS) of > 50% within the stent or in the 5-mm proximal or distal segments adjacent to the stent at follow-up angiography.
For the purpose of this study and to avoid confounding serum data, patients who had acute coronary syndrome (n = 86) during initial angiography and PCI, familial hypercholesterolemia (n = 5), malignant tumor (n = 13), or renal failure requiring hemodialysis (n = 8) were excluded. Another 36 subjects with no hematological and biochemical indices at admission were further excluded. All these patients received a quarterly clinical evaluation, routine analyses and HbA 1c measurements. Follow-up coronary angiography was performed after around 12 months and all the enrolled patients were reminded by telephone in advance. During follow-up, 5 patients died and 68 patients were lost to follow-up. For calculation of HbA 1c variability, subjects (n = 279) without at least three HbA 1c measurements during follow-up (≥ 3 months apart) were also excluded. The remaining 420 subjects constituted the study population (Fig. 1). The diagnosis of type 2 diabetes was made according to the criteria of American Diabetes Association [symptoms of diabetes with casual plasma glucose concentration ≥ 200 mg/dL (11.1 mmol/L) or fasting plasma glucose ≥ 126 mg/dL (7.0 mmol/L), 2 h postprandial glucose ≥ 200 mg/dL (11.1 mmol/L) during an oral glucose tolerance test, and currently or previously treated with insulin and/ Keywords: HbA 1c variability, In-stent restenosis, Type 2 diabetes, Percutaneous coronary intervention, Diameter stenosis or oral hypoglycemic agents] [24]. Hypertension was diagnosed according to seventh report of the Joint National Committee on prevention, detection, evaluation, and treatment of high blood pressure (JNC 7).
This study complies with the Declaration of Helsinki. The study protocol was approved by the local hospital ethics committee, and written informed consent was obtained from all participants.

Baseline clinical and biochemical assessments
Blood samples were obtained at the day of angiography in all patients after an overnight fasting and collected in a quiet, air-conditioned room after at least 20 min supine rest. Serum glucose, insulin, blood urea nitrogen, creatinine, uric acid, total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), triglycerides, apolipoprotein A-I and apolipoprotein B were assessed (HITACHI 912 Analyzer, Roche Diagnostics, Germany). The estimated glomerular filtration rate (eGFR) was computed using the Chronic Kidney Disease Epidemiology Collaboration equation [25]. Blood HbA1c concentration was measured using ion-exchange high performance liquid chromatography with Bio-rad Variant Hemoglobin Testing System (Bio-Rad Laboratories, USA). Serum N-terminal pro-B-type natriuretic peptide (NT-proBNP) was determined using a commercially available electrochemiluminescence immunoassay kit (Roche Diagnostics). Serum levels of high sensitive C-reactive protein (hsCRP) were determined by ELISA (Biocheck Laboratories, Toledo, OH, USA). The detailed information about medical history and lifestyles including smoking status was obtained using a standard questionnaire by trained physicians. Body mass index (BMI) was calculated using the formula of weight/height 2 (kilograms per square meter) [26]. Blood pressure was measured on the non-dominant arm in a seated position after a 10-min rest, using an electronic blood pressure monitor (OMRON Model HEM-752 FUZZY' Omron Co., Dalian, China). Three measurements were taken at 1-min intervals, and the average was used for analysis.

Angiographic analysis
Coronary angiography was performed using standard Judkins technique through radial or femoral approach. For each patient, multiple matched angiographic views were obtained after intracoronary administration of nitrate. Quantitative coronary analysis (QCA) of all angiographic data before and after procedure and during follow-up was performed (TERRA, GE, USA) by two experienced interventional cardiologists (FH Ding and XQ Wang), who were unaware of clinical information of the patients. Using the outer diameter of the contrastfilled catheter as the calibration, the minimal lumen diameter (MLD) and reference diameter (RD) in diastole before intervention was determined from multiple projections by interpolated method. Lesion length was measured as the distance (in millimeters) from the proximal to distal shoulder in the projection with the least amount of foreshortening. The lesion was stented using a normal-to-normal technique, usually including 5-mmlong, angiographically normal segments proximal and distal to the lesion. Net luminal loss was defined as the difference between the MLD immediately after the procedure and that measured during follow-up. Net luminal gain was defined as the difference between the MLD before the procedure and that measured during followup. A value of 0 mm was assigned for MLD in the case of total occlusion at baseline. For patients who underwent multi-lesion coronary angioplasty, the most severe restenotic lesion at follow-up was entered into the analysis.

HbA 1c variability determinations
HbA 1c was measured in the baseline and during followup period for at least three times in 3-month interval. Then the mean and variability of HbA 1c were calculated. Three measures of HbA 1c variability were employed for the analysis. Intraindividual variability of HbA 1c was primarily defined as intraindividual coefficient of variation (CV) of HbA 1c across visits. The alternative variability of HbA 1c includes: 1) standard deviation (SD) and 2) the variability independent of the mean (VIM), which is calculated by the equation as previously reported [27]: VIM = 100 × SD/mean β , where β is the regression coefficient based on natural logarithm of SD on natural logarithm of mean of the study population.

Statistical analysis
Continuous variables were presented as median (interquartile range) or mean ± SD, and categorical data were summarized as frequencies (percentages). Normal distribution of continuous variables was evaluated by Shapiro-Wilk test. For normally distributed variables, differences in tertiles of HbA 1c variability and subgroup analysis were performed by one-way or two-way analysis of variance (ANOVA) followed by post hoc t test with Bonferroni correction. For non-normally distributed continuous variables, differences were analyzed by Mann-Whitney U test or Kruskal-Wallis test. Differences in categorical variables were analyzed by χ 2 test. The association between measures of HbA 1c variability and the incidence of ISR was assessed by Cox regression from which hazard ratios (HR) and 95% confidence interval (CI) were calculated. The assumption of proportionality of the Cox model covariates was tested by plotting Schoenfeld residuals. Five models were constructed for each measure of HbA 1c variability and binary angiographic restenosis (DS ≥ 50%) was employed as the dependent variable. In model 1, sex and age were adjusted. In model 2, we further adjusted admission systolic and diastolic blood pressure, BMI, non-HDL-C and eGFR. In model 3, additional adjustment was performed with the post-PCI RD of target vessel, total stented length and medication use including oral hypoglycemic agent (OHA) and insulin. In model 4 and 5, we further adjusted for baseline HbA 1c and the mean HbA 1c level during follow-up, respectively. Net reclassification improvements (NRI) and integrated discrimination improvements (IDI) were analyzed to assess the improvement in clinical utility of the prediction model by considering HbA 1c variability. All statistical analyses were performed using the R statistical package v.3.6.3 (R Project for Statistical Computing, Vienna, Austria). A 2-tailed < 0.05 was considered statistically significant.

Baseline characteristics of the study population
A total of 420 subjects with 688 lesions, with a mean follow-up period of 12.8 ± 1.3 months, were included in the analysis. The male-to-female ratio was 74:26 and the mean age was 64.5 ± 9.0 years. Among these subjects, 73.8% were with hypertension and 77.6% of the subjects were with multivessel disease. The mean HbA 1c during follow-up was 7.4 ± 1.2%, and CV, SD, VIM of HbA 1c during follow-up were 0.061 (IQR 0.038-0.107), 0.402 (IQR 0.252-0.839) and 0.209 (IQR 0.127-0.297), respectively. CV (Pearson's r = 0.325, P < 0.001) and SD (Pearson's r = 0.445, P < 0.001) were correlated to the mean HbA 1c while there was no significant correlation between VIM and the mean HbA 1c level (Pearson's r = 0.070, P = 0.169).
To analyze the effect of HbA 1c variability on ISR, we divided the population based on tertiles of CV of HbA 1c (Table 1). There was no significant difference in age, sex, history of hypertension, admission blood pressure, smoking status and duration of diabetes between the three tertiles. At admission, subjects with the highest tertile of CV of HbA 1c had higher levels of HbA 1c , fasting and 2 h postprandial glucose, but lower 2 h postparandial insulin level than those with the lowest tertile. Fasting insulin level was similar between the three groups. Meanwhile, HDL-C was lower, whereas serum creatine and hsCRP were higher in subjects with the highest tertile. OHA and insulin were more frequently used in subjects with higher CV of HbA 1c .

Angiographic findings
There were no significant differences in the target vessels, stent counts, stented length, angiographic pre-and post-PCI RD, DS and MLD between the three groups ( Table 2). In the overall population, follow-up coronary angiography showed the prevalence of binary angiographic ISR, defined as ≥ 50% DS, was 8.6%. The mean DS was 22.9 ± 16.8%, and the mean net luminal loss and net luminal gain was 0.42 ± 0.88 mm and 1.66 ± 0.83 mm, respectively.
There was a significant increase in DS across tertiles of CV of HbA 1c (Fig. 2a, P = 0.001). Compared with subjects with the lowest tertile, a higher percentage of DS was found in the highest tertile (26.63 ± 19.08 vs. 19.29 ± 14.47%, P < 0.001). Accordingly, net luminal gain (P < 0.001) was step-wisely decreased in subjects with higher HbA 1c variability as grouped by all the three measures (Fig. 2b). Although there was no difference in net luminal loss between tertiles of CV ( Fig. 2c; P = 0.124), it differed significantly between subjects with different tertiles of SD (P = 0.023) or VIM (P = 0.014) of HbA 1c (Additional file 1: Figures S1 and S2). In addition, comparison of HbA 1c variability between subjects with and without ISR also showed significantly higher HbA 1c variability in ISR patients as analyzed by all the three measures (Additional file 1: Figure S3).
The impact of HbA 1c variability on ISR was analyzed across subgroups of sex, age, dichotomized baseline BMI, eGFR and HbA 1c (Fig. 3). Since the rate of binary ISR was relatively low, DS at follow-up angiography was compared between subgroups. We found DS was increased across tertiles of CV of HbA 1c in male but not female subjects. A trend towards higher percentage of DS across the tertiles was more prominent in subjects with higher BMI and poorer renal function, and was similar between two age groups. Interestingly, compared with subjects with higher HbA 1c at the time of PCI (HbA 1c > 7%), those with lower HbA 1c (≤ 7%) appeared to have more severe restenosis when having higher CV of HbA 1c . There was no significant interaction term between tertiles of CV of HbA 1c and these grouping variables, with the solo exception of basal HbA 1c level (P = 0.010). Dividing subjects by tertiles of SD or VIM yielded similar findings with a little variation (Additional file 1: Figures S4 and S5).

Multivariate analysis
Multivariate analysis was performed to analyze the association between the incidence of ISR and different measures of HbA 1c variability ( Inclusion of HbA 1c variability led to better risk stratification accuracy. After entering tertiles of CV of HbA 1c in the model, 29.4% of subjects with ISR were correctly reclassified to a higher risk category and none was reclassified to a lower category. In patients without ISR, 10.8% were correctly reclassified to a lower risk category and 9.4% were reclassified to a higher category (categories of restenosis: < 10%, 10-20%, ≥ 20%). Accordingly, the categorical NRI was 30.76% [(95% CI 14.78-46.74%), P < 0.001], and IDI was 2.81% [(95% CI 0.81-4.82%), P = 0.006].

Discussion
The major findings of the present study are that patients with type 2 diabetes and high post-procedure HbA 1c variability tend to have greater neointimal hyperplasia and increased rate of ISR in comparison with those with low HbA 1c variability. Evaluation of HbA 1c variability by different measures exhibits consistent findings. Consideration of HbA 1c variability leads to better risk stratification accuracy of ISR in patients with type 2 diabetes after stent implantation.

Impact of glycemic level and stability on ISR
Diabetic patients with obstructive and non-obstructive coronary stenosis generally had poor clinical outcomes, owning to diffuse distribution of atherosclerotic lesions, unstable plaques, microvascular dysfunction and higher incidence of in-stent restenosis (ISR) [28,29]. Compelling evidence has demonstrated a substantially increased rate of ISR in diabetic patients after coronary intervention irrespective of the specific treatment modalities including balloon angioplasty, bare-metal stents (BMS) and DES [7,8,30]. However, very few studies analyzed the association of glucose level and stability with the rate of ISR. Corpus et al. found that optimal glucose control (HbA 1c ≤ 7%) before catherization was associated with a ~ 2-fold decrease in rate of target vessel revascularization compared to those with suboptimal glucose control (HbA 1c > 7%) [10]. A single center prospective study showed that diabetic patients with poor glycemic control at time points both pre-and post-PCI had higher risk of major adverse cardiovascular events (MACE) than nondiabetic patients [14]. In contrast, a retrospective study showed that diabetic patients with good glycemic control (HbA 1c ≤ 6.9%) only at the time of PCI, but not at followup, was associated with significantly lower incidence of MACE compared to those with poor glycemic control (HbA 1c > 6.9%; 18.4% vs. 26.2%, P < 0.05) [31]. These studies unanimously suggest that glycemic control at the time of PCI is of importance to prevent subsequent restenosis and adverse cardiovascular outcomes, but with conflicting findings on the effect of post-procedural glycemic control. Actually, glycemic control in these studies was defined according to the cut-off level of HbA 1c at certain time points without consideration of glycemic variability. A substantial proportion of patients in these studies received coronary intervention based on BMS, which does not necessarily respond in the same way as that of DES in the process of restenosis under hyperglycemic conditions.
In the present study, all the enrolled patients received DES-based PCI, which reflects the predominant treatment modality in current clinical practice. In accordance with previous reports, we found diabetic patients with poor glycemic control at the time of PCI (HbA 1c > 7%) had a 1.49-fold higher rate of ISR than those with good glycemic control (HbA 1c ≤ 7%). By grouping patients based on mean HbA 1c during follow-up instead, there was an even higher (2.54fold) increased rate of ISR in subjects with poor versus good glycemic control. Importantly, we for the first time reported that the rate of ISR and angiographic DS were increased across tertiles of HbA 1c variability parameters. There was also a trend towards greater net luminal loss and less net luminal gain in patients with higher variability of HbA 1c . Therefore, previous reports and our findings suggest that both glycemic level and stability are important in the process of ISR after DES implantation in patients with type 2 diabetes. Interestingly, subgroup analysis showed that the impact of HbA 1c variability on DS was more prominent in subjects with good (HbA 1c ≤ 7%) as compared to those with poor glycemic control (HbA 1c > 7%) at the time of PCI. This might be due to the reason that HbA 1c reflects both fasting and postprandial glucose levels. In wellcontrolled diabetic patients (HbA1c < 7.3%), postprandial glucose level is a predominant contributor (70%) to HbA 1c and this contribution decreases progressively with increasing level of HbA 1c [32]. Hence, variability of postprandial glucose might be more important than fasting glucose in the development of ISR and this hypothesis awaits further investigation. Currently, there is no universally accepted "gold standard" to quantify glycemic variability. In this study, we assessed HbA 1c variability by three different measures. In addition to SD, CV and VIM were employed to adjust for mean HbA 1c during follow-up. VIM was calculated based on logarithmic curve fitting to eliminate its correlation with mean HbA 1c , and CV is relatively simple and more feasible in clinical practice. Analysis of HbA 1c variability by all of these three measures yielded similar findings. After adjusting for mean HbA 1c level during follow-up, different measures of HbA 1c variability remained significantly associated with the incidence of ISR. Inclusion of HbA 1c variability led to significantly increased risk prediction accuracy compared to the model that only included conventional risk factors, lesion and procedure characteristics, and mean HbA 1c . These findings support the notion that HbA 1c variability is independent of glycemic level in association with ISR. Actually, previous secondary analyses of data from DCCT [19] and Finnish Diabetic Nephropathy (FinnDiane) Study [33] revealed that HbA 1c variability is an independent predictor of incident microalbuminuria, progression of renal disease and also incident cardiovascular events in patients with type 1 diabetes. A study analyzing 58,832 patients with type 2 diabetes in a large primary care database in England showed that HbA 1c variability was strongly associated with overall mortality and emergency hospitalization and not explained by mean HbA 1c [34]. A single center prospective study found that elevated admission glycemic variability appears even more important than admission glucose in predicting 1-year MACE in patients with acute myocardial infarction [35]. Therefore, although it is hard to tease out the relative effect of HbA 1c variability after controlling for HbA 1c level in the process of ISR, HbA 1c

Possible mechanisms
It is unclear the specific mechanism by which HbA 1c variability affects the development of restenosis in diabetic patients. Based on previous clinical and basic science Fig. 3 The impact of HbA 1c variability on ISR across subgroups. The impact of HbA 1c variability on ISR was analyzed in the overall population (a) and across subgroups of sex (b), age (c), dichotomized baseline BMI (d), dichotomized baseline eGFR (e) and dichotomized baseline HbA 1c (f). ISR in-stent restenosis, BMI body mass index, eGFR estimated glomerular filtration rate, HbA1c glycated hemoglobin A 1c Table 3 Multivariate analysis Model 1, includes adjustment for age and sex; Model 2: additional adjustment for systolic and diastolic blood pressure, body mass index, non-HDL cholesterol and eGFR; Model 3, additional adjustment for the post-PCI reference diameter of target vessel, total stented length and medication use including oral hypoglycemic agent and insulin; Model 4, model 3 with additional adjustment for baseline HbA 1c ; Model 5, model 3 with additional adjustment for the mean HbA 1c during follow-up. * P for trend CI confidence interval, CV coefficient of variation, eGFR estimated glomerular filtration rate, HbA 1c glycated hemoglobin A 1c , HDL high-density lipoprotein, HR hazard ratios, PCI percutaneous coronary intervention, SD standard deviation, VIM variability independent of the mean studies, potential mechanisms include: First, hyperglycemia and glycemic fluctuation directly and indirectly stimulate the production of reactive oxygen species, inflammatory and metabolic cytokines, which are essential players in the development of adverse myocardial and vascular remodeling, and worse clinical outcomes both in patients with or without diagnosed diabetes [36][37][38][39][40]. Second, glycemic variability is strongly correlated with postprandial β-cell dysfunction in type 2 diabetic patients using OHA. Consistently, we found postprandial insulin level was lower and insulin resistance was higher in patients with the highest tertile of CV than those with the lowest tertile [41]. Given that insulin resistance is an established contributory factor in restenosis, the impact of HbA 1c variability on ISR may also be secondary to insulin resistance. Third, dysregulated glucose homeostasis is associated with endothelial dysfunction and higher risk of cardiovascular events [42,43]. Mounting evidence suggests that endothelial dysfunction is an important predictor of restenosis after stent implantation [44,45]. Hence, endothelial dysfunction and delayed reendothelialization may serve as an important underlying mechanism in the development of ISR in conditions of high glycemic variability.

Study limitation
Our findings should be interpreted in the context of following limitations. First, this study is a retrospective analysis based on prospectively collected data, and all the enrolled patients were from a single center. Second, fluctuations in fasting plasma glucose (FPG) and HbA 1c appear to function differentially in the process of diabetic complications [19,46]. Variability of FPG was not analyzed in this study, which may have different features or function in different phases as compared to that of HbA 1c . Moreover, conditions that affect erythrocyte turnover may also affect HbA 1c level. Third, coronary lesions and restenosis were analyzed by QCA. Intravascular imaging techniques such as intravascular ultrasound would provide more accurate assessments. Fourth, this study was not designed to analyze the predictive value of HbA 1c variability for hard endpoints in diabetic patients underwent PCI. Although we found ISR rate was significantly elevated in patients with high variability of HbA 1c , whether these patients suffer higher risk of cardiovascular mortality remains inconclusive.

Conclusions
In conclusion, our findings suggest that greater visitto-visit HbA 1c variability is associated with higher incidence of ISR in patients with type 2 diabetes after stent implantation. Variability of HbA 1c adds to mean level for risk prediction of ISR. Measures targeting both glycemic level and stability may provide favorable effects to reduce the incidence of ISR and improve clinical outcomes in patients with type 2 diabetes after PCI.
Additional file 1: Figure S1 Cumulative frequency curves for diameter stenosis (DS, A), net luminal gain (B) and net luminal loss (C) at follow-up angiography in subjects with different tertiles of SD of HbA 1c .