Long-term visit-to-visit glycemic variability as predictor of micro- and macrovascular complications in patients with type 2 diabetes: The Rio de Janeiro Type 2 Diabetes Cohort Study

Background Long-term visit-to-visit glycemic variability is an additional measure of glycemic control. We aimed to evaluate the prognostic value of several measures of glycemic variability for the occurrence of micro- and macrovascular complications, and all-cause mortality in patients with type 2 diabetes. Methods 654 individuals were followed-up over a median of 9.3 years. Glycemic variability (SDs and coefficients of variation of HbA1c and fasting glycaemia) was measured during the first 12- and 24-months. Multivariate Cox analysis, adjusted for risk factors and mean HbA1c and fasting glycaemia levels, examined the associations between glycemic variability and the occurrence of microvascular (retinopathy, microalbuminuria, renal function deterioration, peripheral neuropathy) and macrovascular complications [total cardiovascular events (CVE), major adverse CVEs (MACE) and cardiovascular mortality], and of all-cause mortality. Results During follow-up, 128 patients had a CVE (96 MACE), and 158 patients died (67 from cardiovascular diseases); 152 newly-developed or worsened diabetic retinopathy, 183 achieved the renal composite outcome (89 newly developed microalbuminuria and 91 deteriorated renal function), and 96 newly-developed or worsened peripheral neuropathy. Glycemic variability, particularly the 24-month parameters either estimated by HbA1c or by fasting glycemia, predicted all endpoints, except for retinopathy and peripheral neuropathy development/progression, and was a better predictor than mean HbA1c. Glycemic variability predicted retinopathy development/progression in patients with good glycemic control (HbA1c ≤ 7.5%, 58 mmol/mol) and predicted new-incident peripheral neuropathy. Conclusions Long-term visit-to-visit glycemic variability is an additional and frequently a better glycemic parameter than mean HbA1c levels for assessing the risk of future development of micro- and macrovascular complications in patients with type 2 diabetes.


Background
Type 2 diabetes is an important public health problem worldwide, not only by its increasing prevalence, but also because its associated micro-and macrovascular complications severely impact on individuals' quality of life and pose a great burden on healthcare systems [1,2]. Current diabetes treatment aims to reducing chronic complications development and progression, mainly by controlling hyperglycemia, high blood pressure (BP) levels and dyslipidemia; and glycemic control is traditionally monitored by serial mean glycated hemoglobin (HbA 1c ) levels [3]. This rationale comes from several trials and observational studies showing that lowering HbA 1c indeed reduces the risk of micro-and macrovascular complications [4][5][6]. However, particularly for macrovascular disease, there is no consensus that lowering HbA1c to levels below 8.5-8.0% (69-64 mmol/mol) actually reduces such risk [7,8]. Moreover, even for microvascular complications there may be additional glycemic factors associated with increased risks, beyond mean HbA1c levels [9,10]. In this context, the concept of glycemic variability has recently emerged as another measure of glycemic control, which might constitute an additive, or even better predictor of diabetic complications than mean HbA1c levels [11,12].
Therefore, the aim of this study was to evaluate the prognostic value of several measures of long-term glycemic variability for the occurrence of separated microvascular (retinopathy, microalbuminuria, renal function deterioration, and peripheral neuropathy) and macrovascular complications (total cardiovascular events, major cardiovascular events, cardiovascular mortality), and all-cause mortality in The Rio de Janeiro Type 2 Diabetes (RIO-T2D) Cohort Study, an on-going cohort of high cardiovascular risk individuals with type 2 diabetes with a median follow-up of nearly 10 years.

Patients and baseline procedures
This was a prospective study, nested within The Rio de Janeiro Type 2 Diabetes Cohort Study, with 654 patients with type 2 diabetes enrolled between August 2004 and December 2008 and re-evaluated annually until December 2016 in the diabetes outpatient clinic of our tertiarycare University Hospital. All participants gave written informed consent, and the local Ethics Committee had previously approved the study protocol. The characteristics of this cohort, the baseline procedures and the diagnostic definitions have been detailed elsewhere [44][45][46][47]. In brief, inclusion criteria were all adult type 2 diabetic individual up to 80 years old with either any microvascular (retinopathy, nephropathy or neuropathy) or macrovascular (coronary, cerebrovascular or peripheral artery disease) complication, or with at least two other modifiable cardiovascular risk factors. Exclusion criteria were morbid obesity (body mass index ≥ 40 kg/m 2 ), advanced renal failure (serum creatinine > 180 μmol/l or estimated glomerular filtration rate < 30 ml/min/1.73 m 2 ) or the presence of any serious concomitant disease limiting life expectancy. All were submitted to a standard baseline protocol that included a thorough clinical examination, a laboratory evaluation, and a 24-h ambulatory BP monitoring (ABPM). Diagnostic criteria for diabetic chronic complications were detailed previously [44][45][46][47]. In brief, coronary heart disease was diagnosed by clinical, electrocardiographic criteria, or by positive ischemic stress tests. Cerebrovascular disease was diagnosed by history and physical examination, and peripheral arterial disease by an ankle-brachial index < 0.9. The diagnosis of nephropathy needed at least two albuminurias ≥30 mg/24 h or proteinurias ≥0.5 g/24 h or confirmed reduction of glomerular filtration rate (eGFR ≤ 60 ml/ min/1.73 m 2 , estimated by the CKD-EPI equation, or serum creatinine > 130 μmol/l). Peripheral neuropathy was determined by clinical examination (knee and ankle reflex activities, feet sensation with the Semmes-Weinstein monofilament, vibration with a 128-Hz tuning fork, pinprick and temperature sensations) and neuropathic symptoms were assessed by a standard validated questionnaire [45]. Clinic blood pressure (BP) was measured three times using a digital oscillometric BP monitor (HEM-907XL, Omron Healthcare, Kyoto, Japan) with a suitable sized cuff on two occasions 2 weeks apart at study entry. The first measure of each visit was discarded and BP considered was the mean between the last two readings of each visit. Arterial hypertension was diagnosed if mean systolic (SBP) ≥ 140 mmHg or diastolic BP (DBP) ≥ 90 mmHg or if anti-hypertensive drugs had been prescribed. ABPM was recorded in the following month using Mobil-O-Graph, version 12 equipment (Dynamapa, Cardios LTDA., São Paulo, Brazil), and average 24-h SBP and DBP were registered [47]. Laboratory evaluation included fasting glycemia (FG), glycated hemoglobin (HbA 1c ), serum creatinine and lipids. Albuminuria and proteinuria were evaluated in two non-consecutive sterile 24-h urine collections.

Long-term glycemic variability measurements
The patients had at least three annual HbA 1c and FG measurements during follow-up. Long-term visit-to-visit glycemic variability was estimated separately for HbA 1c and FG, and for the first 12 and 24-month periods, as the standard deviation (SD) of all measurements performed during these periods. To account for the possible influence of different number of measurements, the SD was divided by √ [n/(n − 1)] , as previously suggested [10]. The coefficient of variation (SD/mean) was also calculated for each glycemic variability parameter.

Follow-up and outcomes assessment
The patients were followed-up regularly at least 3-4 times a year until December 2016 under standardized treatment. The observation period for each patient was the number of months from the date of the first clinical examination to the date of the last clinical visit in 2016 or the date of the first endpoint, whichever came first. The primary endpoints were the occurrence of any macrovascular or microvascular outcomes. Macrovascular outcomes were total cardiovascular events (CVEs: fatal or non-fatal myocardial infarctions, sudden cardiac deaths, new-onset heart failure, death from progressive heart failure, any myocardial revascularization procedure, fatal or non-fatal strokes, any aortic or lower limb revascularization procedure, any amputation above the ankle, and deaths from aortic or peripheral arterial disease), major adverse cardiovascular events (MACE: non-fatal myocardial infarctions and strokes plus cardiovascular deaths), and all-cause and cardiovascular mortalities [44]. Mortality, as well as non-fatal cardiovascular events occurrence, was ascertained from medical records, death certificates and interviews with attending physicians and patient families, by a standard questionnaire reviewed by two independent observers. In case of disagreement, it was decided by consensus with a third independent consultant. Most of the in-hospital fatal or non-fatal events were attended at our own hospital. Microvascular outcomes were retinopathy development or worsening [46], renal outcomes [47] [new microalbuminuria development, new renal failure development (defined as doubling of serum creatinine or end-stage renal failure needing dialysis or death from renal failure), and a composite of them], and peripheral neuropathy development or worsening [45]. Retinopathy and renal outcomes were evaluated by annual examinations [46,47], whereas peripheral neuropathy was evaluated on a second specific examination performed after a median of 6 years from the baseline examination [45].

Statistical analyses
Continuous data were described as means (SD) or as medians (interquartile range). For initial exploratory analyses, patients were categorized into tertiles of glycemic variability parameters and baseline characteristics compared by ANOVA, Kruskal-Wallis or χ 2 tests, when appropriate. Kaplan-Meier curves of cumulative endpoints incidence during follow-up, compared by log-rank tests, were used for assessing different incidences of outcomes among tertile subgroups. For assessing the prognostic value of each glycemic variability parameter for each macrovascular and microvascular outcome, except for peripheral neuropathy, a time-to-event Cox analysis was undertaken with progressively increasing statistical adjustments for potential confounding. Model 1 was only adjusted for age, sex and number of HbA 1c or FG measurements, model 2 was further adjusted for other potential confounders (diabetes duration, body mass index (BMI), smoking status, physical inactivity, arterial hypertension, number of anti-hypertensive drugs in use, ambulatory 24-h SBP, presence of each micro-and macrovascular complications at baseline, serum mean HDL-and LDLcholesterol, and use of insulin, statins and aspirin), and model 3 was further adjusted for mean FG and HbA 1c levels during the same period of glycemic variability measurement. These results were presented as hazard ratios (HRs) with their 95% confidence intervals (CIs); to allow comparisons among different glycemic variability parameters, their HRs were calculated for standardized increments of 1-SD. As glycemic variability was measured during the first 2 years of follow-up, patients who presented any of the endpoints during this period were excluded from the analysis of this specific outcome. For peripheral neuropathy analyses, a multiple logistic regression was used with the same progressively increasing statistical adjustments, except that height (instead of BMI) and the time interval between the baseline and second neuropathy evaluations were included as adjusting covariates. These results were reported as odds ratios (ORs) with their respective 95% CIs, also estimated for increments of 1-SD in each glycemic variability parameter. The same analyses were performed for patients categorized into tertiles of each glycemic variability parameter, with HRs and ORs calculated for the highest tertile subgroup in relation to the lowest tertile reference subgroup, after adjustments for the same covariates. Interaction between mean HbA 1c and glycemic variability measures were tested for all endpoints and whenever there was evidence of interaction (p < 0.10 for interaction term), stratified analyses for high (> 7.5%, 58 mmol/mol) and low (≤ 7.5%) HbA 1c levels were performed. Statistics were performed with SPSS version 19.0 (SPSS Inc, Chicago, Il., USA), and a 2-tailed probability value < 0.05 was considered significant.

Baseline characteristics
For macrovascular and mortality outcomes, 654 patients without any endpoint occurrence during the first 2 years of follow-up were evaluated. For microvascular outcomes, 615 patients were evaluated for renal, 533 for retinopathy and 471 for peripheral neuropathy outcomes. Patients had a median of 4 HbA 1c (range 3-6) and 5 FG (range 3-7) measurements during the first 12 months of follow-up, and a median of 8 HbA 1c (range 6-11) and 10 FG (range 6-14) measurements during the first 24 months of follow-up. Median time interval between each visit-to-visit HbA 1c and FG measurements was 3 months. Table 1 outlines the baseline characteristics of all patients and of those divided according to tertiles of 24-month HbA 1c variability. Patients with higher HbA 1c -SD were younger, but with longer diabetes duration, and had higher prevalences of microvascular complications than those with lower HbA 1c variability. They also had higher BP levels, particularly at ABPM, and poorer glycemic control, although using insulin more frequently, than those with lower HbA 1c variability. Table 2 shows the same baseline characteristics of patients divided according to 24-month FG variability. In general, they follow the same patterns of HbA 1c variability, except that patients with higher FG-SD also had greater prevalences of macrovascular complications than those with lower FG variability.

Endpoints occurrence during follow-up
Over a median follow-up of 9.3 years (IQR 5.2-10.8 years), 128 patients had a CVE (96 MACE), and 158 patients died (67 from cardiovascular diseases); 152 newly-developed or worsened diabetic retinopathy, 183 achieved the renal composite outcome (89 newly developed microalbuminuria and 91 deteriorated renal function), and 96 newly-developed or worsened peripheral neuropathy. Tables 1 and 2 show that patients with higher long-term glycemic variability had a significantly higher incidence of all endpoints, except of all-cause mortality and new microalbuminuria development for HbA 1c variability and of new peripheral neuropathy development for FG variability. Kaplan-Meier curves of cumulative incidence of endpoints (Figs. 1, 2) shows that for most of the endpoints the increased incidence was mainly observed in the highest tertile variability subgroup in relation to the middle and lowest tertile subgroups, except for retinopathy (for HbA 1c variability) and renal outcomes (for FG variability), where those patients in the middle tertile subgroup already had an increased incidence of these endpoints. Table 3 (for macrovascular and mortality outcomes) and Table 4 (for microvascular outcomes) present the risks associated with a 1-SD increment in each 12-and 24-month glycemic variability parameter after increasing levels of confounding variables adjustments. As a whole, 24-month glycemic variability parameters were better risk predictors than 12-month parameters, and variabilities estimated by SDs and by CVs were roughly equivalent. For cardiovascular endpoints, particularly for MACE, 24-month glycemic variability, either estimated by HbA 1c or by FG, were independent predictors of any macrovascular and mortality outcomes, and were only predictors of retinopathy and peripheral neuropathy development or progression, but not of new neuropathy incidence. Tables 5 and 6 show the same analyses with patients categorized into tertiles of glycemic variability parameters, and the results were mainly consistent with the continuous parameters analyses, although some parameters were non-significant because of the wider confidence intervals associated with the lower number of endpoints in each tertile subgroups. There was evidence of interaction (p < 0.10) between mean HbA 1c levels and glycemic variability in analyses for diabetic retinopathy.

Discussion
This prospective cohort study with a median follow-up of nearly 10 years has some important new findings. First, it demonstrated that for all micro-and macrovascular outcomes, except for retinopathy and peripheral neuropathy development or progression, 24-month visit-tovisit glycemic variability parameters, either estimated for HbA 1c or for FG, were better risk predictors than mean HbA 1c levels during this same time interval. Second, specifically for diabetic retinopathy development or progression, 24-month HbA 1c variability was a significant risk predictor in patients with good glycemic control (with mean HbA1c ≤ 7.5%, 58 mmol/mol); but not in those with poorer controlled diabetes (HbA1c > 7.5%), where mean HbA 1c levels were the main risk predictor. Third, specifically for peripheral neuropathy, mean HbA 1c was the main risk predictor for the composite outcome of developing or worsening neuropathy, whereas HbA 1c variability was a better risk predictor for new-incident peripheral neuropathy. Overall, our findings support the concept that long-term visit-to-visit glycemic variability is an additional and frequently a better glycemic parameter than mean HbA 1c levels for assessing the risk of future development of micro-and macrovascular diabetic complications. Several previous studies evaluated long-term glycemic variability parameters in patients with type 2 diabetes . As a whole, most agree that glycemic variability predicts all-cause mortality [29,32,33,[36][37][38][39][40], fatal or non-fatal cardiovascular diseases [23,[28][29][30][31][32], new microalbuminuria development [16,[18][19][20][21][22]25] and renal function deterioration [16,23,24,26], although there were opposing reports for these outcomes [19,29,[33][34][35][36]41]. A recent meta-analysis including studies published until 2014 confirmed these findings [12]. Our results support these previous investigations. An intriguing finding of our study was that, mainly for cardiovascular and all-cause mortalities, FG variability seemed a stronger risk predictor than HbA 1c variability.

Values are proportions, and means (standard deviations) or medians (interquartile range)
FG fasting glycemia, ACE angiotensin-converting enzyme, AR angiotensin II receptor, SBP systolic blood pressure, DBP diastolic blood pressure, HbA 1c glycated hemoglobin, HDL high-density lipoprotein, LDL low-density lipoprotein, CVEs cardiovascular events a Values are absolute numbers (incidence rate per 100 patient-years of follow-up) b Values are absolute numbers (incidence rate per 100 patient-years of follow-up), except for peripheral neuropathy that are absolute numbers (proportions) The reason can be simple statistical adjustments because mean HbA 1c is expected to attenuate more HbA 1c variability than FG variability. However, models were also adjusted for mean FG levels that would attenuate FG variability, although not at the same extent given that mean HbA 1c was more strongly associated with the outcomes than mean FG. Alternatively, FG variability may have captured more accurately hypoglycemic episodes than HbA 1c variability [36,48]. Severe hypoglycemia is well-known associated with adverse prognosis in type 2 diabetes, particularly with increased mortality [49,50]. Postprandial hyperglycemia is another issue of concern that might not have been adequately captured by either FG or HbA 1c variability parameters. In this regard, low 1,5-anhydroglucitol levels, a potential marker of postprandial hyperglycemia, have been reported to predict worse cardiovascular outcomes in patients with acute coronary syndrome [51] and in patients with stable coronary heart disease submitted to elective angiography [52], both groups with low HbA 1c levels (< 7.0%).
Regarding the value of glycemic variability as risk predictor for future diabetic retinopathy development or progression, previous reports were controversial, with some showing the predictive capacity of FG variability [13,15], whereas others negated the importance of glycemic variability parameters [14,[16][17][18]  meta-analysis also did not demonstrate any prognostic value for retinopathy development [12], but included only two studies. Our study provided new findings, by showing that the predictive power of HbA 1c variability for retinopathy development or progression depends on mean HbA 1c levels, being positive in patients with better-controlled diabetes, but absent in those poorly-controlled. As far as we know, this is the first prospective study to assess the importance of glycemic variability for predicting diabetic peripheral neuropathy development or progression. Only a previous study [27] had evaluated its value for cardiovascular autonomic neuropathy development, with positive findings. We showed that HbA 1c variability was a predictor mainly of new-incident peripheral neuropathy, whereas mean HbA 1c levels mainly predicted its worsening.
From a physiopathological standpoint, there were several potential mechanisms that may link increased glycemic variability to the future occurrence of diabetic micro-and macrovascular complications and to mortality. Acute, short-term glycemia fluctuations induce superoxide overproduction, increased oxidative stress, inflammatory cytokines generation and endothelial dysfunction and damage [11,53,54], all linked to chronic diabetic complications. Moreover, exaggerated glycemic fluctuations were demonstrated to adversely affect endothelial vessel healing, increasing neointimal Number of paƟents at risk: thickness following percutaneous stent implantation [55], and augmenting the risk of periprocedural and short-term cardiovascular complications [56,57], which may be particularly important in such a high cardiovascular risk population as ours. Further, transient hyperglycemia might cause epigenetic changes, inducing cellular metabolic memory [58,59], increasing insulin resistance [60] and pancreatic β-cell dysfunction and apoptosis [61]. Alternatively, but not excluding, increased glycemic variability might simply be a marker of unstable glycemic control due to poor treatment adherence and self-management patient compliance [12,31], multimorbidity, poor quality of life and lack of social support, and frequent infections complications [12]. In this regard, it should be noted that patients with higher glycemic variability at baseline were less physically active, more frequently current or past smokers, and had a greater prevalence of diabetic complications than those with lower variability, particularly evident for FG variability. The study has some limitations that shall be noted. First, it is a prospective observational cohort; hence no cause-and-effect relations, nor physiopathological

Table 3 Results of Cox survival analyses for the excess risks associated with 12-and 24-month glycemic variability parameters, analyzed as continuous variables, for the occurrence of future macrovascular complications and mortality
Values are hazard ratios and 95% confidence intervals, estimated for increases of 1-SD in each glycemic parameter Model 1 is adjusted for age, sex and number of HbA 1c or FG measurements Model 2 is further adjusted for diabetes duration, BMI, smoking status, physical inactivity, arterial hypertension, number of anti-hypertensive drugs in use, ambulatory 24-h SBP, presence of micro-and macrovascular complications at baseline, serum mean HDL-and LDL-cholesterol, and use of insulin, statins and aspirin Model 3 is further adjusted for mean fasting glycemia and HbA 1c HR hazard ratio, CI confidence interval, CV cardiovascular, FG-SD fasting glucose standard deviation, FG-VC fasting glucose variation coefficient, HbA 1c -SD glycated hemoglobin standard deviation, HbA 1c -VC glycated hemoglobin variation coefficient, HbA 1c -MEAN mean glycated hemoglobin during the same time interval * p < 0.001; † p < 0.01; ‡ p < 0.05 a The HR of HbA 1c -MEAN was estimated also for increases of 1-SD in the model with the highest HR of the glycemic variability parameter, whichever it was inferences, can be made, but only speculated. Moreover, as with any cohort study, residual confounding due to unmeasured or unknown factors can not be ruled out. Second, it enrolled mainly middle-aged to elderly individuals with long-standing type 2 diabetes and with a high prevalence of chronic complications followed-up in a tertiary-care university hospital. Hence, our results might not be generalized to younger individuals with Table 4 Results of multivariable analyses for the excess risks associated with 12-and 24-month glycemic variability parameters, analyzed as continuous variables, for the occurrence of future diabetic microvascular complications Values are hazard ratios and 95% confidence intervals, estimated by Cox analyses for increases of 1-SD in each glycemic parameter; except for peripheral neuropathy outcomes that are odds ratios and 95% confidence intervals, estimated by logistic regressions Models 1, 2 and 3 were adjusted for the same covariates as in Table 3, except for peripheral neuropathy that was adjusted for height instead of BMI and further for the time interval between baseline and second neuropathy examination recent onset type 2 diabetes or at primary care management. Third, changes in anti-diabetic medications during follow-up, particularly initiating or increasing insulin dosages, which probably affected glycemic variability during the first 2 years of follow-up, were not taken into account. Forth, peripheral neuropathy assessment was not performed annually during follow-up, as the other outcomes, but on two specific time points (at baseline and after a median of 6 years), which might have affected this endpoint evaluation, although this specific analysis took into account the differential time interval between neuropathy assessments. Finally, we did not adjust for multiple comparisons within each outcome. However, as we have evaluated 4 glycemic variability parameters obtained during 2 time intervals, that is 8 measures for each outcome; if we applied Bonferroni's correction, we would have considered a p value < 0.006 as significant.
With this more conservative approach, only the predictive capacity of glycemic variability for MACE and microalbuminuria incidence would be lost. On the other hand, this study main strength is its well-documented cohort with standardized care and annual outcomes evaluation over a long follow-up, which permitted the most comprehensive analysis of the associations between long-term glycemic variability parameters and risks of separate micro-and macrovascular complications and of mortality in patients with type 2 diabetes.