Impact of Visit-to-Visit Glycemic Variability on the Risks of Macrovascular and Microvascular Events and All-Cause Mortality in Type 2 Diabetes: The ADVANCE Trial

ADVANCE was a factorial randomized controlled trial of blood pressure lowering and intensive blood glucose treatment in patients with type 2 diabetes. The design has been described in detail previously (19,20). In brief, a total of 11,140 patients with type 2 diabetes aged 55 years or older who had a history of major macrovascular or microvascular disease, or at least one other risk factor for vascular disease, were recruited from 215 collaborating centers in 20 countries from Asia, Australasia, Europe, and North America between November 2001 and March 2003. Approval for the trial was obtained from each center’s institutional review board, and all participants provided written informed consent.

Participants were randomly assigned to a fixed combination of perindopril and indapamide (2 mg/0.625 mg for the first 3 months and 4 mg/1.25 mg thereafter) or matching placebo, and to either an intensive glucose control strategy (target HbA_1c ≤6.5%) or a standard glucose control strategy based on local guidelines.

In our primary analysis, we only included patients in the intensive glucose treatment group because our study protocol made it unlikely that we could reliably estimate glucose variability in the standard glucose treatment group, in which few measurements were taken (HbA_1c was only measured at 6, 12, and 24 months during the first 2 years of follow-up, and glucose was first measured at 24 months). After excluding 452 patients who experienced study outcomes of macro- and microvascular events or death during the first 24 months and 720 with missing values of HbA_1c and/or fasting glucose (morning), 4,399 patients were included in the primary analysis (Fig. 1).

Fasting samples for glucose assay and HbA_1c were taken at baseline, at 3, 6, 12, 18, and 24 months, and every 6 months thereafter in the intensive glucose treatment group. HbA_1c and fasting glucose were measured in local laboratories, and each HbA_1c measurement was standardized. Visit-to-visit variability (VVV) of HbA_1c or fasting glucose was evaluated using five measurements at 3, 6, 12, 18, and 24 months (Fig. 1). Measurements during the first 2 months were excluded in order to eliminate the effect of rapid reduction in HbA_1c in this early period. The SD, coefficient of variation (CV), variation independent of mean (VIM), residual standard deviation (RSD), and average real variability (ARV) were used as indices of VVV (21,22). Mean and maximum values of HbA_1c during the 24-month evaluation period were also used in the present analyses.

In all analyses of outcomes, follow-up of patients ranged from their 24-month visit until they experienced events, death, or they completed the final visit at the end of the study. The primary outcome was a composite of major macrovascular and major microvascular events. Secondary outcomes were major macrovascular events, major microvascular events, and all-cause mortality. Major macrovascular events were defined as death from cardiovascular causes, nonfatal myocardial infarction, or nonfatal stroke. Major microvascular events were defined as new or worsening nephropathy (development of macroalbuminuria, defined as a urinary albumin-creatinine ratio of >300 μg of albumin per milligram of creatinine, or doubling of serum creatinine level to at least 200 μmol/L, the need for renal replacement therapy, or death due to renal disease) or retinopathy (development of proliferative retinopathy, macular edema, or diabetes-related blindness or the use of retinal photocoagulation therapy).

In order to test whether glycemic variability increased with time, the difference of the SD in glycemic variables between the first 2 years and thereafter was tested using a paired Student t test. Spearman correlation coefficients were estimated between measures of variability, maximum and mean HbA_1c or fasting glucose during the first 24 months. The effect of SD and maximum value of HbA_1c or fasting glucose were estimated by Cox proportional hazards models using groups defined by the deciles (the tenths). Adjustments were made for 1) age, sex, and randomized blood pressure lowering; 2) variables in model 1 plus region, duration of diabetes, baseline smoking status, baseline alcohol intake, systolic blood pressure, total cholesterol, log-transformed triglycerides, BMI, baseline use of oral glucose-lowering agents, and baseline use of insulin; and 3) variables in model 2 plus mean of HbA_1c or fasting glucose during the first 24 months. A linear trend was tested using a quantitative ordinal variable ranging from 1 to 10 to represent the tenths. Hazard ratios (HRs) and 95% CIs for each tenth relative to lowest category and for each 10-percentile point increase (trend) in variability were estimated in each model. Effects of variability (SD, CV, VIM, RSD, and ARV), maximum value, and average of HbA_1c or fasting glucose as continuous variables were also investigated using Cox proportional hazards models, and HRs per one SD increments were reported.

Sensitivity analyses were performed 1) using five measurements of HbA_1c or fasting glucose during 6–30 months to further eliminate the effect of early reduction in HbA_1c and 2) using the models including the imputation of missing values of HbA_1c or fasting glucose among 720 patients using multiple imputation (23). The effects of SD of HbA_1c or fasting glucose were compared between subgroups defined by change in HbA_1c from 3 to 24 months within the range of −0.3 (−3.3) to 0.3% (3.3 mmol/mol) or larger. Sensitivity analyses in the standard glucose treatment group were also performed using three measurements at 6, 12, and 24 months (Supplementary Data). Finally, in order to assess the effect of randomized treatment, the SD of HbA_1c was evaluated from three measurements at 6, 12, and 24 months and compared between the intensive and standard glucose treatment groups. The heterogeneity between subgroups was tested by adding an interaction term. All statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC). A two-sided P < 0.05 was considered to be statistically significant.

Patient characteristics and visit-to-visit glycemic variability are summarized in Table 1. The patients in the intensive glucose treatment group excluded from the analysis included a group of 452 who experienced events in the first 24 months and was thus a high risk subset. However, the baseline characteristics of patients included in the analyses were similar to those for the total ADVANCE population. Mean age was 65.5 years, 42.6% were female, and 31.9% were recruited in Asia. Maximum of HbA_1c or fasting glucose was strongly correlated with mean HbA_1c or fasting glucose (r > 0.90 for maximum), whereas indicators of VVV had modest to little correlation with mean HbA_1c or fasting glucose during 3–24 months (r = 0.30–0.60 for SD, CV, RSD, and ARV; <0.15 for VIM) (Supplementary Table 1).

The mean SD of HbA_1c or fasting glucose for the intensive group was greater when estimated using five measurements rather than three (0.61 [6.7] vs. 0.56% [6.1 mmol/mol] for SD of HbA_1c; 1.41 vs. 1.32 mmol/L for SD of fasting glucose). The mean SD of HbA_1c or fasting glucose for the intensive group was significantly decreased from the first 2 years to the remaining follow-up period (SD of HbA_1c 0.60% [6.6 mmol/mol] during the first 2 years vs. 0.44% [4.8 mmol/mol] thereafter among 3,192 surviving patients, P < 0.001; SD of fasting glucose 1.38 mmol/L during the first 2 years vs. 1.16 mmol/L thereafter among 3,202 surviving patients, P < 0.001). The mean SD of HbA_1c derived from three measurements was slightly smaller in the intensive group than in the standard group (0.56 [6.1] vs. 0.61% [6.7 mmol/mol]).

Among 4,399 patients in the intensive glucose treatment group, there were 234 patients with major macrovascular events, 309 with major microvascular events, and 211 with deaths (43.1% for cardiovascular disease; 34.1% for cancer) during a median follow-up of 3.0 years (maximum 4.1 years) after completion of the 24-month visit. There were significant linear associations of SD of HbA_1c with combined macro- and microvascular events (P = 0.01 for trend), major macrovascular events (P = 0.02 for trend), and all-cause mortality (P < 0.001 for trend) even after adjusting for mean HbA_1c during the first 24 months and other confounding factors. SD of fasting glucose, adjusted for mean fasting glucose during the first 24 months and other factors, was also continuously associated with combined macro- and microvascular events (P < 0.001 for trend), major macrovascular events (P < 0.001 for trend), and major microvascular events (P = 0.005 for trend) (Fig. 2). These associations between SD of HbA_1c or fasting glucose and all outcomes were broadly similar for measurements during 6–30 months (Supplementary Fig. 1), and when missing values were imputed in the models (Supplementary Fig. 2). There was no heterogeneity in the effects of SD of HbA_1c or fasting glucose in subgroups defined by change in HbA_1c from 3 to 24 months within the range of −0.3 (−3.3) to 0.3% (3.3 mmol/mol) or larger (all P for heterogeneity >0.25 for any of four outcomes) (Supplementary Table 2). Similar results were obtained when different indicators of variability in HbA_1c or fasting glucose were used as continuous variables (Supplementary Tables 3 and 4).

There were no clear associations between maximum HbA_1c among five measurements and any of the four outcomes (Fig. 3). In contrast, maximum fasting glucose among five measurements had significant associations with combined macro- and microvascular events (P < 0.001 for trend), major macrovascular events (P = 0.01 for trend), and major microvascular events (P = 0.007 for trend) (Fig. 3). These associations between maximum HbA_1c or fasting glucose and all outcomes were broadly similar for measurements during 6–30 months (Supplementary Fig. 3) and when missing values were imputed in the models (Supplementary Fig. 4).

Sensitivity analyses were performed to elucidate the effect of VVV of HbA_1c in the standard glucose treatment group in which only three measurements of HbA_1c were available. For comparison, VVV of HbA_1c based on three measurements was also evaluated in the intensive glucose treatment group. There were no clear differences in the effects of SD of HbA_1c on any of four outcomes between the intensive and standard glucose treatment groups (all P > 0.05 for homogeneity) (Supplementary Fig. 5).

This is the first large-scale study to report the effects of VVV in both HbA_1c and fasting glucose on a variety of outcomes in type 2 diabetes, including macrovascular events. The present analyses revealed that VVV in HbA_1c as well as that in fasting glucose predicted future development of macrovascular events, microvascular events, and all-cause deaths independent of cardiovascular risk factors, including mean HbA_1c or fasting glucose, and study treatments. We have also shown that maximum value of fasting glucose was significantly associated with macrovascular and microvascular events. Both VVV and maximum of fasting glucose were stronger predictors than those of HbA_1c in that the HRs per decile increase were greater for both than those for HbA_1c.

Several medium-sized observational studies have reported positive associations between VVV in HbA_1c and development or progression of microalbuminuria in type 2 diabetes (8–10), but we are unaware of the results for other types of outcomes. In the present analyses of ADVANCE, the association between SD of HbA_1c and microvascular events did not reach statistical significance (P = 0.06 for trend) although this is the largest study of type 2 diabetes to date. Instead, there were clear associations of SD of HbA_1c with macrovascular events and all-cause mortality.

The outcomes studied in previous observational studies that assessed the effects of VVV in fasting glucose in patients with type 2 diabetes were retinopathy and mortality only (4–7,24). Most of those studies that investigated the risk of retinopathy reported positive associations (4,6,7). An Italian population-based prospective cohort study also demonstrated a positive association between VVV in fasting glucose and all-cause mortality (24). In the current study, VVV in fasting glucose during the first 2 years was significantly associated with elevated risk of microvascular events and marginally with increased mortality (P = 0.07 for trend). The present results confirmed the findings of the previous observational studies and expanded the current evidence to incident macrovascular events.

Several mechanisms may be involved in the association between visit-to-visit glycemic variability and outcomes. First, glucose fluctuation has been shown to cause overproduction of superoxide (25,26). The oxidative stress generation is a key factor of atherosclerosis (27,28). Second, glycemic fluctuation has also been shown to cause an increase in inflammatory cytokines and monocyte and macrophage adhesion to endothelial cells (28), which are involved in the progression of atherosclerosis. Third, transient hyperglycemia causes long-lasting epigenetic changes (29), which may promote systemic inflammation. Additionally, glucose fluctuation has also been shown to cause loss of pancreatic β-cells due to increased apoptotic cell death (30). The loss of pancreatic β-cells may result in deterioration of glycemic control (31) and subsequent progression of vascular complications or poor prognosis.

The strength of our study includes the large sample size and the rigorous evaluation of glycemic parameters and adjudication of major outcomes including macrovascular events. A limitation is possible selection bias due to exclusion of patients who experienced events during the first 24 months and those with missing values. However, baseline characteristics of patients included in the analyses were almost identical to those of the ADVANCE population (19), and the sensitivity analyses including imputation of missing values for visit-to-visit glycemic variability provided similar results. Second, SD may be overestimated among the patients who experienced continued improvement in blood glucose during the evaluation period from 3 to 24 months. However, sensitivity analyses using RSD, which is an indicator of variability completely independent of changes over time, showed comparable findings, suggesting that this limitation is not likely to invalidate the findings of the present analysis. Third, we are unable to provide the accuracy of the particular assays used in the individual centers to measure HbA_1c and fasting glucose. Fourth, the possibility of residual confounding by unmeasured or unknown risk factors remains. Fifth, since the current study is observational, clinical trials investigating the favorable effects of reducing glycemic variability are needed to elucidate any direct causality. Last, the effect of visit-to-visit glycemic variability could not be reliably explored in the standard glucose treatment group because measurements were few. However, broadly similar results were obtained in the sensitivity analyses using three measurements in both randomized groups.

In conclusion, VVV in HbA_1c and that in fasting glucose were associated with increased risks of macrovascular events, microvascular events, and all-cause deaths in patients with type 2 diabetes. Intensive and consistent glucose control that achieves target HbA_1c/glucose levels recommended by current guidelines (32) with minimum fluctuation may provide further protection against future development of both macro- and microvascular complications and subsequent death in type 2 diabetes.

Funding. The ADVANCE trial was funded by a grant from the National Health and Medical Research Council (NHMRC) of Australia. Y.H. holds a Research Fellowship Award from the Uehara Memorial Foundation for research activities. H.A. holds a Future Fellowship from the Australian Research Council. M.C. holds an Australia Fellowship from the NHMRC. M.W. holds a Senior Research Fellowship from the NHMRC.

Duality of Interest. The ADVANCE trial was funded by a grant from Servier International. S.Z., G.M., and M.W. have received lecture fees from Servier International. M.C. has received lecture fees from Servier International, is a member of advisory boards for Merck, AstraZeneca, Bristol-Myers Squibb, Takeda, Boehringer Ingelheim, Eli Lilly and Company, and has received lecture fees from AstraZeneca and Boehringer Ingelheim. P.H. and S.H. have received lecture and consulting fees from Servier International. N.P. and J.C. have received research grants and lecture fees from Servier International. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. Y.H. and H.A. contributed to the concept and rationale for the study, data analyses, and interpretation of the results. S.Z., T.N., M.W., and J.C. contributed to the concept and rationale for the study and interpretation of the results. M.C., P.H., G.M., N.P., and S.H. contributed to the interpretation of the results. All authors participated in the drafting and approval of the final manuscript and take responsibility for the content and integrity of this article. J.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.