The association between high glycemic variability and all-cause mortality has been widely investigated in epidemiological studies but rarely validated in glucose-lowering clinical trials. We aimed to identify the prognostic significance of visit-to-visit HbA1c variability in treated patients in the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial population.
We studied the risk of all-cause mortality in relation to long-term visit-to-visit HbA1c variability, expressed as coefficient of variation (CV), variability independent of the mean (VIM), and average real variability (ARV), from the 8th month to the transition from intensive to standard glycemic therapy. Multivariable Cox proportional hazards models were used to estimate adjusted hazard ratio (HR) and 95% CI.
Compared with the standard therapy group (n = 4,728), the intensive therapy group (n = 4,755) had significantly lower mean HbA1c (6.6% [49 mmol/mol] vs. 7.7% [61 mmol/mol], P < 0.0001) and lower CV, VIM, and ARV (P < 0.0001). In multivariate adjusted analysis, all three HbA1c variability indices were significantly associated with total mortality in all patients as well as in the standard- and intensive-therapy groups analyzed separately. The hazard ratios for a 1-SD increase in HbA1c variability indices for all-cause mortality were 1.19 and 1.23 in intensive and standard therapy, respectively. Cross-tabulation analysis showed the third tertile of HbA1c mean and VIM had significantly higher all-cause mortality (HR 2.05; 95% CI 1.17–3.61; P < 0.01) only in the intensive-therapy group.
Long-term visit-to-visit HbA1c variability was a strong predictor of all-cause mortality. HbA1c VIM combined with HbA1c mean conferred an increased risk for all-cause mortality in the intensive-therapy group.
Introduction
Type 2 diabetes is a common and potent risk factor for cardiovascular disease (CVD) events, and observational studies have consistently shown an association between the degree of hyperglycemia and the risk of these outcomes (1–4). However, several randomized clinical trials of intensive glycemic control in patients with type 2 diabetes did not demonstrate beneficial effects on these vascular outcomes (5–7). Moreover, the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial showed increased all-cause mortality in the intensive-treatment group compared with those receiving conventional treatment (8). In these clinical trials and in daily diabetes management in clinical practice, levels of glycated hemoglobin A1c (HbA1c) or blood glucose are considered of primary importance.
Recent observational studies in diabetes indicate that greater visit-to-visit glycemic variability is associated with macrovascular events as well as with microvascular complications (9–14). Two components of glycemic variability have been recognized: short-term, over days to weeks, and long-term glycemic variability. The latter may be ascertained by calculating visit-to-visit fluctuations of HbA1c over periods of follow-up lasting months to years. An analysis of 58,832 patients with type 2 diabetes in primary care in the U.K. showed that HbA1c variability was strongly associated with overall mortality and emergency hospitalization, and to a degree, this was not explained by average HbA1c or hypoglycemic episodes (15). Furthermore, a post hoc analysis conducted in the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) showed that greater visit-to-visit variability of fasting blood glucose was associated with increased mortality risk. However, the study just had three glucose measures and included patients with and without diabetes (16). Systematic ascertainment of glycemic variability may provide additional value in the prediction of future complications among patients with diabetes, both in clinical practice and in large clinical trials.
In the current study, we used data from the ACCORD trial to investigate associations of the long-term visit-to-visit variability in HbA1c with outcomes among participants with type 2 diabetes and tested the hypothesis that HbA1c variability might play an important role in outcomes of treated type 2 diabetes.
Research Design and Methods
Study Design
The ACCORD trial design, inclusion criteria, subject characteristics, and main results have been previously described (8,17–19) (online study protocols: https://biolincc.nhlbi.nih.gov/studies/accord/). In brief, the participants were between the ages of 40 and 79 years, had type 2 diabetes and an HbA1c level of ≥7.5% (58.5 mmol/mol), had previous evidence of CVD or risk factors for CVD, and did not have a history of frequent or recent serious hypoglycemic events. The study randomly assigned 10,251 participants to receive comprehensive intensive glucose-lowering therapy targeting an HbA1c level of <6.0% (42 mmol/mol) or standard therapy targeting a level of 7.0% (53 mmol/mol) to 7.9% (63 mmol/mol). HbA1c levels were audited regularly according to treatment group and study center, and feedback was provided to facilitate attainment of the target HbA1c levels. Patients in the intensive-therapy group attended monthly visits for the first 4 months and then every 2 months thereafter, with at least one interim telephone call, with the aim of rapidly and safely reducing HbA1c levels to <6.0%. Patients in the standard-therapy group had glycemic management visits every 4 months. HbA1c was measured every 4 months in the intensive-therapy and standard-therapy groups.
Compared with standard therapy, the use of intensive therapy to target normal HbA1c levels for 3.5 years increased mortality and did not significantly reduce major cardiovascular events (8). The finding of higher mortality in the intensive-therapy group led to a decision to terminate the intensive regimen in February 2008. On 5 February 2008, participants were informed of the decision to discontinue the intensive glucose-lowering regimen after a mean treatment period of 3.7 years (17). Participants in the intensive-therapy group subsequently were switched to standard glycemic therapy. The “transition” variable in the ACCORD “activitystatus” data set specifies whether each visit for a subject was before or after the intensive glycemia intervention was stopped. In this study, the HbA1c mean and variability were calculated using the data from the 8th month to the transition period, and the events of all-cause death were defined as those occurring before transition.
This report represents a post hoc analysis of data available for ACCORD participants in the intensive- and standard-therapy arms. To avoid the high glycemic variability brought about by the early-study stages of glucose lowering, we calculated the variability occurring after the 8th month of the study. After the patients whose HbA1c was measured fewer than three times were excluded, 9,483 patients were included in the final analysis.
Statistics
For database management and statistical analysis, we used SAS 9.4 software (SAS Institute, Cary, NC). Significance was a two-tailed α level of ≤0.05. Means and proportions were compared using the large-sample z test and the χ2 statistic, respectively.
Long-term visit-to-visit HbA1c variability was evaluated using three or more HbA1c measures from the 8th month to the transition, calculating individual participant coefficient of variation (CV), variation independent of the mean (VIM) (20), and average real variability (ARV) (21). CV was calculated as the SD divided by the mean, VIM as the SD divided by the mean to the power x and multiplied by the population mean to the power x, with x derived from curve fitting (20), and ARV as the average of the absolute differences between consecutive HbA1c measurements. The association between HbA1c variability and all-cause mortality was estimated by multivariable Cox proportional hazards models while adjusting for sex, therapy group, history of CVD, diabetes duration, mean of HbA1c during visits, and age, education, waist circumference, BMI, systolic and diastolic blood pressure, triglycerides, and fasting plasma glucose at baseline.
HbA1c variability was investigated as a continuous variable using Cox proportional hazards models, and the HRs of all-cause mortality for 1 SD increment in HbA1c variability indices are reported. To allow for nonlinearity, a linear trend was tested using the variability indices in HbA1c ranging from 1 to 10 to represent deciles. Hazard ratios (HRs) and 95% CIs for each decile relative to the first decile in the standard-therapy group and for each 10-percentile point increase in variability were estimated in a single model.
Results
Characteristics of the Study Participants
Of all 10,251 participants, 9,483 had measurement of HbA1c on at least three visits from the 8th month to the transition and were included in the final analysis. The baseline characteristics of patients included in the analyses were similar to those for the total ACCORD population. Mean age was 62.7 years, and 38.2% were women. Key baseline characteristics were similar in the two therapy groups (Table 1). Mean HbA1c levels were relatively stable from the 8th month to the transition (Supplementary Table 1).
. | All patients . | Therapy status . | HbA1c variability . | ||
---|---|---|---|---|---|
Standard . | Intensive . | VIM <0.45 . | VIM ≥0.45 . | ||
. | (N = 9,483) . | (n = 4,728) . | (n = 4,755) . | (n = 4,752) . | (n = 4,731) . |
At baseline | |||||
Age (years) | 62.7 ± 6.6 | 62.8 ± 6.6 | 62.7 ± 6.6 | 63.0 ± 6.5 | 62.4 ± 6.6* |
Female sex, n (%) | 3,621 (38.2) | 1,802 (38.1) | 1,819 (38.3) | 1,806 (38.0) | 1,815 (38.4) |
Weight (kg) | 93.7 ± 18.4 | 93.7 ± 18.4 | 93.7 ± 18.4 | 92.2 ± 18 | 95.3 ± 18.6* |
BMI (kg/m2) | 32.3 ± 5.4 | 32.3 ± 5.4 | 32.2 ± 5.4 | 31.8 ± 5.3 | 32.8 ± 5.4* |
Waist circumference (cm) | 106.8 ± 13.6 | 106.8 ± 13.5 | 106.8 ± 13.7 | 105.5 ± 13.3 | 108.1 ± 13.7* |
Blood pressure | |||||
Systolic (mmHg) | 136.2 ± 16.9 | 136.4 ± 17.0 | 136.1 ± 16.8 | 135.6 ± 16.7 | 136.9 ± 17.2* |
Diastolic (mmHg) | 74.8 ± 10.6 | 74.9 ± 10.6 | 74.8 ± 10.6 | 74.3 ± 10.2 | 75.4 ± 10.9* |
Fasting serum glucose (mg/dL) | 175.2 ± 55.7 | 175.8 ± 55.9 | 174.7 ± 55.4 | 172.1 ± 53.1 | 178.4 ± 58.0* |
Total cholesterol (mg/dL) | 183.3 ± 41.8 | 183.4 ± 41.8 | 183.2 ± 41.8 | 182.2 ± 41.2 | 184.3 ± 42.3† |
LDL cholesterol (mg/dL) | 104.8 ± 33.7 | 104.9 ± 33.7 | 104.7 ± 33.7 | 104.5 ± 33.4 | 105.1 ± 34.1 |
Serum creatinine (mg/dL) | 0.9 ± 0.2 | 0.9 ± 0.2 | 0.9 ± 0.2 | 0.90 ± 0.22 | 0.92 ± 0.23* |
During follow-up | |||||
Mean HbA1c (%) | 7.1 ± 0.9 | 7.7 ± 0.7 | 6.6 ± 0.7* | 7.1 ± 0.9 | 7.2 ± 0.9* |
HbA1c SD | 0.5 ± 0.3 | 0.6 ± 0.4 | 0.4 ± 0.3* | 0.3 ± 0.2 | 0.7 ± 0.4* |
HbA1c CV | 0.08 ± 0.04 | 0.08 ± 0.04 | 0.06 ± 0.04* | 0.04 ± 0.02 | 0.1 ± 0.04* |
HbA1c VIM | 0.5 ± 0.2 | 0.51 ± 0.25 | 0.49 ± 0.2* | 0.3 ± 0.1 | 0.7 ± 0.2* |
HbA1c ARV | 0.5 ± 0.4 | 0.63 ± 0.39 | 0.39 ± 0.28* | 0.3 ± 0.2 | 0.7 ± 0.4* |
. | All patients . | Therapy status . | HbA1c variability . | ||
---|---|---|---|---|---|
Standard . | Intensive . | VIM <0.45 . | VIM ≥0.45 . | ||
. | (N = 9,483) . | (n = 4,728) . | (n = 4,755) . | (n = 4,752) . | (n = 4,731) . |
At baseline | |||||
Age (years) | 62.7 ± 6.6 | 62.8 ± 6.6 | 62.7 ± 6.6 | 63.0 ± 6.5 | 62.4 ± 6.6* |
Female sex, n (%) | 3,621 (38.2) | 1,802 (38.1) | 1,819 (38.3) | 1,806 (38.0) | 1,815 (38.4) |
Weight (kg) | 93.7 ± 18.4 | 93.7 ± 18.4 | 93.7 ± 18.4 | 92.2 ± 18 | 95.3 ± 18.6* |
BMI (kg/m2) | 32.3 ± 5.4 | 32.3 ± 5.4 | 32.2 ± 5.4 | 31.8 ± 5.3 | 32.8 ± 5.4* |
Waist circumference (cm) | 106.8 ± 13.6 | 106.8 ± 13.5 | 106.8 ± 13.7 | 105.5 ± 13.3 | 108.1 ± 13.7* |
Blood pressure | |||||
Systolic (mmHg) | 136.2 ± 16.9 | 136.4 ± 17.0 | 136.1 ± 16.8 | 135.6 ± 16.7 | 136.9 ± 17.2* |
Diastolic (mmHg) | 74.8 ± 10.6 | 74.9 ± 10.6 | 74.8 ± 10.6 | 74.3 ± 10.2 | 75.4 ± 10.9* |
Fasting serum glucose (mg/dL) | 175.2 ± 55.7 | 175.8 ± 55.9 | 174.7 ± 55.4 | 172.1 ± 53.1 | 178.4 ± 58.0* |
Total cholesterol (mg/dL) | 183.3 ± 41.8 | 183.4 ± 41.8 | 183.2 ± 41.8 | 182.2 ± 41.2 | 184.3 ± 42.3† |
LDL cholesterol (mg/dL) | 104.8 ± 33.7 | 104.9 ± 33.7 | 104.7 ± 33.7 | 104.5 ± 33.4 | 105.1 ± 34.1 |
Serum creatinine (mg/dL) | 0.9 ± 0.2 | 0.9 ± 0.2 | 0.9 ± 0.2 | 0.90 ± 0.22 | 0.92 ± 0.23* |
During follow-up | |||||
Mean HbA1c (%) | 7.1 ± 0.9 | 7.7 ± 0.7 | 6.6 ± 0.7* | 7.1 ± 0.9 | 7.2 ± 0.9* |
HbA1c SD | 0.5 ± 0.3 | 0.6 ± 0.4 | 0.4 ± 0.3* | 0.3 ± 0.2 | 0.7 ± 0.4* |
HbA1c CV | 0.08 ± 0.04 | 0.08 ± 0.04 | 0.06 ± 0.04* | 0.04 ± 0.02 | 0.1 ± 0.04* |
HbA1c VIM | 0.5 ± 0.2 | 0.51 ± 0.25 | 0.49 ± 0.2* | 0.3 ± 0.1 | 0.7 ± 0.2* |
HbA1c ARV | 0.5 ± 0.4 | 0.63 ± 0.39 | 0.39 ± 0.28* | 0.3 ± 0.2 | 0.7 ± 0.4* |
Continuous variables are shown as the means (SD) and categorical variables as indicated. HbA1c of 7.1% converts to 54.1 mmol/mol, 7.7% to 61 mmol/mol, 6.6% to 49 mmol/mol, and 7.2% to 55.2 mmol/mol.
P < 0.001; †P < 0.01.
Compared with the standard-therapy group, the intensive-therapy group had significantly lower follow-up mean HbA1c levels in the reduced subjects in our analysis (6.6% [49 mmol/mol] vs. 7.7% [61 mmol/mol], P < 0.0001) and significantly lower HbA1c variability indices, including SD, CV, VIM, and ARV (all P < 0.0001) (Table 1). CV and ARV were more frequently distributed in higher deciles in the intensive-therapy group than in the standard-therapy group, but VIM was equally distributed in each decile (Supplementary Fig. 1).
Table 1 summarizes the characteristics of the participants by the median of the distribution of VIM. Compared with low HbA1c variability (VIM <0.45 [the median value]), high HbA1c variability (VIM ≥0.45) had a significantly lower baseline age; greater baseline body weight, baseline BMI, and waist circumference; and higher baseline systolic and diastolic blood pressure, fasting serum glucose, total cholesterol, and serum creatinine. High HbA1c variability had a significantly higher mean HbA1c level during follow-up and higher HbA1c variability indices, including SD, CV, VIM, and ARV (Table 1).
Variability Indices and All-Cause Mortality
During the trial, all-cause mortality occurred in 168 and 190 subjects in the standard-therapy group and the intensive-therapy group, respectively. In multiple Cox regression analyses adjusted for sex, age, education, waist circumference, BMI, systolic and diastolic blood pressure, triglycerides, and fasting plasma glucose at baseline, all three HbA1c variability indices were significantly (P < 0.001) associated with all-cause mortality. After further adjustment for the mean of HbA1c during visits, all three HbA1c variability indices were significantly (P < 0.01) associated with all-cause mortality. The hazard ratios for a 1-SD increase in HbA1c variability indices for the all-cause mortality were 1.19–1.50 in the intensive-therapy group and 1.23–1.28 in the standard-therapy group. The ARV of HbA1c showed greater HRs in the intensive therapy group (1.50 for a 1-SD increase) than in the standard-therapy group (1.28) (Table 2).
Correlate (approximate + 1 SD) . | Adjusted model . | Total population (n = 9,483) . | Intensive therapy (n = 4,755) . | Standard therapy (n = 4,728) . | |||
---|---|---|---|---|---|---|---|
HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . | ||
All-cause mortality | |||||||
Mean (+0.9%) | Model 1 | 1.26 (1.11–1.43) | 0.0004 | 1.37 (1.16–1.61) | 0.0002 | 1.13 (0.92–1.39) | 0.23 |
CV (+0.04%) | Model 1 | 1.28 (1.16–1.40) | <0.0001 | 1.32 (1.16–1.51) | <0.0001 | 1.24 (1.08–1.42) | 0.002 |
Model 2 | 1.23 (1.11–1.36) | 0.0001 | 1.22 (1.05–1.42) | 0.01 | 1.23 (1.07–1.42) | 0.005 | |
VIM (+0.25 units) | Model 1 | 1.21 (1.10–1.33) | <0.0001 | 1.21 (1.06–1.37) | 0.003 | 1.22 (1.06–1.40) | 0.007 |
Model 2 | 1.21 (1.10–1.33) | <0.0001 | 1.19 (1.04–1.35) | 0.009 | 1.23 (1.07–1.42) | 0.005 | |
ARV (+0.36) | Model 1 | 1.37 (1.26–1.48) | <0.0001 | 1.50 (1.35–1.66) | <0.0001 | 1.26 (1.12–1.42) | 0.0001 |
Model 2 | 1.36 (1.23–1.50) | <0.0001 | 1.50 (1.30–1.75) | <0.0001 | 1.28 (1.12–1.46) | 0.0004 |
Correlate (approximate + 1 SD) . | Adjusted model . | Total population (n = 9,483) . | Intensive therapy (n = 4,755) . | Standard therapy (n = 4,728) . | |||
---|---|---|---|---|---|---|---|
HR (95% CI) . | P . | HR (95% CI) . | P . | HR (95% CI) . | P . | ||
All-cause mortality | |||||||
Mean (+0.9%) | Model 1 | 1.26 (1.11–1.43) | 0.0004 | 1.37 (1.16–1.61) | 0.0002 | 1.13 (0.92–1.39) | 0.23 |
CV (+0.04%) | Model 1 | 1.28 (1.16–1.40) | <0.0001 | 1.32 (1.16–1.51) | <0.0001 | 1.24 (1.08–1.42) | 0.002 |
Model 2 | 1.23 (1.11–1.36) | 0.0001 | 1.22 (1.05–1.42) | 0.01 | 1.23 (1.07–1.42) | 0.005 | |
VIM (+0.25 units) | Model 1 | 1.21 (1.10–1.33) | <0.0001 | 1.21 (1.06–1.37) | 0.003 | 1.22 (1.06–1.40) | 0.007 |
Model 2 | 1.21 (1.10–1.33) | <0.0001 | 1.19 (1.04–1.35) | 0.009 | 1.23 (1.07–1.42) | 0.005 | |
ARV (+0.36) | Model 1 | 1.37 (1.26–1.48) | <0.0001 | 1.50 (1.35–1.66) | <0.0001 | 1.26 (1.12–1.42) | 0.0001 |
Model 2 | 1.36 (1.23–1.50) | <0.0001 | 1.50 (1.30–1.75) | <0.0001 | 1.28 (1.12–1.46) | 0.0004 |
Model 1 was adjusted for therapy group (if applicable), sex, history of CVD, diabetes duration, and baseline age, education, BMI, systolic and diastolic blood pressure, smoking, drinking, and fasting plasma glucose. Model 2 was adjusted for mean of HbA1c during visits and the variables in model 1.
For all-cause mortality, the 10th decile of HbA1c CV, VIM, and ARV had significantly higher risk in the intensive-therapy group but not in the standard-therapy group (Supplementary Fig. 1).
Cross-Tabulation Analysis of Mean and Variability
Since the HbA1c CV and mean are correlated tightly (CV was calculated as the SD divided by the mean) and VIM was equally distributed in each decile in the intensive-therapy group and in the standard-therapy group, we used the VIM for cross-tabulation analysis. We conducted a cross-tabulation analysis of HbA1c mean tertiles and VIM tertiles through the whole follow-up in relation to all-cause mortality (Table 3). Overall, as tertiles of HbA1c mean and HbA1c VIM increased, so did the incidence of all-cause mortality, and the third tertile of HbA1c VIM combined with the third tertile of HbA1c mean had the highest incidence of all-cause mortality (Fig. 1 and Supplementary Table 2).
Tertiles . | Intensive therapy . | Standard therapy . | ||||
---|---|---|---|---|---|---|
T1 of mean (≤6.22%) . | T2 of mean (6.23–6.74%) . | T3 of mean (>6.74%) . | T1 of mean (≤7.35%) . | T2 of mean (7.36–7.85%) . | T3 of mean (>7.85%) . | |
T1 of VIM (≤0.38 units) | — | 0.74 (0.36–1.52) | 1.18 (0.59–2.39) | — | 0.72 (0.34–1.54) | 0.99 (0.47–2.06) |
T2 of VIM (0.39–0.57 units) | 0.85 (0.44–1.66) | 1.11 (0.59–2.10) | 1.22 (0.64–2.33) | 0.93 (0.46–1.86) | 1.06 (0.53–2.10) | 1.31 (0.65–2.65) |
T3 of VIM (>0.57 units) | 1.19 (0.63–2.25) | 1.40 (0.76–2.60) | 2.05 (1.17–3.61) | 1.38 (0.72–2.66) | 0.97 (0.47–2.03) | 1.83 (0.94–3.55) |
Tertiles . | Intensive therapy . | Standard therapy . | ||||
---|---|---|---|---|---|---|
T1 of mean (≤6.22%) . | T2 of mean (6.23–6.74%) . | T3 of mean (>6.74%) . | T1 of mean (≤7.35%) . | T2 of mean (7.36–7.85%) . | T3 of mean (>7.85%) . | |
T1 of VIM (≤0.38 units) | — | 0.74 (0.36–1.52) | 1.18 (0.59–2.39) | — | 0.72 (0.34–1.54) | 0.99 (0.47–2.06) |
T2 of VIM (0.39–0.57 units) | 0.85 (0.44–1.66) | 1.11 (0.59–2.10) | 1.22 (0.64–2.33) | 0.93 (0.46–1.86) | 1.06 (0.53–2.10) | 1.31 (0.65–2.65) |
T3 of VIM (>0.57 units) | 1.19 (0.63–2.25) | 1.40 (0.76–2.60) | 2.05 (1.17–3.61) | 1.38 (0.72–2.66) | 0.97 (0.47–2.03) | 1.83 (0.94–3.55) |
Data are presented as HR and 95% CI. All models were adjusted for mean of HbA1c during visits, sex, history of CVD, diabetes duration, and baseline age, education, BMI, systolic and diastolic blood pressure, smoking, drinking, and fasting plasma glucose. T1, tertile 1; T2, tertile 2; T3, tertile 3. HbA1c of 6.22% converts to 44.5 mmol/mol, 6.74% to 50.2 mmol/mol, 7.35% to 56.8 mmol/mol, and 7.85% to 62.3 mmol/mol. The third tertile of HbA1c VIM had a significantly higher risk only in the third tertile of HbA1c mean in the intensive-therapy group (boldface value).
Taking the first tertile of mean and VIM as reference, multiple Cox regression analyses were performed to calculate the HRs for other tertiles of mean and VIM. Importantly, for all-cause mortality, the third tertile of HbA1c VIM had a significantly higher risk only in the third tertile of HbA1c mean in the intensive-therapy group but not in the standard-therapy group (Table 3).
Percentage of Uncontrolled HbA1c and All-Cause Mortality
Furthermore, we studied the HR for the risk of all-cause mortality in relation to the maximum and minimum HbA1c and the percentage of uncontrolled HbA1c during follow-up. In multiple Cox regression analyses, maximum and minimum HbA1c were both associated with all-cause mortality in the intensive-therapy group.
The HbA1c of 7.0% (53 mmol/mol) and 8.0% (64 mmol/mol) were taken as the threshold for uncontrolled HbA1c in the intensive-therapy and standard-therapy groups, respectively. Taking the percentage 0–19% as the reference, we found that increasing risks were shown with increased percentage of uncontrolled HbA1c in general for all-cause mortality in the intensive-therapy group but not in the standard-therapy group. The HR for all-cause mortality with ≥80% uncontrolled HbA1c in the intensive-therapy group was 1.67 (95% CI 1.11–2.52) (Supplementary Table 3).
Conclusions
In the current study, CV, VIM, and ARV in HbA1c during visits were analyzed as indices of variability. The key findings can be summarized in two points: 1) long-term visit-to-visit variability in HbA1c was a powerful predictor of all-cause mortality, even when accounting for mean HbA1c; and 2) HbA1c mean and HbA1c variability were both risk factors for all-cause mortality, and higher HbA1c variability combined with higher HbA1c mean conferred an increased risk for all-cause mortality in the intensive-therapy group. These findings raised the issue that visit-to-visit glycemic variability as well as higher levels of glycemia might be important risk factors for all-cause mortality and should be considered in efforts to intensively lower glucose among individuals with type 2 diabetes.
Several previous observational studies in type 2 diabetes have shown HbA1c variability is associated with micro- and macrovascular complications and mortality independently of the HbA1c level. In a prospective cohort study conducted in 2,103 patients with diabetes in outpatient clinics (22), with an average follow-up of 6.6 years and a median of 10 HbA1c measurements, HbA1c CV significantly predicted diabetic nephropathy, defined as increased urinary albumin-to-creatinine ratio (HR 1.03 [95% CI 1.01–1.04] for 1% increase of CV). Lin et al. (23) performed a cohort study in 3,220 Chinese with diabetes living in Taiwan, with an average follow-up of 4.4 years and more than eight measurements of HbA1c, and found that higher HbA1c CV (>13.4%) was associated with greater risk of diabetic nephropathy, defined by an estimated glomerular filtration rate of <60 mL/min/1.73 m2 (HR 1.58, 95% CI 1.19–2.11). Another study also performed in Taiwan in 881 patients with diabetes (24), with an average follow-up of 4.7 years and an average of 12 HbA1c measurements, found that higher HbA1c CV (>50th centile) was associated with greater risk of all-cause mortality (HR 1.06, 95% CI 1.01–1.11). Recently, a systematic review and meta-analysis showed that increased HbA1c variability was associated with an increased risk of renal disease (risk ratio 1.34, 95% CI 1.08–2.25), cardiovascular events (risk ratio 1.27, 95% CI 1.15–1.40), and mortality (risk ratio 1.34, 95% CI 1.18–1.53) in type 2 diabetes (25).
Some studies also examined the HbA1c variability in glucose-lowering clinical trials. There was a strong association between higher visit-to-visit glycemic variability and increased risk of mortality during the Veterans Affairs Diabetes Trial (VADT) that was independent of other traditional risk factors (26). Recent analysis of the Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation (ADVANCE) trial showed that HbA1c variability was positively associated with both total macrovascular events and major adverse cardiovascular events (27,28). That study was the first large-scale study to report the effects of variability in both HbA1c and fasting glucose on a variety of all-cause mortality in type 2 diabetes, including macrovascular events (27). However, the mean number of HbA1c measurements per participant was only five. In this ACCORD population, a mean number of nine measurements of HbA1c in a large-scale clinical trial study allowed assessment of the prognostic significance of HbA1c variability separately in the standard-therapy and intensive-therapy groups. In the intensively treated group, the cross-tabulation analysis shows very clear results: the greater the long-term variability, the higher the incidence of all-cause mortality, regardless of the HbA1c. In addition, this incidence increases with worsening HbA1c. Most interestingly in this study, we found that HbA1c variability combined with HbA1c mean conferred increased risk for all-cause mortality only in the intensive-therapy group, which may be one of the explanations of the higher risk for all-cause mortality among intensive-therapy participants in the ACCORD study.
Going “beyond HbA1c” is an important focus of the current investigation (14). Most traditional clinical trials focus on HbA1c as the definitive measure of efficacy of an intervention. However, mean HbA1c just reflects average glycemia derived without regard to glycemic variability, which is associated with higher risk of hypoglycemia on average (29). Visit-to-visit glycemic variability might be an important confounder for glycemia control and long-term prognosis and should be considered in the management of diabetes.
Our study should be interpreted within the context of its strengths and limitations. The strengths of our study include a large number of high-risk patients and a large number of HbA1c measures, which enable us to accurately calculate HbA1c variability. We used three variability indices, which enable us to study HbA1c variability more comprehensively. The variability index, VIM, can diminish the tight correlation between the CV and mean and was more suitable for the mean and variability cross-tabulation analysis.
The analyses also have limitations. The ACCORD study used HbA1c rather than glucose as the target and evaluation indices, and glucose was not recorded at every visit, so in this study we only focus on HbA1c. Because of the post hoc nature of the analysis and the highly selected study population, which included patients at high risk of CVD, the results should be extended to real-world studies including patients with type 2 diabetes having a variety of risk characteristics.
Conclusion
Our study confirmed that long-term visit-to-visit variability in HbA1c was a strong predictor of a variety of all-cause mortality. HbA1c variability combined with HbA1c mean conferred an increased risk for all-cause mortality in the intensive-therapy group in the ACCORD trial.
See accompanying article, p. 1169.
C.-S.S. and J.T. contributed equally to this work.
Article Information
Acknowledgments. The investigators acknowledge and thank the ACCORD investigators and the National Heart, Lung, and Blood Institute for conducting the trials and making data sets publicly available. Special thanks to Yonghua Xu and Min Hang (Shanghai Jiaotong University) for their data sets preparation.
Funding. This work was supported by Shanghai Pujiang Talents Plan (18PJ1407200), National Natural Science Foundation of China (81770418 and 81270935), National Key R&D Program of China (2016YFC1300103), and Shanghai Municipal Health Bureau Foundation (20114301).
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. C.-S.S. and J.T. contributed to the statistical analysis and wrote the manuscript. C.-S.S., J.T., Y.M., Y.C., and Y.Y. participated in acquisition, analysis, or interpretation of data. P.D.R., Z.T.B., and G.N. reviewed and edited the manuscript. All authors participated in critical revision of the manuscript for important intellectual content. G.N. is the guarantor of the 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.