Visit-to-Visit HbA_1c Variability Is Associated With Cardiovascular Disease and Microvascular Complications in Patients With Newly Diagnosed Type 2 Diabetes

Although there is considerable evidence that intensive blood glucose normalization reduces the risk of both cardiovascular events and microvascular complications of diabetes (1–3), the effects were heterogeneous between trials. For example, the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial was terminated prematurely due to significantly elevated mortality and cardiovascular events (4), suggesting that the near normalization of blood glucose should not be the only target of diabetes treatment. Glycemic variability is one factor that may explain these differences in cardiovascular outcomes.

Glycemic variability can be measured as either the glucose fluctuation within a day or the long-term visit-to-visit variability. The latter has been recently investigated in several studies, although the metrics and definition of the variability measure were inconsistent (5). Most studies evaluating HbA_1c variability using the SD or the coefficient of variation (CV) of HbA_1c suggested that these measures were associated with all-cause mortality and the development of the adverse outcomes of diabetes, after adjusting for the average HbA_1c (6–11). However, neither SD nor CV of HbA_1c can be easily interpreted in clinical practice. Recently, Forbes et al. (12) developed a new scale, namely the HbA_1c variability score (HVS) in the current study, to define HbA_1c variability. The HVS indicates how frequently HbA_1c rises or decreases by >0.5% (5.5 mmol/mol), which is in line with the SD and CV of HbA_1c but clinically more translatable (as it can be interpreted as the percentage of total HbA_1c measures that vary by >0.5% or 5.5 mmol/mol) (6,12). However, the HVS has not been widely used among studies of HbA_1c variability, with previous studies using this scale only focusing on the elderly population without diabetes and evaluating mainly mortality as an outcome (6,12). It is unclear whether HVS is associated with microvascular complications of diabetes and whether the increased cardiovascular risk described could be extended to real-world patients with type 2 diabetes. In this study, we aimed to investigate the association between visit-to-visit HbA_1c variability and both cardiovascular diseases and microvascular complications in a large population database of patients with newly diagnosed type 2 diabetes.

The population was selected from patients from Tayside and Fife in the Scottish Care Information–Diabetes Collaboration (SCI-DC), the electronic health record system used in Scotland for patients with diabetes. The patients were included if they 1) were diagnosed with type 2 diabetes, 2) had their first HbA_1c measurement within 1 year from diagnosis of diabetes, 3) were >40 years old when first diagnosed with diabetes, 4) did not experience any study outcome before or within 3 years since diagnosis of diabetes, and 5) had at least five records of HbA_1c measurement between diagnosis of diabetes and the first episode of the study outcome. Patients were excluded where data were incomplete (for details see the Supplementary Techniques). Data provision and linkage were performed at the University of Dundee Health Informatics Centre (www.dundee.ac.uk/hic), with analyses of anonymized data performed in an ISO 27001– and Scottish Government–accredited secure safe haven. Health Informatics Centre standard operating procedures have been reviewed and approved by the National Health Service (NHS) East of Scotland Research Ethics Service, and consent for this study was obtained from the NHS Fife Caldicott Guardian.

The BMI, estimated glomerular filtration rate (eGFR), and smoking status at baseline were captured from the medical record within 1 year from the diagnosis of diabetes (for details see the Supplementary Techniques). The follow-up was defined by the first event of outcome or the last measurement of HbA_1c before 24 April 2017 in the case of no events. The Charlson Comorbidity Index (CCI) was calculated using the ICD-9 and ICD-10 code within the year after the diagnosis of diabetes (13), while we specifically removed the items of diabetes and cardiovascular events, which were overlapping with our population or outcomes.

To avoid the interaction between the HbA_1c variability parameter and the frequency of HbA_1c measurement and to better fit clinical practice, the HbA_1c variability was evaluated using HVS, which was adopted from a recent publication (12). In brief, HVS is the number of measures within an individual where the HbA_1c has changed by >0.5% (5.5 mmol/mol) from the value prior, as a percentage of the total number of HbA_1c measures between the diagnosis of diabetes and the outcome of interest for that individual (Supplementary Fig. 1). To avoid the impact of multiple HbA_1c measures in a short space of time, we allocated one HbA_1c measure for every 3-month period, using the median of all the HbA_1c measures within that time. The resulting variability measure is termed the binned HVS (b-HVS). We also calculated the time-weighted average HbA_1c, using the area under the curve of HbA_1c from the diagnosis of diabetes to the first event divided by the duration.

We examined 10 outcomes of interest including the following: major adverse cardiovascular events; all-cause mortality; atherosclerotic cardiovascular (ASCV) death; hospitalization or death from coronary artery disease, ischemic stroke, or heart failure; observable background diabetic retinopathy; diabetic peripheral neuropathy; diabetic foot ulcer (DFU); and the new onset of chronic kidney disease (CKD). If the event of interest occurred within the first 3 years from the diagnosis of diabetes, the patient was excluded from the analysis of that outcome, to avoid the outcome occurring close to diagnosis before the HVS could be defined, when the outcome would be unlikely to be related to the HVS. For full definitions of the end points, see the Supplementary Techniques.

The categorical variables were described using frequency and percentage. The continuous variables were described using means and SDs if normally distributed or median interquartile range (IQR) if not. Cox proportional hazards model was used to assess the association between HbA_1c variability and each of the outcomes. The association of the adverse outcome with the HVS categories (≥0 to ≤20, >20 to ≤40, >40 to ≤60, >60 to ≤80, and >80, with the ≥0 to ≤20 as reference) was adjusted for sex, index age, calendar year, Scottish Index of Multiple Deprivation quintiles, ever smoking, hypertension at baseline, BMI at baseline, HDL cholesterol at baseline, eGFR at baseline, antiplatelet therapy at baseline, and CCI (≥1 vs. 0). We used Survival::cox.zph Pack in R to test the proportional hazards assumption for Cox regression models (14) for all our models. We considered the proportional hazards assumptions to be violated if the global P value was <0.01. Because of the violation of proportional hazards assumptions, the stage of CKD (stage 1 or 2) at baseline rather than the eGFR at baseline was stratified in the analysis of the new onset of CKD. Five subgroup analyses were introduced based on the age (<65 vs. ≥65 years), sex, BMI at baseline (>30 vs. ≤30 kg/m²), time-weighted mean HbA_1c (>7% vs. ≤7% or >53 vs. ≤53 mmol/mol), and treatment at baseline (medication/insulin treated vs. lifestyle intervention only). Five sensitivity analyses were performed for each outcome by 1) adjusting for time-weighted average HbA_1c; 2) using the b-HVS instead of HVS; 3) using the HVS based on the HbA_1c measurement solely focusing on the first 3 years after diagnosis of diabetes, prior to the occurrence of any event; 4) using the individual-level SDs of the HbA_1c instead of the HVS; and 5) using individual-level CVs of HbA_1c instead of the HVS. Analyses were undertaken in SAS 9.4 (SAS Institute Inc., Cary, NC) and RStudio for Windows (R version 3.2.5).

As shown in Fig. 1, among the 79,569 patients with type 2 diabetes identified in the population, we included 21,352 patients for further analysis. The average age was 63.3 ± 11.1 years when recruited, and 54.6% were male. The median follow-up duration was 6.8 years (IQR 4.6–11.2). The mean HbA_1c at baseline was 7.7% ± 2.0% (60.7 ± 21.4 mmol/mol), and the median number of HbA_1c measurements throughout the study period was 12 times (IQR 8–19) during the follow-up duration. Supplementary Table 1 shows the baseline patient characteristics for those included for each analysis of outcomes, and Table 1 shows how the baseline characteristics differ across the HVS categories. Sixty-two percent of the patients have an HVS ≤40%; 12.5% have an HVS >60%. As expected, an increasing HVS is associated with younger age of diagnosis, higher BMI, and more intensive diabetes treatment, including greater insulin use.

As shown in Fig. 2, between 13,111 and 19,883 patients were involved in the analyses of each outcome. In comparisons with the reference (lowest HVS category, ≥0 to ≤20), patients with HVS >60 were associated with increased risks of all outcomes in a fully adjusted Cox proportional hazards model. For example, those with HVS >80 to ≤100 had an increased risk of the following: major adverse cardiovascular events (hazard ratio 2.38 [95% CI 1.61–3.53]), all-cause mortality (2.4 [1.72–3.33]), ASCV death (2.4 [1.13–5.11]), coronary artery disease (2.63 [1.81–3.84]), ischemic stroke (2.04 [1.12–3.73]), heart failure (3.23 [1.76–5.93]), DR (7.4 [3.84–14.27]), diabetic peripheral neuropathy (3.07 [2.23–4.22]), DFU (5.24 [2.61–10.49]), and CKD (3.49 [2.47–4.95]).

Given the association between HVS and HbA_1c, we first undertook a sensitivity analysis, including time-weighted average HbA_1c, from diagnosis to event in the models (Fig. 3). The results were similar for most outcomes other than retinopathy where the association of HVS was diminished when adjusting for the time-weighted average HbA_1c.

When comparing the subgroups with time-weighted average HBA_1c higher or lower than 7% (53 mmol/mol), there was a stronger association between HVS and coronary artery disease, ischemic stroke, and progression to CKD in patients with time-weighted average HbA_1c <7% or 53 mmol/mol (Supplementary Fig. 5). Other subgroup analyses were undertaken based on age (Supplementary Fig. 2), sex (Supplementary Fig. 3), obesity at baseline (Supplementary Fig. 4), and treatment at baseline (Supplementary Fig. 6) and did not show significant differences in the trend of the association (except in the case of a very small sample size). Using b-HVS instead of HVS also showed consistent results in all outcomes (Supplementary Fig. 7). However, the sensitivity analysis using the first 3-year HVS suggested a weaker association compared with the main analysis (Supplementary Fig. 8). The sensitivity analysis using the individual-level SD (Supplementary Fig. 9) and CV (Supplementary Fig. 10) of HbA_1c showed a similar pattern of risk for most outcomes but not ischemic stroke for SD and CV or diabetic retinopathy for CV where weaker associations were observed.

To our knowledge, this is the first population-based study to investigate the association between the visit-to-visit HbA_1c variability and comprehensive end points, including cardiovascular events and the microvascular complications of diabetes, in patients with newly diagnosed type 2 diabetes independent of the time-weighted average HbA_1c.

Our study showed clear elevated risks of adverse events in the ∼12.5% of patients with an HVS >60 (meaning those with 60% of HbA_1c measurements increased or decreased by >0.5% [5.5 mmol/mol] compared with the last measurement) after diagnosis of diabetes adjusted for their time-weighted average HbA_1c. The results were consistent with previous studies based on trial (15,16) and observational data sets (6–12,17). Our results indicate that frequent fluctuations of HbA_1c of patients with diabetes may be an independent risk factor for poor prognosis and more stable HbA_1c control may benefit the patients in clinical practice, although it should be emphasized that our results are observational and causal inference cannot be made. Of note, a recent analysis based on the Veterans Affairs Diabetes Trial (VADT) (16) suggested that higher HbA_1c variability was associated with the increased risk of cardiovascular events in the group of intensive glycemic control but not the standard control. It suggested that the increased HbA_1c variability may neutralize the cardiovascular benefits of the sustained 1.5% (16.4 mmol/mol) HbA_1c reduction during the study period (18). We undertook a subgroup analysis looking at HVS in those with good and poor average HbA_1c. It was interesting to note that the HVS association with ASCV events was greater in those with good HbA_1c, in keeping with the VADT finding. However, we need to interpret these results with caution as we cannot account for treatment intensity during the study period.

We previously reported that patients with high variability in HbA_1c have high cardiovascular risk at baseline (19), and thus the association of HbA_1c variability with risk may be not a feature of the HbA_1c variability per se but, rather, a marker of this baseline difference in patient characteristics. In this current study, we adjusted comprehensively for baseline differences in cardiovascular risk, although we acknowledge that there could be residual confounding. It is interesting to note that in the sensitivity analysis where we restrict our analysis to defining HbA_1c variability only in the first 3 years of HbA_1c measures, the association with micro- and macrovascular outcomes is diminished. This suggests that the HbA_1c variability may continuously contribute to the clinical adverse end points beyond the first 3 years and, therefore, that the risk can be less attributable to baseline differences in patient characteristics and more attributable to the HbA_1c variability per se. As a recent study suggested that HbA_1c variability is associated with the quality of patient care (20), it also suggests that it is never too late to reduce the HbA_1c variability in clinical practice. Although infeasible in the current analysis, it would also be interesting to evaluate HbA_1c variability on different antidiabetic treatments to see if reduced variability can explain some of the improved outcomes with some of these agents.

Although we cannot attribute poor prognosis to the HbA_1c variability per se, some underlying mechanisms may explain the association observed in our study. Although oxidative stress is suggested to be the explanation of an association between short-term glycemic variability and adverse outcomes (5), it is not clear whether this is increased in patients with high visit-to-visit HbA_1c variability. An alternative may relate to accumulated epigenetic modification induced by both high and low glycemia (21). Another explanation may simply relate to increased hypoglycemia in these individuals, since some studies suggest that high HbA_1c variability is linked to increased risk of severe hypoglycemic episodes (22) and patients admitted to the hospital due to hypoglycemia have higher HbA_1c variability (23). It will be valuable if a further study could address the frequency of overall and severe hypoglycemia among patients with different HbA_1c variability.

The strengths of our study are clear. First, all the included patients were tracked with their HbA_1c measurement from the diagnosis of diabetes, so there is no period of the patients’ diabetes journey that is not captured. Second, we comprehensively studied 10 clinically important outcomes, including all-cause mortality, cardiovascular events, and major microvascular complications of diabetes, and showed consistent results across these micro- and macrovascular end points. Third, our results were confirmed by a series of subgroup analyses and sensitivity analyses including adjusting for the time-weighted average HbA_1c from the diagnosis of diabetes. Fourth, our study was based on the real-world data of diabetes care in Scotland, making these results directly translatable to clinical practice. Finally, we used the HVS, which we feel is much more clinically tractable than SD or CV. Although SD and CV reflect the dispersion trend of the HbA_1c measures in an individual, they are no more than clinically meaningless statistical parameters. When considering the HVS, the clinicians can review the HbA_1c profile for an individual—those where >60% of measures vary by >0.5% are at high risk.

The study does have limitations. First, as a retrospective cohort study, uncorrected confounding could be possible, and individuals with higher HbA_1c variability may also be at higher cardiovascular risks of other causes (18). Nevertheless, we used Cox proportional hazards models to minimize the possible known confounding factors, including CCI, smoking status, and social deprivation, and used a series of subgroup analyses and sensitivity analyses to confirm our findings to be robust. Second, we did not adjust for or evaluate the contribution of hypoglycemia, which has been reported to be associated with HbA_1c variability (15), in the association between the HbA_1c variability and outcomes because of the limitation of the data. Third, the median follow-up duration of the study was 6.8 years, and this will limit the total incident outcomes. The need to only include patients with newly diagnosed diabetes and other inclusion criteria does limit the total follow-up time in this study population. This relatively short median duration does reduce the number of long-term outcome events, especially for retinopathy and DFU. Studies with a longer follow-up duration in larger populations would be of value.

In conclusion, our study shows that higher HbA_1c variability from the diagnosis of diabetes is independently associated with increased risks of all-cause mortality and major cardiovascular and microvascular complications of diabetes.

Funding. This study was funded by the International Visiting Program for Excellent Young Scholars of Sichuan University and the Department of Science and Technology of Sichuan Province (2019YFH0150). S.L. also received grants from the National Natural Science Foundation of China (81400811 and 21534008) and the 1.3.5 Project for Disciplines of Excellence, West China Hospital, Sichuan University (ZYGD18022). K.Z. was supported by the National Key R&D Program of China (2018YFC200100X). E.R.P. holds a Wellcome Trust New Investigator Award (102820/Z/13/Z).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.L., I.N., L.D., and S.H. performed the statistical analyses. S.L., K.Z., and E.R.P. conceived the study. S.L. and E.R.P. drafted the manuscript. E.R.P. 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.

Prior Presentation. Parts of this study were presented in poster form at the 55th Annual Meeting of the European Association for the Study of Diabetes, Barcelona, Spain, 16–20 September 2019.