Fasting Glucose Variation Predicts Microvascular Risk in ACCORD and VADT

Jin J. Zhou; Juraj Koska; Gideon Bahn; Peter Reaven


J Clin Endocrinol Metab. 2021;106(4):1150-1162. 

In This Article


Study Design and Participants

The Action to Control Cardiovascular Risk in Diabetes trial (ACCORD) was a double-blinded, 2-by-2 factorial, randomized controlled, parallel treatment trial in which 10 251 participants were assigned to receive intensive treatment targeting HbA1c concentration of less than 6.0% (42.1 mmol/mol) or standard treatment targeting HbA1c of 7.0% to 7.9% (53-62.8 mmol/mol)—as well as to distinct blood pressure and lipid interventions arms.[3] The ACCORD study included participants with T2D, HbA1c concentrations of 7.5% (58.5mmol/mol) or more, and who were aged 40 to 79 years with a history of cardiovascular disease or 55 to 79 years with evidence of significant atherosclerosis, albuminuria, left ventricular hypertrophy, or at least 2 risk factors for cardiovascular disease (dyslipidemia, hypertension, smoking, or obesity). During the study, FPG concentrations were measured every 4 months in the initial year, then annually up to a maximum of 84 months. Within 4 months after randomization, the median glycated hemoglobin level had fallen from 8.1% (65 mmol/mol) at baseline to 6.5% (47.5 mmol/mol) in the intensive therapy group and to 7.5% (58.5 mmol/mol) in the standard therapy group. Details of the design and principal results of ACCORD trial were reported previously.[3,12] Our analysis used all in-study FPG measures through the full ACCORD study (Supplementary Figure 1[13]).

The Veteran Affairs Diabetes Trial (VADT) was a randomized trial that enrolled 1791 military veterans (mean age, 60.4 years) who had a suboptimal response to therapy for T2D (HbA1c > 7.5% or 58.5mmol/mol) to receive either intensive or standard glucose control. The design and the principal results have been described previously.[2] Following an established algorithm, the 2 groups were treated with similar medications (but different doses) with a goal of the intensive treatment group of achieving near normal glucose control and an absolute difference in HbA1c of >1.5% between treatment groups. HbA1c and fasting glucose were measured every 3 months up to a maximum of 84 months. At 3 months into the trial median HbA1c levels had decreased in both groups and had stabilized by 6 months, with a level of 8.4% (68 mmol/mol) in the standard-therapy group and 6.9% (52 mmol/mol) in the intensive-therapy group. Similar temporal patterns were seen with FPG.

Primary and Secondary Outcomes

Our primary microvascular outcome for both studies reflects a composite endpoint of advanced kidney and eye disease. In ACCORD, this composite microvascular outcome was its primary outcome and was defined as the development of end-stage renal disease (ESRD, ie, initiation of dialysis or a rise of serum creatinine to 3.3 mg/dL [292 μmol/L]), or retinal photocoagulation or vitrectomy to treat retinopathy. These serious outcomes were previously defined and described as the key microvascular outcomes by ACCORD investigators and were referred to as Neph-3 and Eye-1, respectively.[3] As the VADT did not collect the same renal clinical outcomes as in ACCORD, we defined nephropathy in VADT in a similar, but not identical, fashion as 2 consecutive values of serum creatinine more than 3.3 mg per deciliter or with consecutive values of glomerular filtration rate (GFR) of less than 30 mL per minute.[2] Time to event reflected the first of the 2 consecutive lab values meeting the criteria for either outcome. Retinopathy in the VADT was defined by retinal photocoagulation or vitrectomy, as within ACCORD. The GFR was estimated using the Modification of Diet in Renal Disease (MDRD) equation as described previously.[14] Secondary outcomes for each study included the individual components of the primary composite outcome, ie, time to renal failure (nephropathy) and retinal photocoagulation or vitrectomy (retinopathy).[3] To conduct a meta-analysis of the 2 studies, we redefined the nephropathy outcome in ACCORD by the definition used in VADT in order to make the outcome consistent and further reduce the study heterogeneity.

In addition to the 3 above key microvascular outcomes (eye, renal, or combined events) in ACCORD, there were several other prespecified microvascular outcomes in ACCORD for kidney function and diabetes eye complications. The additional predefined ACCORD microvascular outcomes were also explored for a more comprehensive assessment of less advanced or broader combinations of microvascular outcomes and included:

  • Neph-1: Doubling of baseline serum creatinine or more than 20 mL/min per 1.73 m2 decrease in estimated GFR.

  • Neph-2: Development of macroalbuminuria (urine albumin:creatinine ratio ≥33.9 mg/mmol)

  • [Neph-3: Defined above, as part of the primary composite outcome]

  • Neph-4: Development of Neph-1, Neph-2, or Neph-3

  • Neph-5: Development of microalbuminuria (urine albumin:creatinine ratio ≥3.4 mg/mmol)

  • [Eye-1: Defined above as part of primary composite outcome]

  • Eye-2: Eye surgery for cataract extraction

  • Eye-3: Three-line change in visual acuity

  • Eye-4: Severe vision loss (Snellen fraction <20/200)

A detailed description of the prespecification of outcomes in ACCORD was reported previously.[3]

Fasting Plasma Glucose Variability

Commonly used measures of visit-to-visit glucose variability include SD, coefficient of variation (CV), variability independent of mean (VIM), and average real variability (ARV).[9,15,16] We selected CV and ARV for this analysis as they appear to be complementary measures of variability as previously published.[17] CV measures the spread of the data over time vs while ARV measures smoothness of data over time. Definitions of these 2 variability measures have been described previously[17] and are provided in the Supplementary Table 1.[13] Mean FPG levels revealed substantial treatment group separation was achieved over the initial 4 months of ACCORD and the initial 6 months of VADT. This pattern persisted during the remaining duration of the trials. Therefore, observations from (including) the fourth month and beyond in ACCORD and from (including) the sixth month and beyond in the VADT were used for FPG variability calculation to exclude the rapid change of FPG resulting from the glucose-lowering trial designs (Supplementary Figure 1[13]).

Statistical Analysis

Data are expressed as means (SD) for continuous variables or as numbers and percentages for categorical variables. Differences between patients who did and did not develop an event were analyzed using the Wilcoxon test for continuous variables and the χ2 test or Fisher exact test, as appropriate, for categorical variables shown in Supplementary Table 2 and Supplementary Table 3.[13]

Multivariable analyses were performed by Cox proportional hazard models. We evaluated risk of fasting glucose variability while controlling for average glycemic control (defined as cumulative average of HbA1c). Both were included as continuous and time-dependent covariates[18] in the Cox proportional hazard models. This process dynamically matched the risk variables and time of outcome event, so that we do not use any measures after the event has happened.[19] The proportionality of all model predictors was confirmed in plots of Schoenfeld residuals over time. To ease interpretation of statistical models, hazard ratios (HR) for all variables of glycemic control were standardized to a change of one SD. Nonnormally distributed variables, such as CV, were log-transformed to approach normal distribution. Analyses were performed after adjusting for (i) Model-1: age only; (ii) Model-2: age and covariates reflecting significant baseline differences in characteristics (including blood pressure and lipids treatment arms in ACCORD study) between those who did and did not develop microvascular events during the study (seeTable 1 legend and Supplementary Tables 2 and 3[13]); and (iii) Model-3: covariates in Model-2 plus cumulative mean of HbA1c reflecting average glycemic control to clarify whether variability measures provided information beyond standard glycemic measures.

Five sensitivity analyses were conducted. We first excluded patients with advanced baseline eye or kidney disease. Baseline eye disease was defined as "cataract removal," "retinal laser photocoagulation for diabetic retinopathy," "laser for cataract capsule," or "vitrectomy for diabetic retinopathy" for either eye. Baseline kidney disease was defined as eGFR <45 mL per minute or macroalbuminuria, ie, urine albumin:creatinine ratio ≥33.9 mg/mmol. We also adjusted for "insulin use" in the model to assess whether the risk of glucose variability is independent of insulin use. Non-insulin users were those who reported "No" at all time points; all remaining participants were considered insulin users. We increased the minimum number of glucose measures per individual for inclusion to 5 which preserved 80% of sample. As the intensive-therapy arm was aborted prematurely because of increased mortality in ACCORD,[4] we also tested glucose measures only up to cessation of intensive treatment in the ACCORD study. Finally, we assessed the contribution of adverse lifestyle behaviors on the effect of glucose variability, within a subset sample (n = 2034) in ACCORD, where smoking, dietary patterns, and activity data were available. A summary score of unhealthy behaviors, from 0 to 3, was generated from these factors. A similar score was also generated in VADT, as previously described.[20] As these additional adverse behavior adjustments did not modify the results, they were not included in the presented models. We also used study visit HbA1c measures to capture glycemic variability. However, HbA1c showed a generally weaker association with primary and second microvascular outcomes than did variability in fasting glucose (results not shown).

We also pooled the results from the 2 studies to provide a more precise and generalizable overall effect. As random-effects meta-analysis[21] relies on the estimates of between study variance (which is ideally conducted in more than 2 studies) to apply it correctly,[22] we instead used fixed-effects meta-analysis to integrate the results from the 2 studies for all 3 outcomes (composite, renal, and eye) using the same outcome definitions. Statistical heterogeneity of the 2 studies was assessed with the I2 statistic,[23] where values of 30% to 60% represent a moderate level of heterogeneity. Meta-analyses were used to pool the risk of glucose variability from the 2 cohorts as well as to assess the pooled differential risks between the 2 treatment arms (intensive vs standard glucose control). We tested the null hypothesis (that the difference of glucose variability risks between treatment groups is zero) using the following approach: first, for each trial, we calculated a trial-specific interaction HR by adding an interaction term between treatment group and glucose variability measures in the Cox proportional hazard models; second, we combined these trial-specific interaction HRs across trials using a fixed-effects model. For the stratified analysis and stratified meta-analysis, we derived the HRs of glucose variability for all 3 microvascular outcomes, and their 95% CIs, in each study separately in the intensive and standard treatment group. We then calculated the pooled HR of glucose variability using fixed-effect meta-analysis models.

All statistical analyses were performed using R version 3.5.3 ( Meta-analyses were conducted by a R package, "meta".[24] A 2-sided P < 0.05 was considered statistically significant.