Evaluating the Adequacy of Coefficient of Variation and Standard Deviation as Metrics of Glucose Variability in Type 1 Diabetes Based on Data from the GOLD and SILVER Trials

Abstract

Objective:

Evaluate the adequacy of the coefficient of variation (CV) and standard deviation (SD) as metrics of glucose variability (GV) across mean glucose (MG) levels in individuals with type 1 diabetes.

Methods:

Data from the GOLD and SILVER trials were analyzed. Glucose metrics were derived from continuous glucose monitoring (CGM). Generalized estimating equations were used to assess the relationship between SD and MG, considering intraindividual correlations. Nonlinear associations were evaluated using restricted cubic splines, and glucose values outside the CGM detection range (<2.22 mmol/L and >22.2 mmol/L) were handled using a censored Gamma model.

Results:

In total, 158 individuals with an MG of 10.6 (SD 1.7) mmol/L were included. The SD of glucose values exhibited a nonlinear relationship with the MG during CGM and self-monitoring of blood glucose (SMBG) (both P < 0.001 vs. linear model). The lack of fit of the constant CV model was most distinct at high glucose levels >12 mmol/L. During SMBG, a 25% reduction in MG from 12 to 9 mmol/L was associated with a 16% (95% confidence interval [CI] 10%–21%) reduction in the SD of glucose values. Similar associations were observed during CGM. This deviation was attributed to the censoring of glucose values outside the detection range. After adjusting for censoring, the lack of fit was resolved. When transitioning from SMBG to CGM, the ordinary CV and SD underestimated the treatment effect on GV by 30% and 27%, respectively, compared to estimates adjusted for censoring. Similarly, ordinary CV underestimated the treatment effect by 11% compared with CV adjusted for the nonlinear SD-MG relationship in the GOLD study.

Conclusion:

The SD of glucose values does not increase linearly with the MG during glucose-lowering therapy, suggesting that CV is not an optimal measure of GV. After adjusting for censored glucose values, CV remains reliable. Alternatively, nonlinear SD adjustments relative to MG effectively evaluate glucose-lowering therapies’ impact on GV.