Applications of Variability Analysis Techniques for Continuous Glucose Monitoring Derived Time Series in Diabetic Patients

The analysis of time series by methods from nonlinear dynamics has enhanced understanding of functional dysregulations in various diseases but received yet less attention in diabetes. In this retrospective cross-sectional study of patients with type 1 and type 2 diabetes, we evaluated prominent indices of nonlinear and fractal dynamics and their application in the assessment of dysglycemia in diabetes. We used Poincare plots, multiscale entropy, and detrended fluctuation analysis of continuous glucose monitoring data from 177 subjects with type 1 (n = 22), type 2 diabetes (n = 143), and 12 nondiabetic control subjects. Spearman correlation analysis demonstrated mostly positive, moderate correlations between dynamical indices across different variability domains (p < 0.001). But the multiscale entropy index correlated inversely with SD1, SD2, the shape and area of the Poincare plots and with the exponents α1 and α2 of the detrended fluctuation analysis (r = -0.350 to -0.481, all p < 0.001). In type 2 diabetic patients the Poincare plot parameters and the two α exponents were significantly higher (p = 0.01 to < 0.001) than in nondiabetic subjects but were highest in the type 1 diabetes group. The multiscale entropy index decreased from the nondiabetic to the type 2 and type 1 diabetic group (9.16 ± 2.26 vs. 5.41± 1.84, p < 0.001). Switching from multiple daily insulin injections to continuous subcutaneous insulin infusions in a subgroup of type 1 diabetes patients decreased the dynamical indices, except for the shape of the Poincare plot fitting ellipse, which tended to increase. Overall, on multivariate regression analyses, the study revealed significant associations between the dynamical indices and quality of glycemic control, where the Poincare plot metrics SD1 (s = 0.59), SD2 (s = 0.72, and AFE s = 0.54 (all p < 0.001) emerged as more powerful predictors than glycosylated hemoglobin (s = 0.28, 0.19, and 0.32, respectively, all p < 0.001). As the strength of association between metrics of nonlinear dynamics and quality of glycemic control varies, one metric alone may insufficiently characterize glucose time series. The combination with linear parameters may improve clinical outcomes in diabetes.