Objective: We show how continuous glucose monitoring (CGM) data can be analyzed using a three-level B-spline model, facilitating the estimation of inter-patient variability, within-patient inter-day variability, and measurement error. We propose methods for statistical comparison of glucose profiles among patient groups.
Methods: We applied a three-level random effects model using quadratic B-spline functions to analyze inter-patient and within-patient inter-day variations of the glucose trend. The estimated SD values of the glucose curves are time-dependent and were averaged over a 24-h period. We analyzed CGM data from 322 patients with type 1 diabetes, 223 patients with type 2 diabetes, and 86 subjects without diabetes using interstitial glucose levels measured every 5 min, for approximately 8 days per patient. We compared group-wide glucose profiles from the insulin pump-treated (n = 124) and multiple daily injection (MDI)-treated (n = 144) patients with type 1 diabetes.
Results: The average inter-patient SD values were 49 mg/dL, 43 mg/dL, and 15 mg/dL for type 1 diabetes patients, type 2 diabetes patients, and subjects without diabetes, respectively. The average within-patient, inter-day SD values were 67 mg/dL, 41 mg/dL, and 18 mg/dL, respectively. The residual SD values were 19 mg/dL, 14 mg/dL, and 8 mg/dL, respectively. We identified a statistically significant difference in glucose profiles during the morning between insulin pump-treated and MDI-treated type 1 diabetes patients.
Conclusions: B-spline models facilitate the analysis of CGM data and show that type 1 diabetes is associated with higher inter-day glucose variation than type 2 diabetes or being without diabetes. Pump therapy and MDI have different effects on glucose control during specific time periods.