Estimation of Hemoglobin A1c from Continuous Glucose Monitoring Data in Individuals with Type 1 Diabetes: Is Time In Range All We Need?

Abstract

Objective: To bridge the gap between laboratory-measured hemoglobin A1c (HbA1c) and continuous glucose monitoring (CGM)-derived time in target range (TIR), introducing TIR-driven estimated A1c (eA1c). Methods: Data from Protocol 1 (training data set) and Protocol 3 (testing data set) of the International Diabetes Closed-Loop Trial were used. Training data included 3 months of CGM recordings from 125 individuals with type 1 diabetes, and HbA1c at 3 months; testing data included 9 months of CGM recordings from 168 individuals, and HbA1c at 3, 6, and 9 months. Hemoglobin glycation was modeled by a first-order differential equation driven by TIR. Three model parameters were estimated in the training data set and fixed thereafter. A fourth parameter was estimated in the testing data set, to individualize the model by calibration with month 3 HbA1c. The accuracy of eA1c was assessed on months 6 and 9 HbA1c. Results: eA1c was tracked for each individual in the testing data set for 6 months after calibration. Mean absolute differences between HbA1c and eA1c 3- and 6-month postcalibration were 0.25% and 0.24%; Pearson's correlation coefficients were 0.93 and 0.93; percentages of eA1c within 10% from reference HbA1c were 97.6% and 96.3%, respectively. Conclusions: HbA1c and TIR are reflections of the same underlying process of glycemic fluctuation. Using a model individualized with one HbA1c measurement, TIR provides an accurate approximation of HbA1c for at least 6 months, reflecting blood glucose fluctuations and nonglycemic biological factors. Thus, eA1c is an intermediate metric that mathematically adjusts a CGM-based assessment of glycemic control to individual glycation rates.

Keywords: Continuous glucose monitoring; Estimated A1c; Hemoglobin A1c; Hemoglobin clearance; Hemoglobin glycation; International Diabetes Closed-Loop Trial; Time in range.

FIG. 1.

(Left panel) Correlation between TIR and HbA1c at month 6 (dark gray solid squares) and month 9 (light gray open triangles) of the study. (Right panel) Correlation between eA1c and HbA1c at month 6 (i.e., 3-month postcalibration; dark gray solid squares) and month 9 (i.e., 6-month postcalibration; light gray open triangles) of the study. Fitted linear regression models are shown for the four analyses, with the equations and RMSE values reported in the legends. eA1c, estimated A1c; HbA1c, hemoglobin A1c; RMSE, root mean squared error; TIR, time in target range.

FIG. 2.

Effect of the amount of CGM data before model calibration on method performance. (Left panel) Correlation and R² between eA1c and HbA1c 3 months (dark gray solid squares) and 6 months (light gray open triangles) postcalibration. (Right panel) MAD between eA1c and HbA1c 3 months (dark gray solid squares) and 6 months (light gray open triangles) postcalibration. CGM, continuous glucose monitoring; MAD, mean absolute difference; R², coefficient of determination.