The association between self-monitoring of blood glucose and HbA1c in type 2 diabetes

Abstract

Aims: Fasting capillary blood glucose (FCG) and postprandial capillary blood glucose (PCG) both contribute to HbA1c in diabetes. Due to the collinearity between FCG and PCG, the HbA1c prediction model could not be developed with both FCG and PCG by linear regression. The study aimed to develop an HbA1c prediction model with both FCG and PCG to estimate HbA1c in type 2 diabetes.

Methods: A total of 1,642 patients with type 2 diabetes who had at least three FCG and three PCG measurements in the past 3 months were enrolled in the study. The mean of FCG (MEAN_FCG) and PCG (MEAN_PCG) were calculated for each patient. The patients were randomized into exploratory and validation groups. The former was used for developing HbA1c prediction models and the latter for performance evaluation.

Results: The new HbA1c prediction model using ridge regression expressed as HbA1c (%) = 0.320×MEAN_FCG (mmol/L) + 0.187×MEAN_PCG (mmol/L) + 2.979, R² = 0.668. Compared to linear regression models developed with FCG, PCG, fasting plasma glucose (FPG), and 2-hour postprandial plasma glucose (2-h PPG), respectively, the new HbA1c prediction model showed the smallest mean square error, root mean square error, mean absolute error. The concordance correlation coefficient of the new HbA1c prediction model and the linear regression models with MEAN_FCG, MEAN_PCG, FPG or 2-h PPG were 0.810,0.773,0.749,0.715,0.672.

Conclusion: We have developed a new HbA1c prediction model with both FCG and PCG, which showed better prediction ability and good agreement.

Keywords: HbA1c; fasting capillary blood glucose; postprandial capillary blood glucose; ridge regression; type 2 diabetes.