Objective: To establish an accurate and robust calculation model for predicting hemoglobin A1c (HbA1c) for people with type 2 diabetes (T2D) by using the fewest discrete blood glucose values according to an irregular data set and propose an appropriate cost-effective and scientific scheme for routine blood glucose monitoring. Methods: By using two data sets obtained from 2017 to 2022, which involved 2432 people with T2D, ∼420,000 irregular blood glucose values, and 10,000 HbA1c values, multiple blood glucose monitoring schemes were designed and compared to find the optimal one. The data were structured and then fitted using a regularized extreme learning machine, and the results were evaluated on the basis of indicators such as mean absolute error (MAE), root mean square error, and the relevance analysis (R) value; the optimal scheme for routine blood glucose monitoring was determined by combining the accuracy and the cost and was compared with previous studies in terms of accuracy and stability. Results: Data fitting results for the chosen scheme: R = 0.8029 (P < 0.001), MAE = 0.3181% (95% confidence interval, 0.2666-0.3695%). Within the last 4 weeks before the prediction of HbA1c, a minimum of only seven fasting and seven postprandial blood glucose values are needed, of which are one fasting and one postprandial blood glucose values per 4 days. Compared with previous studies, the prediction model shows better accuracy and stability (P < 0.05), especially under the great glucose fluctuation group. Conclusion: A minimized calculation model for accurately and robustly predicting HbA1c using discrete self-monitoring of blood glucose data within 4 weeks for people with T2D has been established and provides a new reference for the design of a scheme for blood glucose monitoring. The diabetes care clinic of Peking University First Hospital (Registration Number: ChiCTR2300068139).
Keywords: Cost-effectiveness; Glucose monitoring; HbA1c; Prediction model; SMBG.