Key points are not available for this paper at this time.
The Hemoglobin Alc (HbAlc) test reflects the average amount of glucose accumulated in the blood over the last 2-3 months. The HbAlc is considered as one of the fundamental indexes for diabetes management. It is used to monitor long-term glycemic control, adjust therapy, assess the quality of diabetes care, and predict the future risk of complications. Predicting levels of HbAlc in advance holds a critical significance for maintaining long term health of diabetes patients. A higher than the normal value of the HbAlc increases the likelihood of diabetes-related complications. This study aims to predict the HbAlc levels 2-3 months in advance, which will facilitate early intervention and avoiding complications arise from diabetes. We propose novel feature extraction from continuous glucose monitoring (CGM) data using the fractional derivative, glucose variability, time in range, and wavelet decomposition methods. The highly correlated features were identified and used while developing the HbAlc prediction models. The developed framework was evaluated using the CGM data sourced from the Diabetes Research in Children Network (DirecNet). The ensembling of the random forest and extreme gradient boosting models coupled with the feature fusion obtained the best performance of low mean absolute error (MAE) 3.39 mmol/mol and a high coefficient of determination (R-squared) score of 0.81.
Islam et al. (Tue,) studied this question.