Time to Hypoglycemia Prediction for Personalized Diabetes Care and Management

Hypoglycemia is a medical emergency characterized by low glucose levels (70 mg/dL or 3.9 mmol/L), which can be asymptomatic and difficult to predict in patients with type 1 diabetes (T1D). This study leverages the synergistic capabilities of advanced machine learning (ML) models and integrates noninvasive heart rate with continuous glucose monitoring to predict the time to onset of hypoglycemia events. Unlike prior research, this study extends prediction times and customizes personalized prediction models for early detection of hypoglycemia. In this study, we evaluate and develop the Fully Convolutional Networks (FCN) and Residual Networks (ResNet) models. Our results showed that FCN outperformed ResNet with 97% accuracy and robustness across different time classes, along with critical local feature coverage compared to the ResNet model's accuracy of 94%. These results highlight the implications of model architecture, validation techniques, and personalization in predicting hypoglycemia in T1D patients.