Multi-Metric Subgroup Analysis for Glucose Forecasting in Type 1 Diabetes.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Endocrinology

Background:

• Type 1 Diabetes (T1D) management is complex due to intricate glucose regulation.
• Accurate short-term glucose prediction is essential for effective T1D management and reducing complications.
• Existing predictive models often struggle with individual patient variability.

Purpose of the Study:

• To develop and evaluate XGBoost-based predictive models for short-term glucose concentration in T1D patients.
• To compare the performance of population-based versus personalized glucose prediction models.
• To investigate the impact of patient subgroups, including glucose variability, on prediction accuracy.

Main Methods:

• An XGBoost model was developed for glucose prediction at 15, 30, and 60-minute horizons.
• Bayesian optimization was employed to fine-tune the XGBoost model parameters.
• Model performance was assessed across diverse patient subgroups based on glucose patterns and variability.

Main Results:

• The population-based model demonstrated superior performance over personalized models across all prediction horizons.
• Achieved Root Mean Square Error (RMSE) values were 17.39 mg/dL (15 min), 27.28 mg/dL (30 min), and 40.69 mg/dL (60 min).
• Patients with higher glucose variability exhibited reduced prediction accuracy, indicating a challenge for current models.

Conclusions:

• Population-based models show significant potential for enhancing glucose prediction accuracy in T1D.
• Combining population-based approaches with subgroup-specific adjustments can further improve glycemic control.
• Accurate glucose prediction is vital for optimizing insulin therapy and improving patient outcomes in T1D management.