Adaptive System Identification for Estimating Future Glucose Concentrations and Hypoglycemia Alarms

Area of Science:

• Biomedical Engineering
• Endocrinology
• Computational Biology

Background:

• Diabetes mellitus is characterized by significant glucose variability, leading to hyperglycemia and hypoglycemia.
• Predictive models for glucose concentration can aid in preventing dangerous glycemic events.
• Existing models may not fully capture inter- and intra-subject variations or adapt to glycemic disturbances.

Purpose of the Study:

• To develop and validate an adaptive time-series model for predicting glucose metabolism dynamics.
• To enhance glucose prediction by incorporating multi-sensor data for a comprehensive metabolic profile.
• To demonstrate the model's efficacy in early detection of hypoglycemic events.

Main Methods:

• Adaptive system identification using weighted recursive least squares for online parameter estimation.
• Change detection strategy to monitor model parameter variations and adapt to disturbances.
• Comparison of univariate models (continuous glucose monitoring) with multivariate models (multi-sensor body monitor data).

Main Results:

• The proposed adaptive model effectively captures dynamical changes in glucose metabolism.
• Multivariate models incorporating multi-sensor data showed improved predictive capabilities.
• The algorithm demonstrated successful early detection of hypoglycemia 30 minutes in advance in a real-life application.

Conclusions:

• Adaptive time-series modeling offers a robust approach for predicting glucose fluctuations in diabetes.
• Integration of multi-sensor data enhances the accuracy and reliability of glucose prediction models.
• The developed algorithm shows significant potential for real-time clinical decision support in diabetes management.