Glucose estimation and prediction through meal responses using ambulatory subject data for advisory mode model...

Area of Science:

• Biomedical Engineering
• Control Systems
• Diabetes Technology

Background:

• Closed-loop glucose control for type 1 diabetes mellitus (T1DM) faces challenges in adapting to daily activities like meals, stress, and exercise.
• Model-based control requires algorithms balancing simplicity for online prediction with complexity for handling physiological variations.

Purpose of the Study:

• To develop and validate a model-based control strategy for T1DM glucose management applicable across various daily activities.
• To enhance online glucose prediction and control by adapting to meal disturbances and insulin sensitivity variations.

Main Methods:

• Linearized a modified Bergman minimal model for Kalman filter (KF) state estimation in T1DM subjects.
• Developed parameter augmentation methods for online adaptation and assessed model deterioration for prediction horizon determination.
• Validated model predictive control (MPC) strategies using advisory mode simulations.

Main Results:

• Evaluated 20 days of continuous glucose data from three T1DM subjects, including 97 meals.
• Achieved a 45-minute glucose prediction horizon using a constant parameter model with meal announcement.
• Demonstrated that parameter adaptation was crucial for maintaining prediction horizon without meal announcements, effectively capturing glucose disturbances.
• Advisory mode MPC tuning resulted in a controller that responded effectively to meal disturbances.

Conclusions:

• Successfully estimated and predicted glucose levels using a KF based on a modified Bergman model.
• Parameter adaptation enabled closed-loop control implementation even without meal announcements.
• The developed state estimation and model validation framework provides a foundation for advisory mode MPC in T1DM management.