Estimation of future glucose concentrations with subject-specific recursive linear models.

Area of Science:

• Biomedical Engineering
• Data Science in Healthcare
• Diabetes Technology

Background:

• Accurate estimation of future glucose concentrations is vital for diabetes management, enabling timely interventions for hypoglycemia and hyperglycemia.
• Continuous glucose monitoring (CGM) provides high-frequency data, offering detailed insights into glucose variability.
• Developing reliable, subject-specific glucose prediction models is essential for enhancing diabetes care.

Purpose of the Study:

• To develop subject-specific glucose prediction models utilizing continuous glucose monitoring (CGM) data.
• To enhance the reliability and accuracy of glucose prediction for diabetes management applications.
• To validate the developed models under both stable and dynamic daily life conditions.

Main Methods:

• Utilized time-series analysis to create low-order linear models from individual patient CGM data.
• Integrated recursive identification and change detection methods for dynamic model adaptation to subject variability and glycemic disturbances.
• Evaluated prediction performance using glucose prediction error metrics and Clarke Error Grid Analysis (CG-EGA).

Main Results:

• Recursive identification significantly reduced prediction errors in glucose concentration estimation.
• The inclusion of a parameter change detection method further improved prediction accuracy.
• Clarke Error Grid Analysis demonstrated accurate readings in over 90% of cases.

Conclusions:

• A robust subject-specific glucose prediction strategy has been successfully developed.
• The integration of change detection enhances the accuracy of recursive glucose prediction algorithms.
• The proposed model, with its minimal parameter requirements, is suitable for portable devices, offering early alarms and closed-loop insulin pump control.