Personalized blood glucose prediction: A hybrid approach using grammatical evolution and physiological models

Area of Science:

• Biomedical Engineering
• Computational Biology
• Endocrinology

Background:

• Patient variability and external factors complicate diabetes glucose prediction.
• Physiological models offer precision but lack flexibility; data-based models capture relationships but lose physiological meaning.

Purpose of the Study:

• To develop a hybrid prediction model for midterm blood glucose in type 1 diabetes.
• To improve personalized treatment and alarm-control applications.

Main Methods:

• A hybrid approach combining physiological insulin models and grammatical evolution.
• Utilized a penalizing fitness function based on the Clarke error grid to account for clinical harm.
• Trained models on 14-day data from 100 virtual patients using the UVA/Padova T1D simulator.

Main Results:

• Achieved an average of 98.31% of predictions within zones A and B of the Clarke error grid.
• Demonstrated the feasibility of midterm blood glucose predictions using personalized models with the specified grammatical evolution configuration.

Conclusions:

• Personalized midterm blood glucose prediction is feasible using the developed hybrid model.
• The hybrid approach shows potential for predicting short-term glucose, detecting sensor errors, and managing type 1 diabetes treatments.