Personalized Adaptive CBR Bolus Recommender System for Type 1 Diabetes

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Endocrinology

Background:

• Type 1 diabetes mellitus (T1DM) requires careful insulin management to control blood glucose levels.
• Estimating correct insulin dosage is challenging due to complex glucose metabolism, often leading to suboptimal glycemic control.
• Existing insulin recommendation systems may struggle with diverse patient situations and evolving physiological changes.

Purpose of the Study:

• To introduce PepperRec, a case-based reasoning (CBR) system for recommending bolus insulin doses in T1DM.
• To develop a system that handles a wide range of T1DM scenarios and missing data.
• To improve glycemic control by adapting recommendations to individual patient states and concept drift.

Main Methods:

• Utilized case-based reasoning (CBR) methodology for insulin dosage recommendations.
• Incorporated factors affecting glucose metabolism, including physical activity data.
• Developed methods for adapting past recommendations and retaining updated historical patient data to address concept drift.

Main Results:

• PepperRec demonstrated improved glycemic control in simulations using the UVA/PADOVA simulator with 33 virtual subjects.
• The system increased the time T1DM patients spent within their target glycemic range.
• PepperRec reduced time spent in hypoglycemia (low blood glucose) while maintaining or reducing time spent in hyperglycemia (high blood glucose).

Conclusions:

• PepperRec offers a robust approach to personalized insulin therapy for type 1 diabetes.
• The system effectively manages complex glucose metabolism and adapts to individual patient needs over time.
• PepperRec shows significant potential for improving clinical outcomes in T1DM management.