APEA: A Type 1 Diabetes Self-Management Ambient-AI Assistance Tool that Bridges Trajectory Prediction, Interactive...

Area of Science:

• Artificial Intelligence in Healthcare
• Diabetes Technology
• Pediatric Endocrinology

Background:

• Effective self-management of type 1 diabetes (T1D) necessitates advanced informatics tools.
• Current healthcare AI solutions often lack actionable recommendations and clear explanations, hindering clinical utility.

Purpose of the Study:

• To introduce APEA, an Ambient AI assistance tool for pediatric T1D self-management.
• To integrate multi-trajectory glucose prediction, interactive explanations, and adaptive intervention policies for real-time suggestions.

Main Methods:

• Utilized the T1DEXIP dataset comprising 262 pediatric T1D patients from multiple centers.
• Employed reinforcement learning for just-in-time adaptive intervention policy optimization.
• Integrated AI-driven glucose prediction and large language model-based explanations.

Main Results:

• APEA demonstrated significant improvements in glucose control outcomes.
• Achieved a 45% improvement over human management and a 69% improvement over infusion-pump management.
• Addressed implementation gaps in healthcare AI by providing predictive, prescriptive, and explanatory guidance.

Conclusions:

• APEA offers a transferable framework for chronic disease management requiring continuous, personalized interventions.
• The tool bridges the gap between potential health scenarios, actionable interventions, and clinical rationale.
• Despite limitations of sample size and class imbalance, APEA shows promise for enhancing T1D self-management.