APEA: A Type 1 Diabetes Self-Management Ambient-AI Assistance Tool that Bridges Trajectory Prediction, Interactive Explanation, and Just-in-Time Adaptive Intervention Action.

Abstract

Helping patients self-managing diseases like type 1 diabetes (T1D) requires informatics tools delivering real-time predictions with explainable, actionable guidance. However, many healthcare AI solutions lack actionable recommendations and user-friendly explanations, limiting clinical impacts. We introduce APEA, a pediatric T1D self-management <b>A</b>mbient-AI assistance tool, integrating glucose multi-trajectory-scenarios <b>P</b>rediction, interactive, context-aware large language model <b>E</b>xplanations, and just-in-time adaptive intervention policy optimization for <b>A</b>ctionable real-time suggestions through reinforcement learning. Using T1DEXIP dataset (262 pediatric T1D patients, multi-center), our results showed improved glucose control outcomes: 45% over human management, 69% over infusion-pump management. Although constrained by small sample size and severe class imbalance, APEA addresses healthcare AI implementation gaps by bridging what might happen, what can be done about it, and why it makes clinical sense. APEA offers a transferable framework for other chronic conditions that demand continuous, personalized, just-in-time adaptive interventions.

Ähnliche Arbeiten