From Knowledge to Action: An Agentic AI Framework for Diabetes Management

Abstract

Diabetes remains a pressing global health challenge, requiring continuous monitoring, timely interventions, and personalized management strategies. Existing digital health solutions-ranging from rule-based decision support systems to data-driven machine learning models-often stop short at prediction, lacking the ability to transform knowledge into actionable guidance for patients and clinicians. In this paper, we present an Agentic AI framework for diabetes management that bridges this gap by unifying structured knowledge with predictive modeling and autonomous decision support. The framework leverages knowledge graphs as the semantic foundation, integrating biomedical, lifestyle, and psychosocial factors, while Agentic AI components orchestrate reasoning, simulation, and patient interaction. Using the MIDUS dataset, we train machine learning models that serve as predictors and simulators, enabling both individualized risk forecasting and “what-if” scenario analysis. By coupling these predictive models with retrieval-augmented reasoning grounded in knowledge graphs, the system generates context-aware, interpretable, and proactive recommendations. Initial results demonstrate the feasibility of this hybrid approach, highlighting the potential of Agentic AI to advance personalized, explainable, and actionable diabetes care.

Ähnliche Arbeiten