An Agentic AI–Driven End-to-End Clinical Workflow for Diabetes Management

Diabetes Mellitus is a disease that demands persistent monitoring, clinical decisions, as well as comprehensive clinical workflows. Traditionally, traditional Clinical Decision Support Systems (CDSS) as well as machine learning models are considered isolated tools that present clinical workflows with limited illustration and flexibility. However, the latest large language models present clinical reasoning that brings out issues of hallucination.

This paper proposes an agentic AI-driven end-to-end clinical workflow in the management of diabetes, incorporating elements such as patient entry, data integration, evidence retrieval, reasoning, validation, and clinician feedback. In addition, this proposed framework includes elements such as data harmonization, retrieval augmentation, and agent orchestration, all within a human-in-the-loop paradigm.

The system is assessed using simulated clinical workflows and is compared with traditional CDSS and static machine learning methodologies. With the proposed approach, a 35% reduction has been achieved in terms of data-to-decision time, which is greater than 40% for hallucination rate, and a 92% improvement has been achieved for factual consistency, as indicated by the significant potential offered by agentic AI to inform adaptable and trustworthy clinical decision support to manage chronic disease.