Agentic intelligence: multi-agent systems for autonomous healthcare diagnostics and treatment planning

Abstract

**Objective:** Agentic AI represents a paradigm shift from reactive systems to autonomous agents capable of perception, planning, reasoning, memory management, and tool usage without human intervention. This paper proposes the Healthcare Autonomous Reasoning Network (HARN), a multi-agent system designed to assist physicians with real-time diagnosis, treatment recommendations, and drug interaction detection. **Methods:** This paper introduces HARN’s architecture featuring a large language model (LLM) orchestrator coordinating four specialized sub-agents: radiology interpretation, electronic health record (EHR) exploration, pharmacological advice, and risk stratification. Experiments leverage the MIMIC-IV clinical dataset and curated drug interaction databases to evaluate diagnosis accuracy, decision latency, and hallucination rates. **Expected results:** HARN demonstrates an F1-score of 0.873 and 41% reduced clinical decision-making latency compared to baselines. **Conclusions:** The results validate the feasibility of agentic multi-agent systems for high-stakes healthcare applications, demonstrating better performance in autonomous clinical decision support.