Evidence-Grounded Longitudinal Multimodal Clinical Agent for Physician Decision Support

This paper proposes an Evidence-Grounded Longitudinal Multimodal Clinical Agent (ELM-ClinicalAgent) for physician decision support. The proposed agent is designed to retrieve and organize relevant longitudinal patient information from heterogeneous clinical data, identify temporal changes and conflicting evidence, link generated findings to their source records, and explicitly communicate uncertainty. The system is designed as a physician-in-the-loop approach in which the AI assists with information retrieval, clinical context reconstruction, and evidence organization while the final clinical decision remains with the physician. The paper presents the proposed architecture, workflow, research hypotheses, evaluation framework, safety considerations, and directions for prototype implementation and empirical validation.