AgentRx: Benchmark Study of LLM Agents for Multimodal Clinical Prediction

Researchers have introduced AgentRx, a new benchmark study evaluating the effectiveness of Large Language Model (LLM)-based agents in multimodal clinical prediction tasks. The study addresses the challenge of synthesizing heterogeneous healthcare data, including electronic health records, medical images, and clinical notes. While collaborative agent frameworks are proposed to mitigate data sharing issues, this work systematically assesses their performance against single-agent systems using large-scale real-world data. The findings reveal that single-agent frameworks currently outperform naive multi-agent systems in handling multimodal data and demonstrate better calibration. This highlights a critical need for improving multi-agent collaboration mechanisms to effectively process diverse clinical inputs. By open-sourcing their code and evaluation framework, the authors provide a valuable resource for future developments in agentic healthcare systems. The study underscores the potential of LLM agents in clinical decision support while identifying specific areas for technical improvement in multi-agent architectures.