Language Agents for Hypothesis-driven Clinical Decision Making with Reinforcement Learning

Researchers developed LA-CDM, a language agent that uses reinforcement learning to support clinical decision-making by iteratively requesting tests and generating hypotheses for diagnosis. The system was trained using a hybrid approach combining supervised and reinforcement learning, and tested on real-world data covering four abdominal diseases.

• →LA-CDM models clinical decision-making as an interactive, iterative process rather than assuming all patient information is immediately available.
• →The system uses hypothesis-driven uncertainty-aware language agents that converge towards diagnosis through relevant test requests.
• →Training combines supervised and reinforcement learning targeting accurate hypothesis generation, uncertainty estimation, and efficient decision-making.
• →Evaluation on MIMIC-CDM dataset covering four abdominal diseases showed improved diagnostic performance and efficiency.
• →The approach addresses limitations of current LLM applications in clinical decision support that lack task-specific training.