TIR-EC Reinforcement Learning Improves LLM Emergency Care Prediction

TIR-EC addresses LLM failures in clinical risk prediction by treating electronic health record (EHR) feature groups as tools within a reinforcement learning framework. By using a gradient-boosted tree as a teacher, the model learns to select informative features and correctly interpret numerical risk values.