Tool use reinforcement learning improves emergency care predictions
Incentivizing Tool-Use Reinforcement Learning for Post-Arrival Emergency Care Prediction
Artificial Intelligence
Summary
Predicting how patients will do in emergency care is hard because computers often struggle to pick out the best information and understand what numbers mean for health risks. The authors created a method called TIR-EC that helps a language model use special tools to pick and interpret important data from health records. This method learns from a strong existing model and gradually teaches the language model clinical skills to get better at predictions. Their approach beats common methods and works well even compared to much bigger language models.
What this means in practice
- •For hospital data teams: Improve emergency department patient risk predictions by integrating interpretable tool calls with reinforcement learning guided by clinical models.
- •For healthcare software developers: Build AI systems that better combine clinical measurements using tool-based learning to enhance decision support in emergency care.
Authors
Pengfei Hu, Chenwei Wu, Qiao Jin, Jun Deng, Zhiyong Lu, Liyue Shen, Yue Ning
Abstract
Large language models (LLMs) can reason over both structured measurements and free-text records, which makes them attractive for emergency department (ED) prediction. However, most existing LLMs still fall short of conventional ML methods like trees on EHR-based prediction tasks. Our controlled analysis traces this gap to two weaknesses: LLMs do not select informative features, and they misjudge what numerical values imply for risk even when given the right features. We propose TIR-EC, a tool-integrated reinforcement learning framework that addresses both. TIR-EC exposes EHR feature groups as tools the LLM can call. It trains the LLM with rewards from a gradient-boosted tree teacher, which provides a reference set of informative tools and a graded risk target. TIR-EC also distills task-specific clinical skills into the model weights through a reverse curriculum that gradually removes the skills from the prompt during RL process. Evaluation on three clinical prediction tasks shows that our 4B agent achieves the highest macro F1 of all compared models, outperforms conventional ML baselines, and is competitive with much larger frontier LLMs. Further analysis shows that TIR-EC selects more useful evidence with fewer tool calls and combines measurements more effectively.