CAREAgent: Clinical Agent with Structured Reasoning and Tool-Integrated for Order Generation
Ruihui Hou, Ziyue Huai, Chennuo Zhang, Ziyan Liu, Siran Zhao, Yao Yu, Jie Zhai, Tong Ruan
Why It Matters
What makes this one worth your time
This research addresses a critical gap in clinical decision-making by improving the translation of medical decisions into actionable orders, which is essential for effective healthcare delivery.
CAREAgent enhances clinical order generation through structured reasoning and tool integration.
Summary
The paper presents CAREAgent, a clinical order generation agent that utilizes a two-stage reasoning data construction method to improve the generation of executable clinical orders, demonstrating enhanced performance on multiple benchmarks.
Key contributions
- Development of CAREAgent for clinical order generation.
- Introduction of a two-stage reasoning data construction method.
- Demonstration of improved performance on ClinicalBench with specific F1 score enhancements.
Notable insights
- The introduction of a two-stage agentic reasoning data construction method is a novel approach to enhance the quality of training data for clinical agents.
- The use of multi-dimensional reward functions in reinforcement learning for optimizing complex clinical reasoning capabilities is a sophisticated methodology.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2606.01094v1 Announce Type: new Abstract: Clinical order generation serves as a critical bridge between clinical decision-making and real-world practice, translating medical decisions into concrete and executable orders. Existing agents mainly focus on coarse-grained decisions and overlook the fine-grained, executable information required for clinical orders. To address this gap, we propose CAREAgent, an agent for clinical order generation. To support its training, we introduce a two-stage agentic reasoning data construction method. First, we design an agent framework that constructs verifiable reasoning trajectories aligned with realistic clinical tool usage. Second, we filter reasoning trajectories by format compliance, order validity, and clinical plausibility. Building on the constructed data, the model is first trained via supervised fine-tuning to acquire fundamental reasoning formats and medical knowledge, and is subsequently optimized through reinforcement learning with multi-dimensional reward functions to enhance complex clinical reasoning capabilities. Experiments on multiple benchmarks demonstrate the effectiveness of CAREAgent. On ClinicalBench (unseen during training), CAREAgent improves the F1 score by 5.05%, 2.09%, and 0.86% over the single-agent, multi-agent, and agentic reasoning methods, respectively.