LLMs Can Simulate Standardized Patients via Agent Coevolution

TL;DR

EvoPatient introduces a fully autonomous, coevolving multi-agent framework to simulate standardized patient training for medical education. By coupling a patient agent with multiple doctor agents and organizing knowledge through Attention and Trajectories libraries, the system learns standardized presentation patterns and generates high-quality, human-like dialogues with minimal human supervision. Extensive experiments on real patient records demonstrate improved requirement alignment, robust answer quality, and strong transferability across diseases, with optimized resource use. The work offers a scalable path toward autonomous SPs that can enhance doctor training while reducing reliance on human SPs, albeit with acknowledged limitations and ethical considerations.

Abstract

Training medical personnel using standardized patients (SPs) remains a complex challenge, requiring extensive domain expertise and role-specific practice. Previous research on Large Language Model (LLM)-based SPs mostly focuses on improving data retrieval accuracy or adjusting prompts through human feedback. However, this focus has overlooked the critical need for patient agents to learn a standardized presentation pattern that transforms data into human-like patient responses through unsupervised simulations. To address this gap, we propose EvoPatient, a novel simulated patient framework in which a patient agent and doctor agents simulate the diagnostic process through multi-turn dialogues, simultaneously gathering experience to improve the quality of both questions and answers, ultimately enabling human doctor training. Extensive experiments on various cases demonstrate that, by providing only overall SP requirements, our framework improves over existing reasoning methods by more than 10\% in requirement alignment and better human preference, while achieving an optimal balance of resource consumption after evolving over 200 cases for 10 hours, with excellent generalizability. Our system will be available at https://github.com/ZJUMAI/EvoPatient.

Figures (24)

• Figure 1: EvoPatient integrates multiple evolvable agents with distinct roles, collaboratively simulating a real-world diagnostic process that effectively trains doctors with various cases.
• Figure 2: A typical multi-turn dialogue between the patient agent () and the doctor agents (). The agents maintain a continuous memory, and doctor agents can request the recruitment of new doctors. Additionally, the agents continuously store and retrieve knowledge from the library () to facilitate ongoing evolution.
• Figure 3: Multidisciplinary process in our framework.
• Figure 4: An example that standardizes our patient agent through attention requirements and effective few-shot demonstrations for human doctor training.
• Figure 5: Transferability of evolution on five types of diseases before and after patient agent evolution.