Multi-Stage Patient Role-Playing Framework for Realistic Clinical Interactions

TL;DR

We address the realism gap in doctor–patient dialogue simulations for clinical LLMs by introducing Ch-PatientSim, a dataset built from real gastroenterology encounters and balanced with few-shot augmentation. We propose a training-free Multi-Stage Patient Role-Playing (MSPRP) framework that uses a five-dimensional persona vector $P$ partitioned into Communication Style and Expressive Capacity to generate personalized, clinically coherent patient responses across three stages: Basic Information Generation, Communication Style Injection, and Expression Consistency Regulation. Across multiple LLMs, MSPRP yields substantial gains in persona consistency, factual accuracy, naturalness, and contextual relevance, with ablation analyses confirming the importance and sequencing of each stage. The work provides a practical, data-grounded benchmark and a scalable framework to enhance realism in AI-driven medical education and diagnostic tools.

Abstract

The simulation of realistic clinical interactions plays a pivotal role in advancing clinical Large Language Models (LLMs) and supporting medical diagnostic education. Existing approaches and benchmarks rely on generic or LLM-generated dialogue data, which limits the authenticity and diversity of doctor-patient interactions. In this work, we propose the first Chinese patient simulation dataset (Ch-PatientSim), constructed from realistic clinical interaction scenarios to comprehensively evaluate the performance of models in emulating patient behavior. Patients are simulated based on a five-dimensional persona structure. To address issues of the persona class imbalance, a portion of the dataset is augmented using few-shot generation, followed by manual verification. We evaluate various state-of-the-art LLMs and find that most produce overly formal responses that lack individual personality. To address this limitation, we propose a training-free Multi-Stage Patient Role-Playing (MSPRP) framework, which decomposes interactions into three stages to ensure both personalization and realism in model responses. Experimental results demonstrate that our approach significantly improves model performance across multiple dimensions of patient simulation.

Figures (5)

• Figure 1: Comparison of different patient simulation datasets (AI Hospital (fan2025ai), PatientSim (kyung2025patientsim) and ours).
• Figure 2: The construction process of the Chinese patient simulation dataset (Ch-PatientSim).
• Figure 3: Analysis of the persona attributes in the Ch-PatientSim dataset.
• Figure 4: Multi-Stage Patient Role-Playing (MSPRP) framework.
• Figure 5: Case Study. Comparison of Original Responses, Basic Responses, and Responses Applying the MSPRP Framework to Physician Questions.