Intelligent Simulated Patient System for Gastroenterology Education

History-taking is a fundamental clinical skill in gastroenterology because digestive diseases often present with complex and nonspecific symptoms that require systematic information gathering for accurate diagnosis. Traditional history-taking education relies on bedside teaching, standardized patients, and faculty supervision. However, these approaches are limited by high educational costs, restricted access to standardized patients, variability in assessment, and limited opportunities for repeated practice.

This study developed an Intelligent Simulated Patient System (ISPS) powered by a large language model (LLM) and retrieval-augmented generation (RAG) technology for gastroenterology education. The system was built using de-identified electronic health records, clinical guidelines, and textbook-based knowledge to create realistic virtual patients capable of maintaining disease-specific characteristics throughout multi-turn conversations. In addition to virtual patient simulation, the platform includes an automated scoring system that evaluates history-taking completeness using semantic similarity matching and logical completeness assessment based on predefined clinical rubrics.

This prospective randomized controlled trial aims to evaluate whether ISPS-assisted training improves clinical history-taking skills compared with conventional teaching methods. Ninety medical students who had completed courses in diagnostics and internal medicine were randomly assigned in a 1:1 ratio to either an ISPS training group or a conventional teaching group. The intervention lasted four weeks.

Participants in the control group received conventional clinical education, including bedside teaching, case-based learning, and supervised clinical discussions. Participants in the intervention group independently completed repeated history-taking practice using the ISPS platform with virtual patients representing common gastroenterology diseases.

The primary outcomes are history-taking completeness scores and diagnostic accuracy assessed during a standardized Objective Structured Clinical Examination (OSCE). Secondary outcomes include empathy performance during physician-patient communication and participants' acceptance of the ISPS assessed using the Technology Acceptance Model (TAM) questionnaire.

The study also evaluates the reliability of the automated scoring system by comparing AI-generated scores with expert ratings using Pearson correlation, intraclass correlation coefficients (ICC), and Bland-Altman analysis.

The findings are expected to provide evidence regarding the effectiveness of large language model-based virtual patient systems for improving clinical communication skills, standardized assessment, and competency-based medical education in gastroenterology.