Using Generative AI to Simulate Patient History-Taking in a Problem-Based Learning Tutorial: A Mixed-Methods Study

Patient case design in problem-based learning (PBL) must balance realism, educational value, and feasibility. Generative artificial intelligence (genAI)-enabled virtual patients may facilitate this balance, but their use must be examined. In March 2024, 37% of second-year students at a small, community-based medical school in the Midwestern U.S. participated in the pilot implementation of a genAI-enabled virtual patient in a required PBL tutorial. Two groups of students (N = 13) opened the patient case by interviewing a genAI-enabled avatar. For comparison, two groups (N = 13) gathered patient history information using the school’s legacy system, a searchable multimedia database [Electronic PBL Module (ePBLM)]. All groups had the same expert faculty facilitator. The groups’ interactions were examined using descriptive observation. Students’ perceptions and their recall of patient history information were assessed using locally developed instruments. GenAI presented essential case content accurately, occasionally deviating from non-essential content. GenAI students spent approximately 10 min longer on history-taking than ePBLM students, collaboratively troubleshooting interaction with the AI. GenAI students rated the clinical accuracy and teamwork aspects of their experience significantly higher than their previous experiences using ePBLMs, however, they treated the avatar like a sophisticated and engaging “question base,” rather than a virtual patient. Patient history information recall was near ceiling across conditions. Our study provides clues for supporting clinical learning when representing patients realistically within PBL tutorials. Enhancing students’ history-taking from genAI-enabled virtual patients should include framing the experience: prompting students to experiment with AI’s capabilities and helping them recognize and adapt to its limitations.