Development and Evaluation of an Artificial Intelligence-Simulated Patient Tool for Clinical Training in Family Medicine

Background Traditional clinical training in family medicine relies on case–based discussions and standardised patients, which are resource-intensive and limited in scope. We designed and evaluated an artificial intelligence (AI)-simulated patient tool for family medicine training, focusing on usability, realism, and educational value. Methods We used a two-phase developmental research framework (design and development, and evaluation) to design a large language model-based conversational agent powered by GPT-4o, grounded in case reports, patient personas, and normal-range databases. To evaluate the tool, we recruited European family doctors through the European Young Family Doctors’ Movement network and snowball sampling. Participants interacted with the tool and completed nine 5-point Likert items (ranging from 1/Strongly disagree to 5/Strongly agree) and three open-ended questions. Quantitative data were analysed descriptively, while qualitative responses underwent thematic analysis. Results The tool was rated by most of the 21 participants as enjoyable, beneficial for clinical practice, giving evidence-based feedback (all median 4, interquartile range (IQR) 1); easy to use, making appropriate responses and giving a realistic conversational flow (all median 4, IQR 2). Most participants would strongly recommend it to their colleagues (median 5, IQR 1). Strengths included its supportive role, usability, and educational value, while weaknesses involved interaction limitations, realism, and cultural challenges. Suggested improvements included expanding content, adding multilingual and cultural adaptations, and enhancing interactiveness. Conclusion The AI-simulated patient is feasible, well-received, and holds promise for enhancing family medicine education. Future work should add speech-to-text, AI avatars, and measurement of the impact in a randomised controlled trial.