Application of large language models in medical interview training: a study with medical students

This paper explores the application of Large Language Models (LLMs) in medical interview training. While medical interviews remain fundamental in healthcare, training methods often require human interaction, limiting practice opportunities. We investigate if LLMs can effectively simulate patients for training purposes. We examined commercially available models and fine-tuned open-source LLMs using QLoRA techniques on a dataset of medical interviews. We developed a web application employing these models and conducted in-depth interviews with medical students to evaluate its effectiveness. Students found the application helpful, rating conversation quality as good and highlighting advantages over traditional training methods, particularly regarding availability and consistency in patient symptom presentation. While students emphasized that LLMs cannot replace real patient interactions, they recognized significant benefits for supplementary training. Our findings confirm that LLMs can be valuable tools in medical interview training, providing opportunities for skill development without dependency on peer availability or scheduled sessions.