Comprehensive Summary
Augmented reality ocular examinations are effective in medical teaching by allowing students to learn to examine eye health. However, these simulations lack practice with going through symptoms and history with a patient, an essential part of making diagnoses. To incorporate this element in teaching, large language AI models have been adopted for role-play simulation. While ChatGPT has been used to fill this gap, new AI platforms such as SimConverse are designed to enhance healthcare professionals’ communication skills. 16 participants were used in this study and required two years of basic knowledge of symptom taking and slit-lamp operation. Students were tested with three different conditions (blepharitis, keratoconus, and early wet age-related macular degeneration). To examine how students' experiences changed when going through simulations, they took a pre- and post-questionnaire asking about their perceptions of tangibility, reliability, responsiveness, assurance, and empathy. Students reported improvement in all competencies, including identifying diagnoses, offering reassurance to patients, adapting their communication styles, and tailoring management plans to the patient’s needs. ChatGPT performed worse in role-play than SimConverse, as it lacked emotion in responses and often gave wrong information. There was no significant difference between how students rank feedback from an instructor and SimConverse.
Outcomes and Implications
Using AI in slit-lamp examination simulations proved to be incredibly useful for students, with many of them outperforming their competency-based expectations. This study also showed that using AI is just as effective in students' learning as clinically trained actors when providing history and symptoms. Overall, this shows that large language models have great potential in medical teaching, especially SimConverse, as it can provide more realistic patient role-play and present feedback equally as well as clinical supervisors. This study also showed that ChatGPT is not the best AI model to use in role-play during medical teaching, and that other options should be explored if implementing AI in simulations.