AI Virtual Patient Training for Dental Student History Taking

This single-center, parallel-group randomized educational intervention was conducted among third-year undergraduate dental students at the Faculty of Odonto-Stomatology, University of Medicine and Pharmacy, Hue University, Hue, Vietnam.

Eligible students were third-year dental students who had not previously received history-taking training supported by a large language model. Participants were assigned in a 1:1 ratio to either a Gemini-based virtual-patient group or a conventional role-play group. Each group included 28 students.

The Gemini-based virtual patient was developed as a curriculum-bounded learning tool for common oral and maxillofacial clinical scenarios. Its knowledge base included third-year dental curriculum materials, sample medical and dental records, a clinical assessment rubric, and anonymized case scenarios adapted from patient records. The virtual patient was instructed to act only as the patient, disclose information only in response to relevant questions, remain consistent with the predefined case, avoid providing diagnostic guidance during the interview, and provide rubric-aligned feedback after the student completed the encounter.

Students in the virtual-patient group typed questions to the Gemini-based virtual patient, documented the elicited history, and received automated feedback after each case. Students in the conventional role-play group practiced with trained role-players or faculty members using standardized case scripts and the same feedback criteria. Both groups completed four sessions over 2 weeks, with each session lasting approximately 30 to 45 minutes. The groups received comparable scheduled practice time, clinical topics, number of cases, and feedback criteria.

The primary outcome was the change in case-based clinical history-taking competency score from baseline to postintervention assessment. Competency was assessed using five clinical scenarios representing common conditions in the third-year dental curriculum. The possible score range was 0 to 10 points. Each assessment was independently scored by two faculty assessors who were blinded to group allocation and questionnaire responses.

Secondary outcomes included student satisfaction and perceived effectiveness of the training method. Satisfaction covered ease of use, naturalness of interaction, willingness to recommend the method, and diversity of clinical conditions. Perceived effectiveness included self-directed learning, interest in learning, communication and feedback skills, question phrasing, systematic and logical history taking, and stress during history-taking practice. All questionnaire items used a 5-point Likert scale.