Actionable companion · Assessment & reasoning

Learning with agency: Feasibility of Large Language Model–Based Standardized Virtual Patients to Support Clinical Decision-Making Training in Operative Dentistry: Mixed Methods Study

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not claims or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paper Search paper agent CC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Higher education. Health-professions learning must be distinguished from K–12 instruction.

Original authors’ abstract, reproduced under the linked source license:

Abstract Background Clinical decision-making training in operative dentistry commonly relies on real or standardized patients to develop undergraduate students’ ability to deliver safe, effective, and patient-centered care. However, training with real or standardized patients can be limited in scalability, cost-effectiveness, and accessibility. Large language models, with their human-like language capabilities, may have the potential to simulate patients in clinical encounters and help overcome some limitations associated with traditional training approaches. Objective This study aimed to evaluate the feasibility of using large language model–based standardized virtual patients to support undergraduate dental students’ clinical decision-making training in operative dentistry. Methods This mixed methods cross-sectional feasibility study was conducted during a simulation-based clinical decision-making training session in the Operative Dentistry and Cariology course at the College of Dentistry, King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia. Eligible participants were second-year undergraduate dental students enrolled in the course. A convenience sampling approach was used, with all eligible students (N=50) invited to participate. A total of 41 students completed the study, 23 (56%) of whom were male. The students were divided into 8 groups. Each group interacted with 2 standardized virtual patients powered by ChatGPT-4o (OpenAI) through the Chatbase platform to complete comprehensive history-taking and then reviewed the standardized virtual patients’ intraoral photographs and bitewing radiographs. For each standardized virtual patient, students as a group recorded diagnoses, performed a risk assessment, and formulated a treatment plan. Students then completed the Student Satisfaction and Self-Confidence in Learning questionnaire. The quality of the standardized virtual patient responses and overall dialogue realism were evaluated using the Dialogue Authenticity Scale. The dialogues were also thematically analyzed to identify authenticity-undermining response features and explore their context and underlying causes. Results Students perceived the simulation-based training session positively, with all questionnaire items showing high median scores (4.00‐5.00 on a 5-point scale), and both item-level IQRs and 95% CIs spanning no more than 1.0 scale point. In addition, standardized virtual patient responses were largely authentic, with an overall median authenticity rating of 4.50 (IQR 4.00-5.00; 95% CI 4.00-5.00) on a 6-point scale across all interactions. However, several authenticity-undermining response features were identified, including responses that were inconsistent with typical human behavior, contained information beyond a patient’s likely knowledge, or were factually incorrect. Conclusions This proof-of-concept study supports the feasibility of implementing large language model–based standardized virtual patients in undergraduate simulation-based clinical decision-making training in operative dentistry. In a dental context where this application has been only minimally evaluated, this study provides early evidence of positive student perceptions, acceptability, and largely authentic dialogue, while also identifying important performance limitations. Further research is warranted to optimize performance and to evaluate the educational effectiveness of this approach in improving undergraduate students’ clinical skills and knowledge.

What kind of evidence is this?

Supporting AI learning science · Survey / observational study

Stage: Higher education. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Study methods · source passage

This study employed 2 large language model–based standardized virtual patients powered by ChatGPT-4o (OpenAI) within a simulation-based training session to evaluate their feasibility in supporting undergraduate clinical decision-making training in operative dentistry. The 2 standardized virtual patients were designed as high-caries-risk cases with differing risk factors and distinct demeanors and communication styles. Feasibility was evaluated using the Dialogue Authenticity Scale to assess the quality of standardized virtual patient responses during interactions with students and the realism of the overall dialogue in the simulation-based training session [20]. In addition, students’ perceptions and acceptability after the simulation-based training session were assessed with the validated Student Satisfaction and Self-Confidence in Learning questionnaire [27].

Limitations and uncertainty · source passage

To the best of the author’s knowledge, this study is among the first to investigate the feasibility of incorporating large language model–based standardized virtual patients into undergraduate dental education. However, a few limitations should be considered when interpreting the findings. First, this is a proof-of-concept study with a small sample size and therefore provides only preliminary evidence supporting the feasibility of integrating these standardized virtual patients into undergraduate dental simulation-based training, warranting evaluation in larger studies. In addition, the generalizability of these findings to all levels of undergraduate dental students may be limited, as only preclinical students were included in this study. Furthermore, the authenticity ratings and thematic analysis were conducted by a single researcher; therefore, interrater reliability could not be assessed, and the findings may have been influenced by rater-specific bias. Moreover, intrarater reliability for the authenticity ratings, assessed using quadratic-weighted Cohen kappa, indicated only moderate consistency. This level of agreement may have introduced measurement error and reduced confidence in the precision of the authenticity ratings.

Design the system before the pilot

Use this study to inform a learning mechanism or implementation question. It is not direct evidence that autonomous agents improve learning. Keep the proposed agent behavior separate from what the source actually tested.

Use common rubrics, blind scoring where feasible and a fresh unaided reasoning task. Document permissible assistance and avoid treating AI-detector scores as proof of misconduct.
Co-design accommodations with learners and specialists. Check access and error burden across groups without inferring subgroup effectiveness from small or unreported samples.

A 90-day district pilot

Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
Days 16–30 · Design. Choose one disciplinary task. Require a first attempt, a documented AI consultation, a critique of the output and an independent explanation. Align the comparison group and scoring rubric with the same objective. Name an accountable owner and set the baseline, success threshold and budget before implementation.
Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
Days 61–90 · Decide. Blind-score reasoning and accuracy on a new task without AI. Distinguish model examination scores, student perceptions and actual learner gains; only the last directly tests educational benefit. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Fahad BaHammam. Feasibility of Large Language Model–Based Standardized Virtual Patients to Support Clinical Decision-Making Training in Operative Dentistry: Mixed Methods Study. JMIR Formative Research (2026-05-19). https://doi.org/10.2196/91021

Companion prepared 2026-10-30. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.