Artificial intelligence generated virtual patients in health professions education: a scoping review.
4 affiliations
- 1Department of Diagnostic Radiology, Yong Loo Lin School of Medicine, National University of Singapore, 1E Kent Ridge Road, Singapore, 119228, Singapore.
- 2Department of Surgery, Yong Loo Lin School of Medicine, National University of Singapore, 1E Kent Ridge Road, Singapore, 119228, Singapore. [email protected].
- 3National University Health System, 1E Kent Ridge Road, Singapore, 119228, Singapore.
- 4Department of Surgery, Yong Loo Lin School of Medicine, National University of Singapore, 1E Kent Ridge Road, Singapore, 119228, Singapore.
Abstract
Background
Generative artificial intelligence (AI) can enable large scale virtual patient production. However, the health professions education (HPE) community lacks a synthesized overview of how AI generated virtual patients can be best utilized to achieve intended learning outcomes. We sought to synthesize the range of use cases, limitations and approaches to fulfilling these use cases using AI generated virtual patients, including beneficial design features and strategies addressing limitations.Methods
We conducted a scoping review following the Arksey and O'Malley framework, and reported according to PRISMA-ScR guidelines. MEDLINE, Scopus, Web of Science, and CINAHL databases were searched using terms related to AI and virtual patients. After removal of duplicates, titles and abstracts were screened, followed by full-text review of eligible articles. Original research using AI to generate an interactive virtual patient for HPE was included. Data were charted on study characteristics, educational use cases, outcomes evaluated, limitations, beneficial features and development approaches, and qualitatively synthesized.Results
Ninety-seven studies were included. Outcomes were mostly evaluated at Kirkpatrick level one. Use cases included learning and assessing communication skills, diagnostic reasoning, and management reasoning. Limitations included inauthentic conversational flow, excessive agreeableness and clinical inaccuracy. Strategies addressing these challenges included iterative cycles of prompt engineering and review by clinical experts and learners, confining AI platforms to readily available valid information sources, specifying behavioral rules such as adherence to the patient role and limiting reply comprehensiveness, few shot prompting, and building conversational memory. Beneficial features included diverse cases capturing authentic variation in clinical conditions and patient personalities, and feedback that was immediate, individualized, specific, actionable, and transparent.Conclusion
Educators seeking to locally harness the potential of AI generated virtual patients can utilize the range of use cases and design features identified in this review. They should also be cognizant of limitations in clinical accuracy and inauthentic conversational flow, and consider development strategies targeting such limitations. Future research should explore other potentially beneficial design features from the wider educational literature and evaluate outcomes beyond participant reaction and include behavioral transfer to the workplace, cost, and organizational impact of AI generated virtual patients.
Related
- Artificial Intelligence-generated Clinical Images for Plastic Surgery Education: A Prospective Randomized Double-blind Study.
- A WeChat-based, expert knowledge-constrained artificial intelligence health education intervention for patients undergoing gastrointestinal surgery: protocol for a randomized controlled trial
- Effect of an artificial intelligence health education accurately linking system on self-management in young and middle-aged patients after percutaneous coronary intervention: a protocol
- The Influence of Large Language Model-Generated Artificial Intelligence Risk Estimates on Clinical Decision-Making in Breast Imaging: A Simulation-Based Study
- Artificial intelligence in biomedical visualization and education: a narrative review
BibTeX
@article{fang2026artificial,
title = {{Artificial intelligence generated virtual patients in health professions education: a scoping review.}},
author = {Eric Kin Meng Fang and Yuru Boon and Jin Yang Ho and Natasha Tey and Matthew Jian Wen Low},
journal = {BMC medical education},
year = {2026},
month = jul,
doi = {10.1186/s12909-026-10028-w},
url = {https://europepmc.org/article/MED/42928266},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Artificial intelligence generated virtual patients in health professions education: a scoping review.
AU - Eric Kin Meng Fang
AU - Yuru Boon
AU - Jin Yang Ho
AU - Natasha Tey
AU - Matthew Jian Wen Low
PY - 2026
DA - 2026/07/28
JO - BMC medical education
DO - 10.1186/s12909-026-10028-w
UR - https://europepmc.org/article/MED/42928266
AB - Background Generative artificial intelligence (AI) can enable large scale virtual patient production. However, the health professions education (HPE) community lacks a synthesized overview of how AI generated virtual patients can be best utilized to achieve intended learning outcomes. We sought to synthesize the range of use cases, limitations and approaches to fulfilling these use cases using AI generated virtual patients, including beneficial design features and strategies addressing limitations.Methods We conducted a scoping review following the Arksey and O'Malley framework, and reported according to PRISMA-ScR guidelines. MEDLINE, Scopus, Web of Science, and CINAHL databases were searched using terms related to AI and virtual patients. After removal of duplicates, titles and abstracts were screened, followed by full-text review of eligible articles. Original research using AI to generate an interactive virtual patient for HPE was included. Data were charted on study characteristics, educational use cases, outcomes evaluated, limitations, beneficial features and development approaches, and qualitatively synthesized.Results Ninety-seven studies were included. Outcomes were mostly evaluated at Kirkpatrick level one. Use cases included learning and assessing communication skills, diagnostic reasoning, and management reasoning. Limitations included inauthentic conversational flow, excessive agreeableness and clinical inaccuracy. Strategies addressing these challenges included iterative cycles of prompt engineering and review by clinical experts and learners, confining AI platforms to readily available valid information sources, specifying behavioral rules such as adherence to the patient role and limiting reply comprehensiveness, few shot prompting, and building conversational memory. Beneficial features included diverse cases capturing authentic variation in clinical conditions and patient personalities, and feedback that was immediate, individualized, specific, actionable, and transparent.Conclusion Educators seeking to locally harness the potential of AI generated virtual patients can utilize the range of use cases and design features identified in this review. They should also be cognizant of limitations in clinical accuracy and inauthentic conversational flow, and consider development strategies targeting such limitations. Future research should explore other potentially beneficial design features from the wider educational literature and evaluate outcomes beyond participant reaction and include behavioral transfer to the workplace, cost, and organizational impact of AI generated virtual patients.
SN - 1472-6920
ER -