AI Chatbots Sweep Healthcare Simulation Research, But Rigorous Outcome Studies Lag Behind

A bibliometric review published in Global Surgical Education has mapped nearly five years of research on large language models in healthcare simulation education, and the picture it reveals is one of explosive growth paired with a striking gap in rigorous evaluation. The analysis, covering studies published from 1 January 2020 to 13 August 2026, identified 483 unique publications, of which 411 — roughly 85 percent — appeared in just the first nineteen and a half months of 2025 and 2026.

The review drew records from six databases — OpenAlex, PubMed, Europe PMC, Semantic Scholar, DOAJ and Scopus — and applied a written eligibility codebook through two operationally separate screening processes powered by a GPT-5.6 Sol-based system, labelled Reviewer A and Reviewer B. Records were deduplicated and grouped so each study contributed once, with lead-author checks conducted on eligibility in two samples and on non-technical-skill classifications in a subset.

What the mapped corpus shows is a field heavily weighted toward building rather than testing. Development or evaluation of models was the most frequent study design, accounting for 163 of the 483 studies, or 34 percent. Scenario-based approaches dominated the simulation landscape, appearing in 294 studies (61 percent), closely followed by virtual patients at 290 studies (60 percent).

Explicit non-technical-skill domains were identified in 360 studies, or 75 percent of the corpus, with communication and decision-making the most frequently addressed. Communication was the most commonly examined domain, named in 288 studies (60 percent), while decision-making appeared in 130 studies (27 percent).

Yet the review’s conclusions temper enthusiasm about the field’s maturity. Comparative designs — studies that directly tested whether AI-enhanced simulation improves learning relative to alternatives — formed a minority of the corpus, just 66 of 483 studies, or 14 percent. Educational effects were not synthesised, meaning the review could not and did not draw conclusions about whether large language models actually improve training outcomes. The authors call for future research to prioritise comparative learning outcomes, retention and transfer of skills, cognitive training, and team-based simulation, signalling that the field’s next challenge is moving from demonstration to evidence.

Subject of Research: Bibliometric mapping of large language model research in healthcare simulation education and non-technical skills

Article Title: Large language models in healthcare simulation education: a bibliometric analysis of simulation approaches and non-technical skills

Article References: Pears, M., Wadhwa, K., Payne, S. R., Konstantinidis, S. T. H., & Biyani, C. S. (2026). Large language models in healthcare simulation education: a bibliometric analysis of simulation approaches and non-technical skills. Global Surgical Education – Journal of the Association for Surgical Education, 5(1), Article 208. https://doi.org/10.1007/s44186-026-00592-6

Image Credits: AI Generated

DOI: 10.1007/s44186-026-00592-6

Keywords: large language models, healthcare simulation, bibliometric analysis, non-technical skills, virtual patients, medical education, communication training, decision-making, systematic review, scenario-based learning, Large, language