EUVIMEDEUVIMED entdecken
Vergleichsstudie

Evaluating the performance of large language models versus prosthodontic residents on the 2024 and 2025 National Prosthodontic Resident Examination

Soni Prasad, Qiao Fang, Merve Koseoglu, Marwa Shembesh, Maryam Gheisarifar, Cortino Sukotjo

The Journal of Prosthetic Dentistry · 2026

DOI: 10.1016/j.prosdent.2026.03.043 ↗

Kurz gefasst

Vergleichsstudie in der Zeitschrift The Journal of Prosthetic Dentistry, 2026. Verfasst von Soni Prasad und 5 weiteren Autor:innen. Beteiligte Einrichtung: Associate Professor, Adams School of Dentistry, University of North Carolina at Chapel Hill, Chapel Hill, N.C. Electronic address: soniprasad@yahoo.c…. Hauptthemen laut MEDLINE: Large Language Models, Prosthodontics, Internship and Residency, Educational Measurement. Die Arbeit nennt 29 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

STATEMENT OF PROBLEM: Recent advancements in large language models (LLMs) offer expanded reasoning, multimodal processing, and improved factual grounding. However, their performance relative to prosthodontic residents remains unknown, and comparisons can inform whether current training fosters higher-order reasoning beyond the capabilities of artificial intelligence (AI). PURPOSE: This study compared the performance of 3 advanced LLMs, ChatGPT-5, Gemini 2.5 Pro, and Claude Sonnet 4, with prosthodontic residents on the 2024 and 2025 National Prosthodontic Resident Examinations (NPREs). MATERIAL AND METHODS: Three hundred multiple-choice questions (MCQs) were obtained from the 2024 and 2025 NPRE (150 items per year). Each MCQ was entered verbatim into the 3 LLMs using zero-shot prompting in isolated sessions. Responses were scored against the American College of Prosthodontists answer key. Aggregate resident performance by postgraduate year and subject domain was obtained from the ACP database. Descriptive statistics summarized accuracy, and differences among LLMs were evaluated using Cochran Q, chi-squared, or Fisher exact tests (α=.05). RESULTS: Across both years, all LLMs outperformed residents in most content areas, including materials, esthetics, implant prosthodontics, pharmacology, and pathology. Residents performed better than all LLMs only in diagnosis (2024) and in infection control and microbiology (2025). Overall LLM performance for the 2024 and 2025 were statistically similar (P=.280 and P=.173, respectively). Gemini 2.5 Pro achieved the highest accuracy in 2024 (78.7%) and 2025 (85.3%). All LLMs scored higher on the 2025 examination compared with 2024, although year-to-year differences were statistically similar (overall, for Gemini 2.5 Pro P=.193, for ChatGPT-5 P=.314, for Claude Sonnet 4 P=.282). CONCLUSIONS: LLMs demonstrated high accuracy on consecutive national prosthodontic examinations and surpassed residents in most domains. Resident advantages in a few areas requiring contextual interpretation indicate domains in which resident accuracy exceeded model accuracy. LLMs show promise as supplemental educational tools, but responsible use requires AI literacy, verification skills, and assessment strategies that emphasize higher-order reasoning.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Large Language ModelsProsthodonticsInternship and ResidencyEducational MeasurementHumansGenerative Artificial Intelligence

Autor:innen und Institutionen

  • Associate Professor, Adams School of Dentistry, University of North Carolina at Chapel Hill, Chapel Hill, N.C. Electronic address: soniprasad@yahoo.com.
  • Clinical Assistant Professor, Department of Restorative Dentistry, College of Dentistry, University of Illinois Chicago, Chicago, Ill.
  • Associate Professor, Department of Prosthodontics, Faculty of Dentistry, University of Sakarya, Sakarya, Turkey.
  • Professor and Chair, Department of Prosthodontics, School of Dental Medicine, University of Pittsburgh, Pittsburgh, Pa.

Zitieren

APA

Prasad, S., Fang, Q., Koseoglu, M., Shembesh, M., Gheisarifar, M., & Sukotjo, C. (2026). Evaluating the performance of large language models versus prosthodontic residents on the 2024 and 2025 National Prosthodontic Resident Examination. The Journal of Prosthetic Dentistry. https://doi.org/10.1016/j.prosdent.2026.03.043

Vancouver

Prasad S, Fang Q, Koseoglu M, Shembesh M, Gheisarifar M, Sukotjo C. Evaluating the performance of large language models versus prosthodontic residents on the 2024 and 2025 National Prosthodontic Resident Examination. The Journal of Prosthetic Dentistry. 2026. doi:10.1016/j.prosdent.2026.03.043

BibTeX
@article{Prasad2026,
  author = {Prasad, Soni and Fang, Qiao and Koseoglu, Merve and Shembesh, Marwa and Gheisarifar, Maryam and Sukotjo, Cortino},
  title = {Evaluating the performance of large language models versus prosthodontic residents on the 2024 and 2025 National Prosthodontic Resident Examination},
  journal = {The Journal of Prosthetic Dentistry},
  year = {2026},
  doi = {10.1016/j.prosdent.2026.03.043}
}
Kostenlos starten →