Evaluating ChatGPT Performance on the Orthopaedic In-Training Examination
JBJS Open Access · 2023-07 · 8 (3)
DOI: 10.2106/jbjs.oa.23.00056 ↗Kurz gefasst
Übersichtsarbeit in der Zeitschrift JBJS Open Access, 2023-07. Verfasst von Justin E Kung und 4 weiteren Autor:innen. Beteiligte Einrichtung: Department of Orthopedic Surgery, Prisma Health-Midlands University of South Carolina, Columbia, South Carolina. Die Arbeit wurde bisher 94-mal zitiert (Crossref) und nennt 12 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Background
Artificial intelligence (AI) holds potential in improving medical education and healthcare delivery. ChatGPT is a state-of-the-art natural language processing AI model which has shown impressive capabilities, scoring in the top percentiles on numerous standardized examinations, including the Uniform Bar Exam and Scholastic Aptitude Test. The goal of this study was to evaluate ChatGPT performance on the Orthopaedic In-Training Examination (OITE), an assessment of medical knowledge for orthopedic residents.
Methods
OITE 2020, 2021, and 2022 questions without images were inputted into ChatGPT version 3.5 and version 4 (GPT-4) with zero prompting. The performance of ChatGPT was evaluated as a percentage of correct responses and compared with the national average of orthopedic surgery residents at each postgraduate year (PGY) level. ChatGPT was asked to provide a source for its answer, which was categorized as being a journal article, book, or website, and if the source could be verified. Impact factor for the journal cited was also recorded.
Results
ChatGPT answered 196 of 360 answers correctly (54.3%), corresponding to a PGY-1 level. ChatGPT cited a verifiable source in 47.2% of questions, with an average median journal impact factor of 5.4. GPT-4 answered 265 of 360 questions correctly (73.6%), corresponding to the average performance of a PGY-5 and exceeding the corresponding passing score for the American Board of Orthopaedic Surgery Part I Examination of 67%. GPT-4 cited a verifiable source in 87.9% of questions, with an average median journal impact factor of 5.2.
Conclusions
ChatGPT performed above the average PGY-1 level and GPT-4 performed better than the average PGY-5 level, showing major improvement. Further investigation is needed to determine how successive versions of ChatGPT would perform and how to optimize this technology to improve medical education. Clinical Relevance: AI has the potential to aid in medical education and healthcare delivery.
Autor:innen und Institutionen
- Department of Orthopedic Surgery, Prisma Health-Midlands University of South Carolina, Columbia, South Carolina.
- University of South Carolina School of Medicine, Columbia, South Carolina.
ORCID-Profile: Justin E Kung, Christopher Marshall, Chase Gauthier, Tyler A Gonzalez
Zitieren
APA
Kung, J. E., Marshall, C., Gauthier, C., Gonzalez, T. A., & Jackson, J. B. (2023). Evaluating ChatGPT Performance on the Orthopaedic In-Training Examination. JBJS Open Access, 8(3). https://doi.org/10.2106/jbjs.oa.23.00056
Vancouver
Kung JE, Marshall C, Gauthier C, Gonzalez TA, Jackson JB. Evaluating ChatGPT Performance on the Orthopaedic In-Training Examination. JBJS Open Access. 2023;8(3). doi:10.2106/jbjs.oa.23.00056
BibTeX
@article{Kung2023,
author = {Kung, Justin E and Marshall, Christopher and Gauthier, Chase and Gonzalez, Tyler A and Jackson, J Benjamin},
title = {Evaluating ChatGPT Performance on the Orthopaedic In-Training Examination},
journal = {JBJS Open Access},
year = {2023},
volume = {8},
number = {3},
doi = {10.2106/jbjs.oa.23.00056}
}
EUVIMED