CoChat
Sign Up
PreprintCited by 177Artificial Intelligence in Healthcare and EducationInnovations in Medical EducationArtificial Intelligence in Education

How Does ChatGPT Perform on the Medical Licensing Exams? The Implications of Large Language Models for Medical Education and Knowledge Assessment

Open access
Gilson, Aidan, Safranek, Conrad, Huang, Thomas, Socrates, Vimig, Chi, Ling, Taylor, R. Andrew, Chartash, David·2022·medRxiv
DOI 10.1101/2022.12.23.22283901
Open this paper in CoChat — free accountOpen PDF

Abstract

Background: ChatGPT is a 175 billion parameter natural language processing model which can generate conversation style responses to user input. Objective: To evaluate the performance of ChatGPT on questions within the scope of United States Medical Licensing Examination (USMLE) Step 1 and Step 2 exams, as well as analyze responses for user interpretability. Methods: We used two novel sets of multiple choice questions to evaluate ChatGPT’s performance, each with questions pertaining to Step 1 and Step 2. The first was derived from AMBOSS, a commonly used question bank for medical students, which also provides statistics on question difficulty and the performance on an exam relative to the userbase. The second, was the National Board of Medical Examiners (NBME) Free 120-question exams. After prompting ChatGPT with each question, ChatGPT’s selected answer was recorded, and the text output evaluated across three qualitative metrics: logical justification of the answer selected, presence of information internal to the question, and presence of information external to the question. Results: On the four datasets, AMBOSS-Step1, AMBOSS-Step2, NBME-Free-Step1, and NBMEFree-Step2, ChatGPT achieved accuracies of 44%, 42%, 64.4%, and 57.8%. The model demonstrated a significant decrease in performance as question difficulty increased (P=.012) within the AMBOSSStep1 dataset. We found logical justification for ChatGPT’s answer selection was present in 100% of outputs. Internal information to the question was present in > 90% of all questions. The presence of information external to the question was respectively 54.5% and 27% lower for incorrect relative to correct answers on the NBME-Free-Step1 and NBME-Free-Step2 datasets (P<=.001). Conclusion: ChatGPT marks a significant improvement in natural language processing models on the tasks of medical question answering. By performing at greater than 60% threshold on the NBME-FreeStep-1 dataset we show that the model is comparable to a third year medical student. Additionally, due to the dialogic nature of the response to questions, we demonstrate ChatGPT’s ability to provide reasoning and informational context across the majority of answers. These facts taken together make a compelling case for the potential applications of ChatGPT as a medical education tool.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1101/2022.12.23.22283901

Cites

3 of 3 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Studies comparing ChatGPT's performance against previous large language models all indicated a significant improvement in the task of question answering.

    “While these studies demonstrated mixed results with regards to ChatGPT’s performance in standardized exams against that of students, studies which specifically compared the model’s performance against that of previous large language models all indicated a significant improvement in the task of question answering20,21,23.”

    From Perception, performance, and detectability of conversational artificial intelligence across 32 university courses (2023) · cites this paper as [20]

  • The literature lacks a systematic study comparing the performance of ChatGPT against that of students across various disciplines at the same institution.

    “Despite these examples, the literature still lacks a systematic study comparing the performance of ChatGPT against that of students across various disciplines at the same institution.”

    From Perception, performance, and detectability of conversational artificial intelligence across 32 university courses (2023) · cites this paper as [20]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 8

  1. Savery Max, Abacha Asma Ben, Gayen Soumya, Demner-Fushman Dina. Question-driven summarization of answers to consumer health questions Scientific Data. 2020;7:1-9.
  2. Logé Cécile, Ross Emily, Dadey David Yaw Amoah, et al. Q-Pain: A Question Answering Dataset to Measure Social Bias in Pain Management arXiv preprint arXiv:2108.01764. 2021. · arXiv
  3. Gutiérrez Bernal Jiménez, McNeal Nikolas, Washington Clay, et al. Thinking about GPT- 3 In-Context Learning for Biomedical IE? Think Again arXiv preprint arXiv:2203.08410. 2022. · arXiv
  4. Google Scholar https://scholar.google.com/ 2022.
  5. Jin Qiao, Dhingra Bhuwan, Liu Zhengping, Cohen William W, Lu Xinghua. PubMedQA: A dataset for biomedical research question answering arXiv preprint arXiv:1909.06146. 2019. · arXiv
  6. Jin Di, Pan Eileen, Oufattole Nassim, Weng Wei-Hung, Fang Hanyi, Szolovits Peter. What Disease Does This Patient Have? A Large-Scale Open Domain Question Answering Dataset from Medical Exams Applied Sciences. 2021;11:6421.
  7. Springer Leonard, Stanne Mary Elizabeth, Donovan Samuel S.. Effects of Small-Group Learning on Undergraduates in Science, Mathematics, Engineering, and Technology: A Meta-Analysis Review of Educational Research. 1999;69:21-51.
  8. Neville Alan J, Norman Geoff R. PBL in the undergraduate MD program at McMaster University: three iterations in three decades Academic Medicine. 2007;82:370-374.

Cited by

Sorted by citations

Related

Authors
Gilson, A., Safranek, C., Huang, T., Socrates, V., Chi, L., Taylor, R. A., & Chartash, D.
Affiliations
  • Yale University (US)
  • University College Dublin (IE)
Corresponding author
Chartash, David
ORCID
Venue
medRxiv
Publisher
openRxiv
Published
2022-12-26
DOI
10.1101/2022.12.23.22283901
OpenAlex
W4313232683
Cited by
172
References
3
Citations by year
2026: 9 · 2025: 29 · 2024: 52 · 2023: 82
Type
posted content
Language
en
Open access
Licence
cc-by-nc-nd
Locations
medRxiv (acceptedVersion, cc-by-nc-nd)
Funders
  • Yale University
  • National Institutes of Health
  • National Institute of Diabetes and Digestive and Kidney Diseases (T35DK104689)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Innovations in Medical Education
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
Keywords
ChatGPT, large language models, USMLE Step 1, USMLE Step 2, medical licensing examinations, multiple-choice question answering, medical education, knowledge assessment, question difficulty, AMBOSS, medical students

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review