MétaCan
Menu
← Back to cohort
Record W4416618182 · doi:10.2196/79465

The Validity of Generative Artificial Intelligence in Evaluating Medical Students in Objective Structured Clinical Examination: Experimental Study

2025· article· en· W4416618182 on OpenAlexvenueno aff
Masashi Yokose, Takanobu Hirosawa, Tetsu Sakamoto, Ren Kawamura, Yudai Suzuki, Yukinori Harada, Taro Shimizu

Bibliographic record

VenueJMIR Formative Research · 2025
Typearticle
Languageen
FieldMedicine
TopicArtificial Intelligence in Healthcare and Education
Canadian institutionsnot available
Fundersnot available
KeywordsClinical trialArtificial neural networkMedical informationMedical researchMEDLINEReliability (semiconductor)

Abstract

fetched live from OpenAlex

BACKGROUND: The Objective Structured Clinical Examination (OSCE) has been widely used to evaluate students in medical education. However, it is resource-intensive, presenting challenges in implementation. We hypothesized that generative artificial intelligence (AI) such as ChatGPT-4 could serve as a complementary assessor and alleviate the burden of physicians in evaluating OSCE. OBJECTIVE: By comparing the evaluation scores between generative AI and physicians, this study aims to evaluate the validity of generative AI as a complementary assessor for OSCE. METHODS: This experimental study was conducted at a medical university in Japan. We recruited 11 fifth-year medical students during the general internal medicine clerkship from April 2023 to December 2023. Participants conducted a mock medical interview with a patient experiencing abdominal pain and wrote patient notes. Four physicians independently evaluated the participants by reviewing medical interview videos and patient notes, while ChatGPT-4 was provided with interview transcripts and notes. Evaluations were conducted using the 6-domain rubric (patient care and communication, history taking, physical examination, patient notes, clinical reasoning, and management). Each domain was scored using a 6-point Likert scale, ranging from 1 (very poor) to 6 (excellent). Median scores were compared using the Wilcoxon signed-rank test, and the agreement between ChatGPT-4 and physicians was assessed using intraclass correlation coefficients (ICCs). All P values <.05 were considered statistically significant. RESULTS: Although ChatGPT-4 assigned higher scores than physicians in terms of physical examination (median 4.0, IQR 4.0-5.0 vs median 4.0, IQR 3.0-4.0; P=.02), patient notes (median 6.0, IQR 5.0-6.0 vs median 4.0, IQR 4.0-4.0; P=.002), clinical reasoning (median 5.0, IQR 5.0-5.0 vs median 4.0, IQR 3.0-4.0; P<.001), and management (median 6.0, IQR 5.0-6.0 vs median 4.0, IQR 2.5-4.5; P=.002), there were no significant differences in the scores of patient care and communication (median 5.0, IQR 5.0-5.0 vs median 5.0, IQR 4.0-5.0; P=.06) and history taking (median 5.0, IQR 4.0-5.0 vs median 5.0, IQR 4.0-5.0; P>.99), respectively. ICC values were low in all domains, except for history taking, where the agreement was still poor (ICC=0.36, 95% CI -0.32 to 0.78). CONCLUSIONS: ChatGPT-4 produced higher evaluation scores than physicians in several OSCE domains, though the agreement between them was poor. Although these preliminary results suggest that generative AI may be able to support assessment in some domains of OSCE, further research is needed to establish its reproducibility and validity. Generative AI like ChatGPT-4 shows potential as a complementary assessor for OSCE. TRIAL REGISTRATION: University Hospital Medical Information Network Clinical Trials Registry UMIN000050489; https://center6.umin.ac.jp/cgi-open-bin/ctr/ctr_his_list.cgi?recptno=R000057513.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.035
metaresearch head score (Gemma)0.102
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesMetaresearch
Consensus categoriesnone
DomainCandidate signal: Evaluation · Consensus signal: none
Study designCandidate signal: Bench or experimental · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.965
Threshold uncertainty score0.185

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0350.102
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0010.001
Science and technology studies0.0010.003
Scholarly communication0.0020.002
Open science0.0010.002
Research integrity0.0010.001
Insufficient payload (model declined to judge)0.0010.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.502
GPT teacher head0.673
Teacher spread0.171 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

Study designBench or experimental
DomainEvaluation
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations8
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Formative Research→Same topicArtificial Intelligence in Healthcare and Education→French-language works237,207→