Assessment of Artificial Intelligence Chatbot Performance on the Canadian Otolaryngology and Head and Neck Surgery In-Training Exam: Insights from a Comparative Analysis (Preprint)
Notice bibliographique
Résumé
BACKGROUND The introduction of large language models (LLM) has rapidly transformed the field of healthcare. Its performance, often compared to that of physicians, has been greatly scrutinized. ChatGPT-4, a finely tuned supervised model, offers improved reasoning capabilities and visual input analysis. OBJECTIVE The purpose of this study is to evaluate the performance of ChatGPT-4 in the field of otolaryngology and head and neck surgery (OTOHNS) residency training. METHODS A total of 351 questions from the OTOHNS National In-Training Exam (NITE) for 2022 and 2023 were submitted to ChatGPT-4 from April 22nd, 2024, to May 12th, 2024, using a new account. New sessions were used for every question, except for follow-up questions. Answers were independently graded by two reviewers using the official grading rubric and the average score was used. Cohen’s kappa coefficient was used for inter-rater reliability. Anonymized mean exam results from residents who have previously taken this exam were obtained from the lead faculty of the NITE. The sample size was calculated based on the total number of enrolled residents, as indicated on each university’s program website. Z-tests were used to compare ChatGPT-4’s performance to that of residents per sub-specialty and training level. The questions were categorized by type (image or text), task (diagnosis, additional exams, treatment or guidelines), sub-specialty, taxonomic level and prompt length. A one-way ANOVA, independent t-test and two-tailed Pearson correlation was used to examine variations between question categories. IBM SPSS 29 was used. RESULTS ChatGPT-4 scored 66.19% and 64.84% on the 2022 and 2023 exams, respectively. Inter-rater reliability between the two raters was 89.8% (standard error 0.018, P < .001). ChatGPT-4 outperformed the residents on both exams, amongst all training levels and within all sub-specialties except for the general/pediatrics section of the 2023 exam (Z-test -2.37). There were decreasing performance gaps with increasing residency training as per the following Z-scores: PGY-2 16.08, PGY-3 9.31, PGY-4 3.49 in 2022 and PGY-2 15.57, PGY-3 8.60, PGY-4 3.21 in 2023. For the 2022 exam, ChatGPT-4 would rank in the 99th percentile amongst PGY-2, 95th percentile amongst PGY-3 and 73rd percentile amongst PGY-4 classmates. For the 2023 exam, it would rank in the 99th percentile amongst PGY-2, 94th percentile amongst PGY-3 and 71st percentile amongst PGY-4 classmates. ChatGPT-4 performed best on text-based questions (74.3%, P<.001), level one taxonomic questions (75.1%, P<.001) and guideline-based questions (70%, P=.048). It had no significant difference in performance based on sub-specialty (P=.364) or prompt length (P=.385). CONCLUSIONS ChatGPT-4 not only achieved passing grades on two versions of the Canadian OTOHNS NITE, but it also outperformed residents in an outstanding manner, underscoring a critical need to redesign residency assessment methods. CLINICALTRIAL N/A
Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.
Comment cette classification a été obtenuedéplier
Prédiction machine sur la base complète
Imitation des enseignantsNi prévalence calibrée, ni vérité terrain. Validation humaine à venir. Le volet Gemma est une étiquette directe du modèle pour chaque travail de la base, lue sur la notice réduite au titre. Le volet Codex est un classifieur appris des 10 348 étiquettes directes de Codex et calibré sur les taux pondérés de l'échantillon; les champs sans appui suffisant ne portent aucun appel Codex. Le mode candidate est l'union des deux volets; le consensus est leur intersection. Ces sorties portent le statut machine_predicted_unvalidated et ne sont pas des étiquettes humaines.
Scores du classifieur distillé par catégorie (deux têtes)
| Catégorie | Codex | Gemma |
|---|---|---|
| Métarecherche | 0,012 | 0,062 |
| Méta-épidémiologie (sens strict) | 0,000 | 0,000 |
| Méta-épidémiologie (sens large) | 0,000 | 0,001 |
| Bibliométrie | 0,003 | 0,002 |
| Études des sciences et des technologies | 0,001 | 0,001 |
| Communication savante | 0,002 | 0,001 |
| Science ouverte | 0,001 | 0,001 |
| Intégrité de la recherche | 0,001 | 0,001 |
| Charge utile insuffisante (le modèle a refusé de juger) | 0,003 | 0,001 |
Scores machine (provisoires)
Les deux têtes enseignantes du modèle étudiant, lues sur ce travail. Un score ordonne la base pour la relecture; il n'affirme jamais une catégorie, et le statut de validation accompagne chaque rangée tel quel.
Scores de référence d'un modèle non mature (critères de maturité non atteints, 7 itérations). Un score ordonne; il n'affirme jamais une catégorie.
score_only:v0-immature-baseline · tel quel depuis la passe de notation : score_only signifie que le nombre peut ordonner les travaux, et qu'aucune étiquette de catégorie n'en découleClassification
machine, non validéePrédiction automatique; un appel candidat d’une seule source (Gemma direct ou Codex distillé), pas un consensus.
Le détail, modèle par modèle et score par score, se trouve en fin de page sous « Comment cette classification a été obtenue ».