Diagnostic Accuracy of GPT-4o and Claude 4.6 Sonnet in Turkish ED Anamnesis Notes (LLM-ED-DX-TR)
3 giugno 2026 aggiornato da: Emir Ünal, Marmara University Pendik Training and Research Hospital
Diagnostic Accuracy of Large Language Models From Emergency Department Anamnesis Notes: A Comparison of GPT-4o and Claude 4.6 Sonnet With Emergency Medicine Specialists
This retrospective diagnostic accuracy study evaluates the ability of two large language models (LLMs) - GPT-4o (gpt-4o-2024-11-20; OpenAI) and Claude 4.6 Sonnet (claude-sonnet-4-6; Anthropic) - to generate correct diagnoses from anonymized Turkish-language emergency department (ED) anamnesis notes, and compares their performance with the diagnosis entered by the treating emergency physician.
A consensus gold standard is established by three independent board-certified emergency medicine specialists who blindly review each note and vote on the primary diagnosis using ICD-10 three-character codes; the majority vote (at least 2 of 3 specialists agreeing) constitutes the reference standard.
Both LLMs are evaluated using a standardized zero-shot direct prompting strategy (temperature=0, stateless API sessions).
The primary outcome is diagnostic accuracy (proportion of ICD-10 chapter-level matches) and Cohen's kappa for each LLM against the gold standard.
Secondary outcomes include top-3 accuracy, treating physician accuracy, inter-model agreement, and subgroup analyses by ESI triage level and ICD-10 chapter.
Inter-rater reliability among the three specialists is quantified using Fleiss' kappa.
Analyses are performed in Jamovi.
This study represents the first evaluation of LLM diagnostic accuracy using Turkish-language clinical notes and the first to benchmark LLM performance against an independent three-specialist majority-vote gold standard rather than against the treating physician's own diagnosis.
Panoramica dello studio
Stato
Non ancora reclutamento
Descrizione dettagliata
STUDY DESIGN: Retrospective diagnostic accuracy study, STARD-AI 2025 reporting, single center, cohort design.
AI INDEX TESTS: (1) GPT-4o (model version gpt-4o-2024-11-20; OpenAI API). (2) Claude 4.6 Sonnet (model version claude-sonnet-4-6; Anthropic API). Both accessed via Python (Google Colab). Temperature=0 for reproducibility. Zero-shot, stateless sessions - no cross-case context. No task-specific fine-tuning or additional training applied; models used as-is via API.
MODEL INTERPRETABILITY: Model interpretability analyses (such as SHAP, Grad-CAM, or layer-attribute visualizations) are not applicable to this study. Because GPT-4o and Claude 4.6 Sonnet are accessed as black-box models through proprietary, closed-source commercial APIs, internal model weights, gradients, and attention architectures are structurally inaccessible for post-hoc interpretability computations.
REFERENCE STANDARD: Three board-certified emergency medicine specialists independently evaluate each anonymized note, blinded to the original physician diagnosis and to each other. Primary diagnosis assigned by at least 2/3 specialists (majority vote) constitutes the gold standard. A 5-case calibration session precedes the main evaluation.
DATA PRIVACY: All anamnesis notes are fully de-identified (name, ID number, date of birth, physician name removed) prior to processing. De-identified notes are stored in a password-protected encrypted database. Only de-identified text is transmitted to LLM APIs - no personal health data. Compliant with Turkish Personal Data Protection Law (KVKK No. 6698).
PATIENT AND PUBLIC INVOLVEMENT: Not applicable. This retrospective study uses fully anonymized existing records; no patient or public involvement in design or conduct.
DATA SHARING: Anonymized dataset will be shared via Zenodo upon article acceptance. Statistical analysis code (Jamovi project files and Python prompt scripts) will be available on GitHub.
Tipo di studio
Osservativo
Iscrizione (Stimato)
600
Contatti e Sedi
Questa sezione fornisce i recapiti di coloro che conducono lo studio e informazioni su dove viene condotto lo studio.
Contatto studio
- Nome: Emir Ünal, Assistant Professor
- Numero di telefono: +905327766010
- Email: emirunal@gmail.com
Backup dei contatti dello studio
- Nome: Emir Unal, Assistant Professor
- Email: emirunal@gmail.com
Luoghi di studio
-
-
Istanbul
-
Istanbul, Istanbul, Turchia (Türkiye), 34899
- Marmara University Pendik Training and Research Hospital
-
Contatto:
- Emir ünal
- Email: emirunal@gmail.com
-
-
Criteri di partecipazione
I ricercatori cercano persone che corrispondano a una certa descrizione, chiamata criteri di ammissibilità. Alcuni esempi di questi criteri sono le condizioni generali di salute di una persona o trattamenti precedenti.
Criteri di ammissibilità
Età idonea allo studio
- Adulto
- Adulto più anziano
Accetta volontari sani
No
Metodo di campionamento
Campione non probabilistico
Popolazione di studio
The study population comprises consecutive adult patients (aged 18 years and older) who presented to the emergency department of a tertiary care training and research hospital and had their encounters fully documented in the hospital information system (HBYS).
Eligible individuals must have a complete electronic anamnesis note containing the chief complaint, history of present illness, and clinical presentation, alongside a definitive primary ICD-10 diagnosis finalized by the treating emergency physician at file closure.
The population excludes pediatric cases, patients triaged to high-acuity resuscitation areas (ESI level 1), and clinical notes with fewer than 50 words or insufficient clinical content.
Descrizione
INCLUSION CRITERIA:
- Adult patients (aged 18 years and older) presenting to the emergency department.
- Complete electronic health record available in the hospital information system (HBYS) containing a detailed anamnesis note with chief complaint, symptom duration, associated symptoms, and relevant medical history.
- A definitive primary diagnosis recorded by the treating emergency physician using ICD-10 codes at the time of patient file closure.
EXCLUSION CRITERIA:
- Emergency department anamnesis notes containing fewer than 50 words or completely lacking substantive clinical content[cite: 1].
- Pediatric cases (age under 18 years)[cite: 1].
- Patients critically ill and triaged to high-acuity resuscitation areas (Emergency Severity Index [ESI] level 1)[cite: 1].
- Clinical notes containing residual identifying information that cannot be fully de-identified, preventing compliance with data privacy regulations[cite: 1].
- Non-independent clinical notes consisting solely of a brief cross-reference to a prior hospital visit without a new history entry[cite: 1].
Piano di studio
Questa sezione fornisce i dettagli del piano di studio, compreso il modo in cui lo studio è progettato e ciò che lo studio sta misurando.
Come è strutturato lo studio?
Dettagli di progettazione
Coorti e interventi
Gruppo / Coorte |
|---|
|
Emergency Department Patient Cohort
Consecutive adult patients presenting to the emergency department with a fully documented electronic anamnesis note and a definitive primary ICD-10 diagnosis
|
Cosa sta misurando lo studio?
Misure di risultato primarie
Misura del risultato |
Misura Descrizione |
Lasso di tempo |
|---|---|---|
|
Diagnostic Accuracy of GPT-4o for ICD-10 Chapter-Level Diagnosis
Lasso di tempo: At the time of single-session algorithmic evaluation (each case evaluated once following data extraction in June 2026).
|
Proportion of cases in which GPT-4o primary (rank 1) diagnosis matches the 3-specialist majority-vote gold standard at the ICD-10 chapter level (22 categories).
Range: 0 to 1.00.
|
At the time of single-session algorithmic evaluation (each case evaluated once following data extraction in June 2026).
|
|
Diagnostic Accuracy of Claude 4.6 Sonnet for ICD-10 Chapter-Level Diagnosis
Lasso di tempo: At the time of single-session algorithmic evaluation (each case evaluated once following data extraction in June 2026).
|
Proportion of cases in which Claude 4.6 Sonnet primary (rank 1) diagnosis matches the 3-specialist majority-vote gold standard at the ICD-10 chapter level (22 categories).
Range: 0 to 1.00.
|
At the time of single-session algorithmic evaluation (each case evaluated once following data extraction in June 2026).
|
Misure di risultato secondarie
Misura del risultato |
Misura Descrizione |
Lasso di tempo |
|---|---|---|
|
Cohen's Kappa Between GPT-4o Primary Diagnosis and Gold Standard
Lasso di tempo: At the time of algorithmic evaluation (June-July 2026)
|
Kappa coefficient measuring agreement between GPT-4o rank-1 ICD-10 chapter and the 3-specialist gold standard .
Interpreted per Landis & Koch (1977): <=0.20 slight; 0.21-0.40
fair; 0.41-0.60
moderate; 0.61-0.80
substantial; >0.80 almost perfect .
Range: -1.00 to 1.00 .
|
At the time of algorithmic evaluation (June-July 2026)
|
|
Cohen's Kappa Between Claude 4.6 Sonnet Primary Diagnosis and Gold Standard
Lasso di tempo: At the time of algorithmic evaluation (June-July 2026)
|
appa coefficient measuring agreement between Claude 4.6 Sonnet rank-1 ICD-10 chapter and the 3-specialist gold standard .
Interpreted per Landis & Koch (1977): <=0.20 slight; 0.21-0.40
fair; 0.41-0.60
moderate; 0.61-0.80
substantial; >0.80 almost perfect .
Range: -1.00 to 1.00
|
At the time of algorithmic evaluation (June-July 2026)
|
|
Top-3 Diagnostic Accuracy of GPT-4o
Lasso di tempo: At the time of algorithmic evaluation (June-July 2026)
|
Proportion of cases in which the 3-specialist gold standard diagnosis appears within GPT-4o's ranked list of three differential diagnoses .
Range: 0 to 1.00
|
At the time of algorithmic evaluation (June-July 2026)
|
|
Top-3 Diagnostic Accuracy of Claude 4.6 Sonnet
Lasso di tempo: At the time of algorithmic evaluation (June-July 2026)
|
Proportion of cases in which the 3-specialist gold standard diagnosis appears within Claude 4.6 Sonnet's ranked list of three differential diagnoses[cite: 1].
Range: 0 to 1.00
|
At the time of algorithmic evaluation (June-July 2026)
|
|
Treating Physician Diagnostic Accuracy Against Gold Standard
Lasso di tempo: At the time of the original clinical encounter (retrospective data spanning August-December 2025)
|
Proportion of cases in which the ICD-10 code entered by the treating emergency physician at file closure matches the 3-specialist majority-vote gold standard at the chapter level[cite: 1].
Range: 0 to 1.00
|
At the time of the original clinical encounter (retrospective data spanning August-December 2025)
|
Collaboratori e investigatori
Qui è dove troverai le persone e le organizzazioni coinvolte in questo studio.
Investigatori
- Investigatore principale: Emir Ünal, Marmara University
Pubblicazioni e link utili
La persona responsabile dell'inserimento delle informazioni sullo studio fornisce volontariamente queste pubblicazioni. Questi possono riguardare qualsiasi cosa relativa allo studio.
Pubblicazioni generali
- Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HC, Kressel HY, Rifai N, Golub RM, Altman DG, Hooft L, Korevaar DA, Cohen JF; STARD Group. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015 Oct 28;351:h5527. doi: 10.1136/bmj.h5527.
- Newman-Toker DE, Peterson SM, Badihian S, Hassoon A, Nassery N, Parizadeh D, Wilson LM, Jia Y, Omron R, Tharmarajah S, Guerin L, Bastani PB, Fracica EA, Kotwal S, Robinson KA. Diagnostic Errors in the Emergency Department: A Systematic Review [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2022 Dec. Report No.: 22(23)-EHC043. Available from http://www.ncbi.nlm.nih.gov/books/NBK588118/
- Wei J et al. Chain-of-thought prompting elicits reasoning in LLMs. NeurIPS. 2022;35:24824-24837.
- Sounderajah V, Guni A, Liu X, Collins GS, Karthikesalingam A, Markar SR, Golub RM, Denniston AK, Shetty S, Moher D, Bossuyt PM, Darzi A, Ashrafian H; STARD-AI Steering Committee. The STARD-AI reporting guideline for diagnostic accuracy studies using artificial intelligence. Nat Med. 2025 Oct;31(10):3283-3289. doi: 10.1038/s41591-025-03953-8. Epub 2025 Sep 15.
- Niset A, Melot I, Pireau M, Englebert A, Scius N, Flament J, El Hadwe S, Al Barajraji M, Thonon H, Barrit S. Grounded large language models for diagnostic prediction in real-world emergency department settings. JAMIA Open. 2025 Oct 21;8(5):ooaf119. doi: 10.1093/jamiaopen/ooaf119. eCollection 2025 Oct.
- Williams CYK, Miao BY, Kornblith AE, Butte AJ. Evaluating the use of large language models to provide clinical recommendations in the Emergency Department. Nat Commun. 2024 Oct 8;15(1):8236. doi: 10.1038/s41467-024-52415-1.
- Hoppe JM, Auer MK, Struven A, Massberg S, Stremmel C. ChatGPT With GPT-4 Outperforms Emergency Department Physicians in Diagnostic Accuracy: Retrospective Analysis. J Med Internet Res. 2024 Jul 8;26:e56110. doi: 10.2196/56110.
- Kanjee Z, Crowe B, Rodman A. Accuracy of a Generative Artificial Intelligence Model in a Complex Diagnostic Challenge. JAMA. 2023 Jul 3;330(1):78-80. doi: 10.1001/jama.2023.8288.
- Takita H, Kabata D, Walston SL, Tatekawa H, Saito K, Tsujimoto Y, Miki Y, Ueda D. A systematic review and meta-analysis of diagnostic performance comparison between generative AI and physicians. NPJ Digit Med. 2025 Mar 22;8(1):175. doi: 10.1038/s41746-025-01543-z.
- Shan G, Chen X, Wang C, Liu L, Gu Y, Jiang H, Shi T. Comparing Diagnostic Accuracy of Clinical Professionals and Large Language Models: Systematic Review and Meta-Analysis. JMIR Med Inform. 2025 Apr 25;13:e64963. doi: 10.2196/64963.
- Taylor RA, Sangal RB, Smith ME, Haimovich AD, Rodman A, Iscoe MS, Pavuluri SK, Rose C, Janke AT, Wright DS, Socrates V, Declan A. Leveraging artificial intelligence to reduce diagnostic errors in emergency medicine: Challenges, opportunities, and future directions. Acad Emerg Med. 2025 Mar;32(3):327-339. doi: 10.1111/acem.15066. Epub 2024 Dec 15.
Studiare le date dei record
Queste date tengono traccia dell'avanzamento della registrazione dello studio e dell'invio dei risultati di sintesi a ClinicalTrials.gov. I record degli studi e i risultati riportati vengono esaminati dalla National Library of Medicine (NLM) per assicurarsi che soddisfino specifici standard di controllo della qualità prima di essere pubblicati sul sito Web pubblico.
Studia le date principali
Inizio studio (Stimato)
1 giugno 2026
Completamento primario (Stimato)
1 luglio 2026
Completamento dello studio (Stimato)
1 ottobre 2026
Date di iscrizione allo studio
Primo inviato
3 giugno 2026
Primo inviato che soddisfa i criteri di controllo qualità
3 giugno 2026
Primo Inserito (Effettivo)
8 giugno 2026
Aggiornamenti dei record di studio
Ultimo aggiornamento pubblicato (Effettivo)
8 giugno 2026
Ultimo aggiornamento inviato che soddisfa i criteri QC
3 giugno 2026
Ultimo verificato
1 giugno 2026
Maggiori informazioni
Termini relativi a questo studio
Termini MeSH pertinenti aggiuntivi
Altri numeri di identificazione dello studio
- 09.2026.26-0514
Informazioni su farmaci e dispositivi, documenti di studio
Studia un prodotto farmaceutico regolamentato dalla FDA degli Stati Uniti
No
Studia un dispositivo regolamentato dalla FDA degli Stati Uniti
No
Queste informazioni sono state recuperate direttamente dal sito web clinicaltrials.gov senza alcuna modifica. In caso di richieste di modifica, rimozione o aggiornamento dei dettagli dello studio, contattare register@clinicaltrials.gov. Non appena verrà implementata una modifica su clinicaltrials.gov, questa verrà aggiornata automaticamente anche sul nostro sito web .