Erschienen in:
24.10.2023 | Laryngology
ChatGPT performance in laryngology and head and neck surgery: a clinical case-series
verfasst von:
Jerome R. Lechien, Bianca M. Georgescu, Stephane Hans, Carlos M. Chiesa-Estomba
Erschienen in:
European Archives of Oto-Rhino-Laryngology
|
Ausgabe 1/2024
Einloggen, um Zugang zu erhalten
Abstract
Objectives
To study the performance of ChatGPT in the management of laryngology and head and neck (LHN) cases.
Methods
History and clinical examination of patients consulting at the Otolaryngology-Head and Neck Surgery department were presented to ChatGPT, which was interrogated for differential diagnosis, management, and treatment. The ChatGPT performance was assessed by two blinded board-certified otolaryngologists using the following items of a composite score and the Ottawa Clinic Assessment Tool: differential diagnosis; additional examination; and treatment options. The complexity of clinical cases was evaluated with the Amsterdam Clinical Challenge Scale test.
Results
Forty clinical cases were submitted to ChatGPT, accounting for 14 (35%), 12 (30%), and 14 (35%) easy, moderate and difficult cases, respectively. ChatGPT indicated a significant higher number of additional examinations compared to practitioners (p = 0.001). There was a significant agreement between practitioners and ChatGPT for the indication of some common examinations (audiometry, ultrasonography, biopsy, gastrointestinal endoscopy or videofluoroscopy). ChatGPT never indicated some important additional examinations (PET–CT, voice quality assessment, or impedance-pH monitoring). ChatGPT reported highest performance in the proposition of the primary (90%) or the most plausible differential diagnoses (65%), and the therapeutic options (60–68%). The ChatGPT performance in the indication of additional examinations was lowest.
Conclusions
ChatGPT is a promising adjunctive tool in LHN practice, providing extensive documentation about disease-related additional examinations, differential diagnoses, and treatments. The ChatGPT is more efficient in diagnosis and treatment, rather than in the selection of the most adequate additional examination.