Accuracy of ChatGPT in head and neck oncological board decisions: preliminary findings

Authors:

Abstract
Objectives To evaluate the ChatGPT-4 performance in oncological board decisions. Methods Twenty medical records of patients with head and neck cancer were evaluated by ChatGPT-4 for additional examinations, management, and therapeutic approaches. The ChatGPT-4 propositions were assessed with the Artificial Intelligence Performance Instrument. The stability of ChatGPT-4 was evaluated through regenerated answers at 1-day interval. Results ChatGPT-4 provided adequate explanations for cTNM staging in 19 cases (95%). ChatGPT-4 proposed a significant higher number of additional examinations than practitioners (72 versus 103; p = 0.001). ChatGPT-4 indications of endoscopy–biopsy, HPV research, ultrasonography, and PET–CT were consistent with the oncological board decisions. The therapeutic propositions of ChatGPT-4 were accurate in 13 cases (65%). Most additional examination and primary treatment propositions were consistent throughout regenerated response process. Conclusions ChatGPT-4 may be an adjunctive theoretical tool in oncological board simple decisions.

This publication has 15 references indexed in Scilit: