AI-assisted treatment decisions for femoral neck fractures: a simulated assessment of ChatGPT-4’s accuracy and comprehensiveness

Abstract

As technology advances, medical knowledge has become more democratized and accessible.An increasing number of patients now rely on search engines and AI chatbots for medical information [13,14], a trend that reduces dependence on traditional healthcare systems and sometimes even challenges physicians'recommendations.ChatGPT-4 is not specifically designed for healthcare and lacks domain-specific training [15,16].The illusion that ChatGPT-4 can generate seemingly reasonable but evidence-based treatment suggestions is particularly dangerous in specialized decision-making [17].Existing studies have raised concerns about its accuracy in triage and diagnosis [9,18].Although AI holds substantial potential in healthcare, integrating it into clinical decision support systems, especially for decisions regarding FNF treatment, remains insufficiently studied.ChatGPT-4 can process vast amounts of data, clinical parameters, and demographic information to provide personalized treatment recommendations [19,20].Currently, research on the application of AI in FNF treatment decisions is limited, with a lack of evaluation reports on the accuracy of ChatGPT-4 recommendations in this field.This study aims to explore this issue.Previous research primarily focused on single-option and closed-ended questions, failing to address the needs of individualized treatment decisions [21,22].In this study, orthopedic surgeons from various medical centers were recruited to conduct a preliminary assessment of ChatGPT-4 accuracy and comprehensiveness in responding to medical queries from simulated patients with FNF.Additionally, our study provides subjective, open-ended explanations to inform treatment decisions for FNF.Professional physicians rated the treatment recommendations and explanations generated by ChatGPT-4, allowing for an analysis of its limitations in producing medical information.This evaluation provides preliminary evidence for the reliability of ChatGPT-4 in clinical fracture treatment.

Ähnliche Arbeiten