<![CDATA[Triage is an essential process in emergency care services aimed at determining patient treatment priorities based on the severity of clinical conditions. This study aimed to analyze the accuracy and level of agreement of Artificial Intelligence (AI), particularly ChatGPT, in triage classification at the Emergency Department of RSUD Gunung Jati compared with medical personnel as the reference standard. This study employed a quantitative observational approach with a retrospective cross-sectional design. Data were obtained from 420 adult patient medical records collected between March and May 2025. Triage classification was conducted using the Simple Triage and Rapid Treatment (START) method by both medical personnel and AI ChatGPT. Data analysis was performed using a confusion matrix and diagnostic tests, including sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), as well as agreement analysis using Cohen’s kappa and weighted Cohen’s kappa. The results showed that AI ChatGPT achieved an overall accuracy of 72.1% (95% CI: 67.8–76.4). The Cohen’s kappa value was 0.313, while the weighted Cohen’s kappa values were 0.338 using linear weighting and 0.384 using quadratic weighting, indicating a fair level of agreement. These findings suggested that ChatGPT had the potential as a supportive system in triage classification; however, it could not yet fully replace the role of medical personnel and still required further development, particularly in cases involving higher levels of emergency severity.]]>
Copyrights © 2026
