525. Evaluating ChatGPT’s Potential as a Burn Evaluation Tool in Acute Care Settings

Abstract Introduction ChatGPT is a widely accessible and versatile large language model. It has the potential to aid burn diagnosis by helping assess severity and guide initial care, particularly among first responders or underserved areas with limited access to specialists. Treatment depends on burn depth as a measure of injury severity. This study evaluates how well ChatGPT agrees with burn surgeons in identifying and managing burns to assess its potential in acute care. Methods A dataset of 20 burn images, previously classified by a burn surgeon by depth, was used to survey other burn surgeons and ChatGPT on burn thickness, need for referral, and treatment approach. Inter-rater reliability was measured using Fleiss’s and Cohen’s Kappa and percent agreement, comparing surgeons’ responses, then surgeon consensus to ChatGPT’s responses. Results There was 80% agreement among the three respondents (Fleiss’s kappa = 0.766 ± 0.046, p