WOFAPS 2025 8th World Congress of Pediatric Surgery

Asya Eylem Boztaş, İncinur Genişol, Ayşe Demet Payza, Özkan Okur, Arzu Şencan

Behçet Uz Çocuk Hastalıkları Eğitim ve Araştırma Hastanesi Çocuk Cerrahisi Kliniği, İzmir/ Türkiye

Introduction:Artificial intelligence(AI)is playing an increasingly prominent role in advancing the field of medicine. Among AI technologies, ChatGPT stands out as a potential tool in clinical support and education. This study aims to evaluate, the accuracy and quality of responses generated by ChatGPT-4o® to frequently asked questions(FAQs) posed by practicing physicians regarding the initial assessment of pediatric burn injuries, with assessment of pediatric burn specialists.

Methods:Thirty-four FAQs about pediatric burn care were posed to ChatGPT-4o twice, one week apart, blindedly by four experienced pediatric surgeons who work at a national tertiary referral burn center. Questions were divided into five subgroups; initial assessment and triage, fluid resuscitation and hemodynamic management, wound care and infection prevention, pain management and sedation, special situations and follow-up. The results given by chatbot were evaluated by pediatric surgeons using modified five-point DISCERN tool(mDISCERN) for reliability and the Global Quality Scale(GQS) for comprehensive quality of the answers. Inter-rater reliability was measured using intraclass correlation coefficients(ICC).

Results:ChatGPT-4o® demonstrated high-quality and reliable responses to pediatric burn care questions. The median GQS was 4.75(3.50–5.00), with 67.7% of responses scoring ≥4.75, and 41.2% receiving a perfect score of 5.00. The mDISCERN median score was 9.25(7.00–10.00), and 74% of responses scored ≥9.25, reflecting strong informational reliability. There was a very strong correlation between GQS and mDISCERN scores(r=0.858,p<.001), indicating consistent alignment between content quality and reliability. Inter-rater reliability analysis showed good agreement for individual scores(ICC= 0.63) and excellent consistency for average scores(ICC= 0.87, p< .001), supporting the robustness of the reviewers' assessments.

Conclusions:ChatGPT-4o proves to be a high-quality and reliable source of information for the initial evaluation of pediatric burn patients, providing substantial support for healthcare professionals in clinical decision-making. Its consistent accuracy and relevance position it as a promising adjunct tool in pediatric burn care.