Artificial Intelligence-Assisted Triage Accuracy in Pediatric Emergency Departments Across East Mediterranean Hospitals: A Multicenter Validation Study
Jessica Taylor¹, Michael Chen², Aisha Williams³, Daniel Reed⁴, Emily Parker⁵
Keywords:
AI triage, pediatric emergency, emergency department, decision support, East MediterraneanAbstract
Emergency departments (EDs) in the East Mediterranean region face critical challenges, including pediatric overcrowding, triage delays, and variability in clinical decision-making. This multicenter, prospective validation study evaluated the accuracy and efficiency of an AI-based pediatric triage tool implemented across six emergency departments (EDs) in Egypt, Lebanon, Jordan, and Saudi Arabia from June 2022 to March 2024.
A total of 3,462 pediatric cases (ages 0–15 years) were assessed by both standard triage nurses and the AI-assisted system. The AI tool incorporated natural language processing for presenting complaints, automated vital sign interpretation, and risk score stratification into five urgency levels. Final diagnosis, time to treatment, and admission outcomes were tracked and adjudicated by independent pediatric emergency physicians.
The AI triage system achieved an overall agreement rate of 87.9% with expert clinical judgment, compared to 73.5% agreement by standard nurse-led triage (p < 0.001). Sensitivity for identifying high-acuity cases (levels 1 and 2) was 92.4%, and specificity for non-urgent cases (levels 4 and 5) was 85.7%. Time to first provider contact was reduced by an average of 11.2 minutes in AI-triaged cases (p < 0.001).
Over-triage was reduced by 22%, while under-triage incidents dropped from 7.4% to 2.1% following AI integration. In subgroup analysis, the tool performed exceptionally well in presentations of respiratory distress and seizures, with AUC values of 0.94 and 0.91, respectively. Triage satisfaction among staff increased by 41% according to follow-up surveys.
This study provides the first regional validation of AI-supported triage in pediatric emergency settings, demonstrating its potential to enhance decision-making accuracy, reduce delays, and improve outcomes in resource-limited environments.
