From Algorithmic Accuracy to Clinical Accountability: Reframing Human-AI Collaboration in Emergency Triage

Artificial intelligence (AI) is increasingly being explored to improve patient prioritisation in emergency departments by supporting faster decision-making and consistency in clinical assessment. However, its clinical value cannot be established through algorithmic accuracy alone. This opinion article builds on the review by Mayerhoffer and Ili? to examine the challenges of reliability, transparency and performance in complex triage situations, particularly when AI-generated recommendations conflict with nurses’ clinical assessments. Evidence of variable accuracy and the risk of under-triage and over-triage highlight the need to evaluate AI through clinically meaningful safety outcomes rather than aggregate performance metrics alone. The article emphasises that human oversight must be translated into explicit clinical procedures addressing verification, disagreement, escalation and professional accountability. It advocates a shift in research from comparing AI systems with individual clinicians to evaluating whether AI-assisted nursing triage improves patient prioritisation compared with nurse-only assessment without increasing clinically important errors. Future research should examine triage errors, delays in appropriate care, nurse–AI disagreements, override decisions, resource utilisation and patient outcomes across diverse clinical settings. Ultimately, the safe integration of AI into emergency triage requires structured human–AI collaboration, continuous performance monitoring and preservation of nursing judgement. The priority should be to establish whether AI strengthens clinical decision-making while maintaining patient safety and professional accountability.