Emergency departments worldwide face immense pressure from increasing patient volumes and demands for rapid, accurate care. Inaccurate or delayed triage can lead to suboptimal patient outcomes, prolonged wait times, and inefficient use of costly resources. Artificial intelligence (AI) offers a transformative approach to augment human capabilities in high-stakes environments like the ED. Developing and validating AI-powered triage systems is crucial for modernizing emergency care, ensuring patient safety, and improving operational efficiency in overstretched healthcare facilities.
The aim of the study.
This study aimed to assess the impact and efficacy of a bespoke AI-powered triage system on patient wait times, triage accuracy, and overall emergency department throughput in a multi-center, real-world clinical setting.
Material and methods.
A prospective, quasi-experimental study was implemented in three diverse academic emergency departments in Beijing, Geneva, and Riyadh over a six-month period. The study involved 9,500 adult patients presenting to the ED. For each patient, standard human triage was performed, and concurrently, an AI-powered triage system, trained on extensive historical patient data (symptoms, vital signs, chief complaints, past medical history), generated a recommended triage category. The primary outcome was the discordance rate between human and AI triage, specifically focusing on cases where AI recommended a higher acuity level. Secondary outcomes included patient wait times to initial physician assessment, ED length of stay, and the incidence of critical patient deterioration during the waiting period. Statistical analyses included chi-square tests for categorical comparisons and t-tests for continuous variables, with a significance level of p < 0.05.
The AI triage system demonstrated a high degree of concordance with human triage (overall accuracy 88%). Notably, the AI system reclassified 15.3% (n=1454) of patients to a higher acuity level than initial human assessment, with 7.1% (n=674) of these being subsequently confirmed by physician review as true higher acuity cases (e.g., severe sepsis, acute coronary syndrome miscategorized). This resulted in an average reduction of 25 minutes (SD 11) in time to physician assessment for these high-acuity, AI-identified patients (p < 0.001). Overall ED length of stay decreased by an average of 18 minutes (SD 7) across all patients when the AI system's recommendations were considered by human triagers (p = 0.04). The rate of critical adverse events occurring during the waiting period was significantly lower in the AI-supported group (0.8% vs. 1.5% in control periods, p = 0.02).
The integration of AI-powered triage systems in emergency departments can significantly enhance the accuracy of patient prioritization and demonstrably reduce wait times for high-acuity cases. This technology offers a robust solution to alleviate ED overcrowding, improve patient safety, and optimize resource utilization. Further research is warranted to explore the long-term impact on patient outcomes and to refine the integration of AI recommendations into clinical workflows.
1. Chen L, Wang M. AI in Emergency Medicine: A Paradigm Shift. Ann Emerg Med. 2023;82(1):45-53.
2. Dubois S, Moreau P. Overcoming ED Crowding: Digital Solutions. J Trauma Emerg Care. 2022;5(2):89-97.
3. Al-Sayed A. Triage Accuracy: The Role of Technology in High-Volume Settings. Middle East J Emerg Med. 2021;18(3):210-218.
4. World Health Organization. Accelerating access to digital health innovations. Geneva: WHO; 2020.
5. Johnson D, Smith R. Machine learning models for predicting patient severity in emergency care. IEEE J Biomed Health Inform. 2024;28(2):401-410.