Researchers in South Korea are developing an AI-powered smartphone application designed to help emergency department staff determine whether children require hospital admission, with the goal of improving triage during periods of overcrowding and physician shortages. The project is being led by a collaboration involving the Catholic University of Korea, Korea University, Asan Medical Center and medical AI company VUNO. The application is based on a natural language processing model that analyzes clinical notes recorded during the earliest stages of a patient's emergency department visit, before laboratory or imaging results become available.
The AI model was trained using electronic medical record data from nearly 88,000 pediatric emergency visits collected over almost a decade. Instead of relying solely on traditional triage scores, the researchers classified patients based on the care they ultimately received, distinguishing between cases requiring significant medical intervention and those discharged after minimal treatment. Built on a Korean adaptation of the BERT language model and further trained on medical records, the system analyzes symptom descriptions and clinician notes to identify subtle patterns that may indicate a child requires more intensive care. In testing, the model outperformed several conventional machine learning approaches as well as the Korean Triage and Acuity Scale in predicting which patients would require emergency-level treatment.
The researchers believe free-text clinical notes contain valuable early information that is often overlooked by traditional triage systems based primarily on structured data such as vital signs. They are now translating the model into a smartphone application that will provide clinicians with real-time decision support during pediatric emergency assessments. Additional multicenter validation studies are planned to evaluate performance across different hospitals and patient populations, including post-pandemic datasets. More broadly, the project reflects growing interest in applying AI and natural language processing to emergency medicine, where earlier identification of high-risk patients could improve resource allocation, reduce overcrowding and enhance patient safety.
Click here to read the original news story.
