Abstract
Introduction: In the emergency room, it is essential to quickly and accurately classify the patients' various severities. However, existing five-stage classification systems, such as the Korean Emergency Patient Classification Tool (KTAS), do not sufficiently reflect the physiological and clinical heterogeneity of all patients, so there is a possibility of under-classification in some age groups or specific symptom groups. Methods: A retrospective cross-sectional study was conducted using KTAS and the physiological and clinical data of 41,728 patients who visited the emergency room of a university hospital in Incheon in 2022. K-prototypes unsupervised cluster analysis incorporating demographic, physiological, and clinical variables was applied, and the number of clusters was determined as the optimal value through the Silhouette, Dunn, and Davies-Bouldin indicators. Dimension reduction was performed by UMAP, and differences between clusters were compared by t-test, Mann-Whitney U, and chi-square test. Results: Two different clusters were identified. Cluster 0 was a stable patient group with a mean age of 58 years and an average arterial pressure of 104 mmHg. On the other hand, Cluster 1 was a young but physiologically unstable patient group with an average age of 46 years and an average arterial pressure of 90 mmHg. There were significant differences in age, MAP, heart rate, respiratory rate, body temperature, and pain scores between clusters (p < 0.001), and a moderate association was observed between KTAS classification and clusters (Cramer's V = 0.208). Discussion: This study suggested the possibility of early identification of high-risk groups in the emergency room and efficient resource allocation by identifying potential patient heterogeneity that KTAS cannot detect through unsupervised learning. This approach can be used as a basis for precision triage and patient-centered emergency medical policy establishment by supplementing rather than replacing the existing classification system.