Emergency Department Admissions Overflow Modeling by a Clustering of Time Evolving Clinical Diagnoses

摘要

Emergency Departments (ED) of hospitals are greatly impacted by winter epidemics of respiratory diseases. To detect the underlying overcrowding, it is essential to study patient flow. In this paper we propose to model the admission flow corresponding to clinical diagnoses encoded with ICD-10 which are more likely linked with respiratory diseases. To achieve this, clustering algorithms are applied on time evolving diagnoses in the adult ED of Saint-Etienne and benchmarked regarding a time series of laboratory-confirmed influenza data. For both K-Means and Hierarchical algorithms, the cluster containing the laboratory-confirmed series is composed of ICD-10 codes of diagnoses representing respiratory diseases and diseases linked with cardiac disorders, showing that these diseases present similar variations overtime. The information contained in such a cluster makes it possible to plot the average number of arrivals of these diagnoses overtime and the average length of stay of the patients in the ED who only have one or several of these diagnoses. Such an acknowlegdement about its patient flow will allow an ED staff to detect the underlying overcrowding.