Clustering Emergency Department patients - an assessment of group normality

This paper presents an investigation into clustering of vital signs from Emergency Department patients with an intention of uncovering distinct thresholds for groups of patients. Emergency Department clinicians have to deal with an enormous spectrum of symptoms and diseases. The variety in patients is a cause for false alarms which greatly burden clinicians. Better targeted alarm thresholds may mitigate the risk of alarm fatigue. The study is based on vital signs from a prospective cohort study at a Danish Hospital coupled with health registry data, and utilizes k-means clustering and novel evaluation metrics for cluster assessment. All combinations of 5 key vital signs are clustered in a range from 2..20. We evaluate the clustering of respiration and arterial peripheral oxygen saturation for k=7. The study fails to identify distinct groups, but does uncover relevant traits and contribute with an evaluation strategy for further studies.