Cardiorespiratory instability in monitored step-down unit patients: using cluster analysis to identify patterns of change

Cardiorespiratory instability (CRI) in monitored step-down unit (SDU) patients has a variety of etiologies, and likely manifests in patterns of vital signs (VS) changes. We explored use of clustering techniques to identify patterns in the initial CRI epoch (CRI 1 ; first exceedances of VS beyond stability thresholds after SDU admission) of unstable patients, and inter-cluster differences in admission characteristics and outcomes. Continuous noninvasive monitoring of heart rate (HR), respiratory rate (RR), and pulse oximetry (SpO 2 ) were sampled at 1/20 Hz. We identified CRI 1 in 165 patients, employed hierarchical and k -means clustering, tested several clustering solutions, used 10-fold cross validation to establish the best solution and assessed inter-cluster differences in admission characteristics and outcomes. Three clusters (C) were derived: C1) normal/high HR and RR, normal SpO 2 (n = 30); C2) normal HR and RR, low SpO 2 (n = 103); and C3) low/normal HR, low RR and normal SpO 2 (n = 32). Clusters were significantly different based on age (p