Analyzing Temporal Patterns of ICU Alarms: A Time- Series Clustering Approach

Abstract

Objective This study employs a machine learning approach, namely time-series clustering with Dynamic Time Warping (DTW), to empirically identify subgroups of intensive care unit (ICU) patients and examine the relationships between different alarm types. Utilizing unsupervised learning algorithm, the study aims to uncover insights into temporal alarm patterns in large datasets without the need for labeled data. The primary objectives are to identify distinct patterns of patient complications within the initial hours of ICU stay, regardless of admission diagnosis, and to understand how different types of alarms interact over time.

Design The time-series clustering was performed using data from Wilhelmina Kinderziekenhuis (WKZ) hospital. The patient population included individuals from the pediatric and neonatology departments.