Cardiac arrest in a devastating complication in critically ill children, associated with substantial mortality and lifelong morbidity. Identifiable changes in the behavior of physiological variables preceding cardiac arrest has generated interest in assessing whether or not physiological signal analysis using techniques like machine learning can be used to accurately predict this event in critically ill children. In collaboration with the Goldenberg lab at the Hospital for Sick Children, we are utilizing convolutional and recurrent neural networks and patient physiological data to generate cardiac arrest prediction models for deployment at the point of care. The attached manuscript describes a proof of concept approach using 5 second data archived from Etiometry’ Tracking, Trajectory and Triggering (T3) system.
