Abstract 13443: Multi-Center Independent Validation of an Automated Algorithm for Predicting Cardiorespiratory Deterioration Events in Single Ventricle Patients

Abstract

Introduction: Sudden acute deterioration is common in patients with hypoplastic left heart syndrome and comparable lesions. Previous work by the investigators using a single-center cohort at Texas Children’s Hospital (TCH) resulted in a novel algorithm which can predict such events 1-2 hours in advance. Hypothesis: The TCH algorithm for predicting deterioration events in single ventricle patients 1-2 hours in advance can be translated to heart centers across the country without significant performance loss. Methods: A prospective study on infants undergoing cardiovascular surgical palliation at Children’s Colorado, Nationwide Children’s Hospital, and Children’s Hospital of Philadelphia between Oct 2018 to Apr 2020 was conducted. Subjects were admitted to the cardiovascular ICU (CVICU) immediately after stage-1 palliation and enrollment continued until stage-2 palliation was performed or the subject was discharged. All physiologic data from all subjects were recorded from bedside monitoring equipment using the Sickbay system (Medical Informatics Corp, Houston, TX) in both CVICU and step-down units. Instances of cardiorespiratory deterioration, defined as the need for CPR or endotracheal intubation, were identified. The existing predictive algorithm was applied without re-training to the data obtained from study sites. A 12-hour lead-in period was used to initialize the algorithm. The ROC area, true and false positive rates (TRP & FPR) were used to quantify performance. Results: There were 58 subjects who met inclusion criteria across all sites. Median time between stage I & II surgeries was 141 days (126-155 IQR). A total of 30 in-hospital deterioration events were observed. Twenty-one subjects experienced at least 1 event. Most (72%) events occurred 40 days after stage I surgery. Over 29K hours of data was collected from subjects. The ROC area was found to be 0.927 for predicting deterioration events 1-2 hours in advance of overt extremis. Conclusions: This study validated the performance of an existing predictive algorithm which provides a 1-2 hour early warning for cardiorespiratory deterioration events in children with single ventricle physiology during their interstage period using a large, independent, multi-center patient cohort.