Feasibility of Automated Vital Sign Instability Detection in Children Admitted to the Pediatric Intensive Care Unit

Abstract

Children admitted to a Pediatric Intensive Care Unit (PICU) are at risk of deterioration, which can lead to a cardiac arrest if undetected. Outcomes after pediatric cardiac arrest remain poor, even for witnessed, in-hospital events. Thus, early detection of deterioration is paramount; ideally long before the risk of harm increases. Vital signs trends of patients admitted to the PICU at BC Children’s Hospital were extracted from local outcomes registries (n=96). A rule-based algorithm (RBA) for detecting vital signs instabilities was developed; we did so in the expectation of enhancing clinician trust compared to black box approaches such as deep neural networks. Two PICU physicians provided expert classifications for episodes indicative of vital signs instability or their absence. The RBA’s best result generated 91.6% correct, 6% false negatives, and 3% false positives on the test data (n=29) showing promise for eventual application in a clinical setting. Future research is needed to refine the algorithm and implement it clinical practice.