Use of Predictive Analytics to Quantify Neonatal Hypothermia Burden After Cardiac Surgery

点击翻译按钮获取中文摘要

ABSTRACT Neonates (infants ≤ 28 days), especially those with congenital heart disease (CHD), are among the most vulnerable populations cared for by critical care nurses. Approximately, two out of three CHD neonates experience unintentional hypothermia after cardiopulmonary bypass (CPB). Unintentional hypothermia impairs cellular function, which can be linked to poor outcomes frequently reported in this population. To date, there are no studies examining the association between the burden of unintentional hypothermia and clinical outcomes in neonates with CHD. This knowledge would render future opportunities to improve nursing care and prevent avoidable safety events in these vulnerable neonates. To address this gap, we propose to use retrospective data from CardioAccess (database local to the Children’s Hospital of Philadelphia [CHOP]), which includes one of the largest multicenter repositories of neonatal cardiac surgery data available to date (Pediatric Cardiac Critical Care Consortium [PC4]), as well as, the electronic health record. Using data from at least 432 neonates who have undergone CPB between 2015 and 2019, we will quantify the time course of hourly temperature trajectories within the initial 24–48 hours after CPB and evaluate their relation to key clinical outcomes. We will specifically study the temporal trends of unintentional hypothermia burden (temperature depth and duration), which challenges current practice, where care is based on maintaining a single, preselected temperature threshold that is driven by consensus, rather than evidence. Single threshold values are not dynamic representations of the complexity that makes up temperature. A more robust output, such as an accumulative hypothermia burden index, is needed to assist clinicians with interpretation of this dynamic indicator of overall health. Our Specific Aims are: 1) Identify distinct temporal temperature patterns in CHD neonates after CPB using both: a multilevel model for intensive longitudinal data with group-based trajectory modeling; and an unsupervised machine learning technique using principal component analysis followed by k- means clustering of longitudinal data. 2) Determine the relationship between hypothermia burden subgroups / clusters and important clinical outcomes in this population. Our team has a demonstrated expertise in building clinically relevant and physiologically plausible markers of adverse outcomes in critically ill patients. This study aligns with the NINR’s priorities of promoting wellness and preventing illness across the lifespan, as well as, using recent advances in precision medicine. The research conducted under this award will take place at the University of Pittsburgh School of Nursing, a research-intensive institution (data analysis), and CHOP (data provision). The personalized training plan outlined in this application, supports the applicant’s career and academic development goals to become an independent nurse researcher.