Team Shiba

ECMO is an advanced method of life support for pediatric patients with cardiac and/or pulmonary failure. In a recent multicenter study of pediatric ECMO, the cumulative proportions of hemorrhagic and thrombotic events were 70% and 37.5%, respectively. Early prediction of these complications is important for patient recovery. Using multivariable logistic regression models and random forest classification, our model attempts to determine the most relevant demographic, physiologic, laboratory, and transfusion-related risk factors for hemorrhagic and thrombotic complications in ECMO patients, using a time-series data set. In this study, we analyze electronic health record data from 169 pediatric patients (age 0-18 years) admitted to a single center between 2011–2018. This study demonstrates that comprehensive health data can be used to establish an effective prediction model for ECMO-related complications.