A Machine Learning Pipeline for Mortality Prediction in the ICU

Mortality risk prediction for patients admitted into the intensive care unit (ICU) is a crucial and challenging task, so that clinicians are able to respond with timely appropriate clinical intervention. This becomes more urgent under background of COVID-19 as global pandemic. In recent years, electronic health records (EHR) have been widely adopted, potential greatly improve services diagnostics. However, large proportion missing data in EHR poses challenges may reduce accuracy methods. We propose cohort study builds pipeline extracts ICD-9 codes laboratory tests from public available ICU databases, in-hospital mortality using combination neural network imputation approach decision tree based outcome algorithm. show proposed achieves higher area ROC curve, ranging 0.88-0.98, compared other well-known machine learning methods applied similar target population. It also offers interpretations through variable selection. Our analysis shows neonates was than adults, decreases stayed longer ICU.