Prediction of CPAP Failure in the Neonatal Icu

Abstract

Continuous Positive Airway Pressure (CPAP) is now the preferred treatment for neonatal patients suffering from less severe respiratory distress upon admission to the Neonatal Intensive Care Unit (NICU). If a patient “fails” CPAP, meaning the patient does not respond to CPAP therapy and needs to be transitioned to invasive mechanical ventilation, this could cause later complications for the infant’s development. This study suggests predictors for CPAP failure based on clinical data from the MIMIC II (Multi-parameter Intelligent Monitoring for Intensive Care, Version 2) database. A reference data set was created by selecting neonatal patients who were <36 weeks gestational age, born with Respiratory Distress Syndrome, placed on CPAP within 24 hours of NICU admission, and on CPAP for >4 hours. From the 22 features considered as possible predictors of CPAP failure, the final features were chosen based on a minimum number of patients who had the feature recorded, a statistically significant difference of that feature between the pass/fail groups of patients using F-score, and consultation with a neonatologist. Logistic regression with 100X bootstrapping was used to assess combinations of one, two and three features with and without interaction terms. The set of three features that showed both the highest AUC (Area Under ROC Curve) and highest sensitivity at a specificity of 90% were Oxygen Saturation (SaO2), Total Cycle Time, and Indirect Bilirubin. This set of three features resulted in an AUC of 96.5%, Sensitivity of 88.2%, and Specificity of 90%.