Abstract
Background: Preterm neonates with respiratory distress syndrome
(RDS) who fail continuous positive airway pressure (CPAP) ventilation
have higher risks for increased morbidity and mortality.
Objective: To assess if machine learning, on multicenter data,
may predict CPAP failure in preterm infants with RDS and allow proactive
intervention to minimize CPAP failure burden and improve clinical
outcomes.
Methods: This study was conducted using the Oracle EHR
Real-World Data (OERWD) database including preterm NICU admits between
2002-2023. CPAP failure was defined as the need for invasive mechanical
ventilation within 72 hours of life. Demographics, admit vital signs,
and laboratory values were retrieved to develop an explainable machine
learning model using extreme gradient boosting (XGBoost).
Results: 24,127 neonates from 27 NICUs qualified for the study
with CPAP failure rate of 64.1%. FiO2 was the strongest predictor of
CPAP failure followed by systolic blood pressure, temperature,
birthweight, PaO2, oxygen saturation, heart rate, and gestational age
followed in importance. Resulting XGBoost model attained an area under
the receiver operator characteristic curve of 0.91 (95% CI: 0.90, 0.92)
and an F-1 score of 0.87.
Conclusions: CPAP failure can be predicted with high accuracy
at admission to the NICU creating opportunities for early intervention
and prevention of RDS related complications.