Early Prediction of Mechanical Ventilation Needs in Very Preterm Neonates Using Machine Learning

Area of Science:

• Neonatal Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Invasive mechanical ventilation (IMV) is critical for some neonates but increases morbidity.
• Delayed or unnecessary IMV initiation negatively impacts preterm neonates.
• Continuous positive airway pressure (CPAP) failure necessitates escalation of respiratory support.

Purpose of the Study:

• To develop a machine learning model for predicting IMV need in neonates experiencing CPAP failure.
• Utilize early clinical data from electronic health records for prediction.
• Enable proactive interventions to improve outcomes for preterm neonates.

Main Methods:

• Utilized the Oracle EHR Real-World Data (OERWD) database (2012-2022).
• Included preterm neonatal intensive care unit (NICU) admissions.
• Trained an extreme gradient boosting (XGBoost) model using demographics, vital signs, and laboratory values.

Main Results:

• The study included 20,363 neonates from 27 NICUs; 69.0% experienced CPAP failure.
• Key predictors for CPAP failure included FiO2, systolic blood pressure, PaO2, birthweight, diastolic blood pressure, gestational age, and oxygen saturation.
• The XGBoost model achieved an AUC of 0.90 and an F1 score of 0.88.

Conclusions:

• Accurate early prediction of IMV requirement is feasible at birth.
• Early prediction facilitates timely IMV initiation for indicated patients.
• This approach can help minimize unnecessary intubations and improve patient care.