Neonatology-Perinatology Fellow Nationwide Children's Hospital Columbus, Ohio, United States
Background: Bronchopulmonary dysplasia (BPD) is a chronic lung disease due to the disruption of pulmonary development and injury in premature neonates, associated with mechanical ventilation. Although invasive mechanical ventilation allows for improved oxygenation and ventilation, prolonged use is associated with an increased risk of BPD and significant comorbidities. Conversely, premature extubation resulting in reintubation is also associated with significant morbidity, mortality, and longer hospitalization. Patients with BPD are discharged on a wide variety of respiratory support ranging from nasal cannula to tracheostomy, depending on if they can be successfully extubated. There are no current guidelines to predict if an extubation attempt in a patient with established BPD is going to be successful, and no prior studies to guide clinical judgement. Objective: Our objective is to develop a predictive machine learning model to determine if an extubation attempt will be successful or not in patients with BPD. We plan to validate this model prospectively with the eventual goal of integration into the electronic health record (EHR) clinical decision-making workflow. Design/Methods: This study was approved by the IRB at Nationwide Children’s Hospital. It is a retrospective study taking place at a single center neonatal intensive care unit and uses data from the Children’s Hospital Neonatal Consortium database, and data pulled from the EHR. The model includes both modifiable and non-modifiable variables from pregnancy and throughout the patient’s hospitalization. We used LASSO regularized logistic regression and random forest to build our predictive models. We finished training and validating the model on retrospective data in September 2023. We are currently building a prospective platform to collect data and validate the model prospectively. This will be completed by mid-March 2024. By late March 2024 we plan to perform data analysis. If results look promising, we will begin integration into the EHR.
