Developing a Machine Learning Model and Unified Protocol Approach to Predict Neonatal Extubation Readiness

About This Grant

Significance: Premature infants can suffer significant morbidity and mortality from both prolonged mechanical ventilation (MV) and extubation failure, however, currently no prediction models or standard tests of extubation readiness exist to aid with the high-risk decision on appropriate timing of extubation in neonates in the Neonatal Intensive Care Unit (NICU). Bronchopulmonary dysplasia, a form of chronic lung disease, is the most common morbidity of prematurity that may affect intubated NICU patients for the rest of their lives. Thus, there is a critical need to develop, evaluate, and implement tools to determine extubation readiness in the NICU to mitigate infant mortality and morbidity, especially chronic lung disease. Approach: The overall objective of this proposal is to design a clinical decision support (CDS) tool that will predict extubation readiness for intubated preterm infants and can be integrated into the clinical care workflow. The central hypothesis is that including physiologic information provided by continuous vital sign and ventilator data along with clinical data into a machine learning (ML) prediction model will enhance personalized decision making and clinical adoption. This study will enhance preliminary work with an ML-based prediction model by including data from a second site to (Aim 1) develop and validate a prediction model for extubation readiness using data from two clinically distinct centers. Model performance will be compared to those of published logistic regression algorithms. To begin the future work to implement a protocol in a larger multicenter trial, this study will use the consolidated framework for implementation research to (Aim 2) develop a unified protocol via surveys and interviews for evaluating neonatal extubation readiness and determine effective informatics-based implementation strategies for adopting these guidelines into the clinical workflow. Because approaches to neonatal respiratory care vary internationally, inclusion of neonatal clinicians from both the United States and Canada in Aim 2 will provide broader perspectives on barriers, facilitators, and implementation strategies for extubation readiness protocols. These insights will improve the generalizability and future adoption of standardized protocols within U.S. NICUs. Innovation: This study is integrating continuous physiologic data that is not routinely incorporated into prediction models and performing pre-implementation methodology to learn how to implement an extubation readiness protocol. The expected outcomes of this project are: 1) a clinically useful ML model for predicting extubation readiness in the NICU, and 2) a targeted informatics-based implementation strategy for an extubation readiness protocol. The broader impact is a reduction in extubation failure rates and associated morbidities, including MV duration and chronic lung disease.