Mechanical Ventilation-Driven Machine Learning for Predicting the Risk of Pediatric Respiratory Failure

Acute respiratory conditions are the most common reason children are admitted to the hospital, and respiratory failure is the most costly diagnosis in acute care pediatrics. Identifying which children are at risk of decline is crucial to clinicians and health systems. In this study, we developed machine learning models to predict 12th-hour Oxygenation Saturation Index (OSI) using 1st-hour ventilator-derived data, combining clinician domain knowledge with advanced signal processing. The Random Forest classifier achieved the best performance, with an AUROC of 0.87 and an AUPRC of 0.52, significantly outperforming the baseline value of 0.167. Lasso and Ridge emerged as the leading regression models, yielding RMSE and MAE values of 2.41 and 1.70, respectively. These findings demonstrate the feasibility of integrating ventilator data with computational tools to improve risk stratification and optimize pediatric intensive care unit resource allocation.Clinical relevance- This study addresses the gap in the limited application of invasive ventilation data combined with clinician-informed features for predicting hypoxia risk in mechanically ventilated children. The approach underscores the potential to enhance early risk stratification and guide critical care decisions, ultimately improving patient outcomes and optimizing resource utilization in pediatric intensive care.