Development and validation of a model for endotracheal intubation and mechanical ventilation prediction in PICU patients

Predicting the need for intubation in pediatric intensive care units (PICUs) is critical for timely intervention and resource allocation. This study aimed to develop and externally validate a machine learning model to predict intubation in children admitted to a PICU using routinely available electronic medical record (EMR) data. A retrospective observational cohort study was conducted across two PICUs within the same healthcare system, including a quaternary academic center and a tertiary community hospital. Clinical data were extracted from the EMR, focusing on PICU stays where mechanical ventilation occurred for at least 24 hours within 1–7 days of hospital admission. In the derivation cohort (n = 13,208), 8.90% of stays included an intubation event, while in the validation cohort (n = 17,841), 6.53% of stays required intubation. A Categorical Boosting (CatBoost) model was trained using vital signs, laboratory results, demographic data, medications, and organ dysfunction scores to predict intubation within a 24-hour observation window. The model outperformed extreme gradient boosting, random forest, and logistic regression models, achieving an area under the receiver operating characteristic curve (AUC) of 0.88 (95% CI, 0.88–0.89) in the derivation cohort and 0.92 (95% CI, 0.91–0.92) in the validation cohort. The study demonstrates that an interpretable machine learning model can effectively predict the need for intubation in PICU patients, offering a valuable tool to enhance clinical decision-making. Implementation of this model may improve the timing of intubation and optimize resource allocation for mechanically ventilated children.