Predicting Flow Rate Escalation for Pediatric Patients on High Flow Nasal Cannula Using Machine Learning

Area of Science:

• Pediatric critical care medicine
• Respiratory physiology
• Machine learning in healthcare

Background:

• High flow nasal cannula (HFNC) is a common non-invasive respiratory support for critically ill children.
• Optimal HFNC parameters and escalation criteria remain unclear, lacking predictive models for flow rate adjustments.
• Existing clinical scores like the ROX index are limited in predicting HFNC non-response beyond invasive mechanical ventilation escalation.

Purpose of the Study:

• To evaluate tree-based machine learning algorithms for predicting HFNC flow rate escalations in pediatric acute respiratory failure.
• To compare the predictive performance of machine learning models against the ROX index.

Main Methods:

• Retrospective cohort study of children (<24 months) with acute respiratory failure on HFNC.
• Machine learning models, including gradient boosting, were trained to predict HFNC flow rate escalation.
• Area Under the Receiver Operating Characteristic (AUROC) analysis was used to compare model performance against the ROX index.

Main Results:

• The gradient boosting model achieved an AUROC of 0.810 ± 0.003 for predicting HFNC flow rate escalation.
• The ROX index demonstrated a significantly lower AUROC of 0.525 ± 0.000.
• Tree-based machine learning models significantly outperformed the ROX index in predicting HFNC flow rate escalations.

Conclusions:

• Tree-based machine learning models show superior ability to predict HFNC flow rate escalations compared to the ROX index in pediatric patients.
• These findings suggest potential for machine learning to guide respiratory support adjustments in critically ill children.
• Further validation is required for widespread clinical application of these predictive models.