Prediction of Adverse Events in Single Ventricle Physiology Infants Using Artificial Intelligence Tools

Area of Science:

• Pediatric Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Adverse events (AEs) in cardiac intensive care units (CICUs) carry high mortality risks.
• Infants with single ventricle (SV) physiology are especially vulnerable to AEs before the bidirectional Glenn procedure.
• Early detection and management of AEs are crucial for improving outcomes in this population.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting AEs in infants with SV physiology.
• To identify AEs, including cardiac arrest (CA), extracorporeal membrane oxygenation (ECMO) cannulation, and endotracheal intubation, up to 8 hours prior to occurrence.
• To utilize continuous physiologic data for predictive modeling in a vulnerable pediatric population.

Main Methods:

• A retrospective cohort of 158 SV infants (324 admissions) was analyzed.
• Supervised ML classifiers, including Random Forest (RF), were trained on physiologic data across 1-, 2-, 4-, and 8-hour windows preceding AEs.
• Model performance was assessed using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• The RF model demonstrated high performance in predicting AEs, with AUROCs ranging from 0.996 to 0.998 across all time windows.
• For multiclass classification at a 1-hour window, the RF model achieved AUROCs of 0.819 for intubation, 0.804 for ECMO-CA, and 0.840 for no-event prediction.
• Six high-quality variables were identified and included in the predictive ML models.

Conclusions:

• ML models can effectively predict and differentiate various AEs in SV infants prior to the bidirectional Glenn surgery.
• Accurate AE prediction facilitates timely interventions, potentially decreasing morbidity, mortality, and healthcare costs.
• Continuous physiologic data combined with ML offers a promising approach for proactive patient management in pediatric cardiac care.