A pilot study to predict cardiac arrest in the pediatric intensive care unit

Area of Science:

• Pediatric critical care medicine
• Biomedical informatics
• Machine learning in healthcare

Background:

• In-hospital cardiac arrest (IHCA) is a significant cause of mortality in pediatric intensive care units (PICUs).
• Predicting IHCA in critically ill children remains a challenge for clinicians.
• Early prediction is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and evaluate machine learning models for the early prediction of IHCA in pediatric intensive care settings.
• To identify key clinical and physiological features indicative of impending cardiac arrest.
• To assess the performance of various machine learning algorithms in predicting IHCA up to three hours in advance.

Main Methods:

• Utilized a dataset of 1,145 pediatric ICU patients including ECG, physiological time series, medications, and demographics.
• Extracted 23 heart rate variability (HRV) metrics from ECG and 96 summary statistics from 12 vital signs.
• Classified medications into 42 therapeutic drug classes and evaluated six machine learning models, including XGBoost.

Main Results:

• The XGBoost model demonstrated superior performance on an independent test set.
• Achieved an area under the receiver operating characteristic curve (auROC) of 0.971 and an area under the precision-recall curve (auPRC) of 0.797.
• The model achieved 99.5% sensitivity and 69.6% specificity for predicting IHCA.

Conclusions:

• High-performing machine learning models can identify subtle signatures of IHCA.
• These models integrate HRV, vital signs, and medication data for prediction.
• Early IHCA prediction up to three hours in advance is feasible, enabling earlier clinical intervention and potentially improving patient outcomes.