Early Prediction of Multiple Organ Dysfunction in the Pediatric Intensive Care Unit

Area of Science:

• Pediatric critical care medicine
• Machine learning in healthcare
• Predictive analytics

Background:

• Multiple organ dysfunction (MOD) is a significant cause of mortality in pediatric intensive care units (PICUs).
• Early identification of patients at risk for MOD is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop and evaluate machine learning models for the early prediction of MOD risk in PICU patients.
• To assess the lead time and accuracy of these predictive models.

Main Methods:

• Retrospective observational cohort study of pediatric patients (<18 years) admitted to a single academic PICU.
• Organ dysfunction was labeled using International Pediatric Sepsis Consensus Conference (IPSCC) and Proulx et al. criteria.
• Four machine learning models (random forest, XGBoost, GLMBoost, Lasso-GLM) were trained to predict MOD using 24-hour rolling time windows.

Main Results:

• All models achieved an area under the receiver operating characteristics curve (AUC) ≥0.91.
• XGBoost model showed the highest performance (F1-score: 0.81) for predicting MOD based on Proulx criteria, with a median early warning time of 37 hours.
• Random forest model achieved a sensitivity of 0.72 and positive predictive value of 0.70 for IPSCC criteria, with a median warning time of 22.7 hours.
• Spectral clustering identified a high-risk group with a positive predictive value (PPV) of ≥0.93 for MOD onset.

Conclusions:

• Risk-based patient monitoring using machine learning can provide early warnings for MOD in PICU patients.
• The developed models offer a significant lead time (over 22 hours) for MOD onset.
• High-risk patient identification with high PPV (≥0.93) enables proactive clinical management.