Development of a machine learning model for predicting pediatric mortality in the early stages of intensive care unit...

Area of Science:

• Pediatric critical care medicine
• Machine learning in healthcare
• Predictive analytics for patient outcomes

Background:

• Pediatric mortality prediction is crucial for timely intervention.
• Existing models may lack accuracy in early intensive care unit (ICU) stages.
• Machine learning offers potential for improved predictive capabilities.

Purpose of the Study:

• To develop and validate a machine learning model for predicting pediatric mortality within 72 hours of ICU admission.
• To compare the performance of the developed model against the Pediatric Index of Mortality 3 (PIM 3).

Main Methods:

• Development of a random forest (RF) model using data from three tertiary referral hospitals.
• Internal validation of the RF model on the derivation cohort.
• External validation of the RF model on a separate hospital cohort.
• Comparison of RF model performance (AUROC) with PIM 3.

Main Results:

• The RF model achieved an AUROC of 0.942 in the derivation cohort and 0.906 in the external validation cohort.
• The PIM 3 model achieved an AUROC of 0.892 in the derivation cohort and 0.845 in the external validation cohort.
• The RF model demonstrated superior predictive performance compared to PIM 3 in both internal and external validation.

Conclusions:

• The developed random forest model shows high accuracy in predicting early pediatric mortality in the ICU.
• This machine learning approach surpasses the predictive performance of the established PIM 3 score.
• The findings suggest a valuable tool for enhancing clinical decision-making and patient management in pediatric critical care.