Machine Learning-Based Prediction Model for ICU Mortality After Continuous Renal Replacement Therapy Initiation in...

Area of Science:

• Pediatric critical care medicine
• Nephrology
• Biomedical data science

Background:

• Continuous renal replacement therapy (CRRT) is vital for critically ill children.
• Predicting outcomes in CRRT patients is challenging due to heterogeneity and limited data.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting survival in pediatric and young adult CRRT patients.
• To identify key predictors of intensive care unit (ICU) and hospital survival.

Main Methods:

• A retrospective international multicenter study used an 80/20 training/testing split of 933 patients under 25.
• Models included logistic regression (LR), decision tree, random forest (RF), gradient boosting, and support vector machine.
• Performance was assessed using AUROC and AUPRC due to imbalanced data.

Main Results:

• Random forest (RF) showed the highest performance for predicting ICU mortality (AUROC 0.791).
• Logistic regression (LR) performed best for hospital mortality (AUROC 0.777).
• Key predictors for ICU survival included Pediatric Logistic Organ Dysfunction-2 score and respiratory failure diagnosis.

Conclusions:

• This study presents the first ML models for predicting CRRT survival in pediatric and young adult populations.
• RF models demonstrated superior performance for ICU mortality prediction.
• Future research should incorporate more variables and explore deep learning for enhanced precision.