A machine learning model to predict intradialytic hypotension in pediatric continuous kidney replacement therapy

Area of Science:

• Pediatric critical care medicine
• Nephrology
• Biomedical engineering

Background:

• Intradialytic hypotension (IDH) is a known mortality risk in adult hemodialysis patients.
• The association between IDH and mortality in critically ill children on continuous kidney replacement therapy (CKRT) remains unclear.
• Early identification of IDH is crucial for improving outcomes in pediatric CKRT.

Purpose of the Study:

• To investigate the relationship between IDH and hospital mortality in pediatric CKRT patients.
• To determine if pressure data from dialysis machines can predict IDH.
• To explore machine learning models for IDH prediction.

Main Methods:

• Retrospective cohort study of 23 pediatric patients (38 admissions) in a pediatric intensive care unit and NICU.
• Analysis of intradialytic hypotension (IDH) proportion and its association with mortality.
• Development and evaluation of machine learning models (Random Forest, CatBoost) using dialysis pressure data and clinical variables.

Main Results:

• IDH proportion was significantly associated with increased hospital mortality (RR: 4.40, p=0.02).
• Machine learning models, particularly CatBoost, demonstrated high accuracy in predicting IDH (AUC up to 88.18%).
• Dialysis machine-derived pressure parameters were identified as key predictors of IDH, outperforming traditional blood pressure measures.

Conclusions:

• Intradialytic hypotension (IDH) is a significant predictor of mortality in pediatric CKRT patients.
• Dialysis machine pressure data offers a promising avenue for early IDH detection.
• Integrating pressure monitoring into CKRT management systems can enhance patient safety and outcomes.