Machine learning-based prediction of circuit clotting during pediatric continuous kidney replacement therapy sessions

Area of Science:

• Pediatric Critical Care Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Continuous kidney replacement therapy (CKRT) is vital for critically ill children with acute kidney injury (AKI).
• Circuit clotting is a common complication, leading to treatment interruptions and adverse events.
• Predicting clotting events could significantly improve patient management and outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting circuit clotting in pediatric CKRT.
• To identify key variables associated with premature circuit clotting.
• To assess the potential of ML for enhancing CKRT management in pediatric intensive care units.

Main Methods:

• Retrospective analysis of de-identified CKRT session data from 23 pediatric patients.
• Development of an ML classification model to predict clotting 60 minutes in advance.
• Utilized time-series data, feature selection (LightGBM), and cross-validation (Extra Trees classifier).

Main Results:

• The ML model achieved an AUROC of 0.99 on the training set.
• Successfully predicted clotting events 60 minutes prior in 148 instances during validation.
• Key predictors included effluent volume, treatment duration, fluid removal, and dialysate flow.

Conclusions:

• Machine learning models can effectively predict circuit clotting during pediatric CKRT.
• Early prediction facilitates timely clinical intervention, reducing complications.
• Integrating predictive algorithms into clinical workflows can optimize treatment and improve outcomes for critically ill children.