Machine Learning and Artificial Intelligence Approaches for Predicting Arteriovenous Fistula Dysfunction in...

Area of Science:

• Nephrology
• Biomedical Engineering
• Data Science

Background:

• Arteriovenous fistula (AVF) is crucial for hemodialysis, but dysfunction is a common issue impacting patient outcomes.
• Traditional risk prediction models struggle with complex interactions, highlighting the need for advanced methods.

Purpose of the Study:

• To systematically review and synthesize evidence on artificial intelligence (AI) and machine learning (ML) for predicting AVF dysfunction.
• To evaluate the performance of AI/ML models compared to conventional methods.

Main Methods:

• Systematic literature search in major databases (PubMed, Embase, Scopus, Web of Science) for studies from January 2021 to December 2025.
• Inclusion of studies applying AI/ML to predict AVF dysfunction in adult hemodialysis patients.
• Risk of bias assessment using PROBAST and narrative synthesis due to heterogeneity.

Main Results:

• Ten studies met inclusion criteria, primarily retrospective cohorts from China and the USA.
• Ensemble tree-based models (Random Forest, XGBoost, LightGBM) achieved high predictive accuracy (AUC-ROC 0.80-0.98), outperforming traditional models.
• Key predictors included prior surgeries, inflammatory markers, and imaging parameters.

Conclusions:

• AI and ML, especially ensemble tree methods, show strong potential for predicting AVF dysfunction.
• Further research requires external validation, prospective studies, and standardized reporting for clinical implementation.