Development of a Risk Prediction Model for Arteriovenous Fistula Restenosis Following Percutaneous Transluminal Angioplasty in Maintenance Hemodialysis Patients Using SMOTE Algorithm

Abstract

Background

Arteriovenous fistula (AVF) restenosis after percutaneous transluminal angioplasty (PTA) affects hemodialysis in maintenance hemodialysis (MHD) patients. Data imbalance limits predictive model accuracy.

Objective

To identify risk factors and develop a SMOTE-based risk prediction model for AVF restenosis within 12 months after PTA.

Methods

This retrospective cohort study enrolled 241 consecutive MHD patients with AVF stenosis undergoing PTA between July 2022 and July 2025. Each patient contributed one index PTA. Patients were divided into restenosis (45 cases) and nonrestenosis (196 cases) groups according to whether AVF restenosis occurred within 12 months after PTA. Univariate and multivariate logistic regression identified risk factors. The SMOTE algorithm addressed data imbalance. Model performance was evaluated by ROC curves, calibration curves, and decision curve analysis (DCA).

Results

Original-data multivariate analysis identified multisegmental stenosis, cannulation method, platelet count, and C-reactive protein (CRP) as independent risk factors. After SMOTE, intradialytic hypotension and serum albumin also became significant; however, these variables should be interpreted as exploratory because SMOTE does not create additional independent clinical information. The SMOTE-based model (P2) achieved AUC of 0.934 (training set) and 0.868 (validation set), outperforming the original model (P1).

Conclusion

The SMOTE-based model may help predict AVF restenosis within 12 months after PTA, but it is an exploratory internally validated tool and requires external validation before clinical use.