Objective: To develop an interpretable machine learning model for predicting arteriovenous fistula (AVF) stenosis in hemodialysis patients using inflammatory biomarkers and identify key influencing factors. Methods: A retrospective cohort study was conducted on 974 end-stage renal disease patients undergoing hemodialysis with AVF at The Central Hospital of Wuhan (2017– 2024). Clinical data (demographics, comorbidities, inflammatory markers) were collected. After data preprocessing (imputation, normalization, feature selection), eight machine learning models including Logistic Regression (LR) were built and validated via 10-fold cross-validation. SHAP (SHapley Additive Explanations) was used to interpret model outputs. Results: The LR model outperformed others with an AUC of 0.833 (95% confidence interval CI: 0.796– 0.868), an accuracy of 0.782 (95% CI: 0.751– 0.811), and an F1 score of 0.756 (95% CI: 0.718– 0.791). Key factors associated with AVF stenosis included AVF surgical history, thrombosis history, comorbidities, smoking, alcohol consumption, monocyte-to-high-density lipoprotein cholesterol ratio (MHR), and platelet-to-high-density lipoprotein cholesterol ratio (PHR) (p < 0.05). SHAP visualization showed these factors significantly impacted model predictions, with MHR/PHR correlating with reduced stenosis risk when elevated. Conclusion: The LR model based on inflammatory biomarkers effectively predicts AVF stenosis. Integrating SHAP (SHapley Additive Explanations) values enhances the interpretability of the model, thus providing a practical tool for clinical risk stratification and early intervention of AVF stenosis in hemodialysis patients. Keywords: arteriovenous fistula stenosis, hemodialysis, metabolism-integrated inflammatory biomarkers, machine learning, SHAP value
Wang et al. (Sun,) studied this question.