Machine learning-based prediction model for arteriovenous fistula thrombosis risk: a retrospective cohort study from 2017 to 2024
Abstract Background Thrombosis of arteriovenous fistulas represents a prevalent complication among patients undergoing hemodialysis, characterized by a notably high incidence rate. Presently, there is an absence of robust assessment tools capable of predicting thrombosis occurrence. This study seeks...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-07-01
|
| Series: | BMC Nephrology |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12882-025-04201-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Background Thrombosis of arteriovenous fistulas represents a prevalent complication among patients undergoing hemodialysis, characterized by a notably high incidence rate. Presently, there is an absence of robust assessment tools capable of predicting thrombosis occurrence. This study seeks to develop an interpretable machine learning model to forecast the risk of arteriovenous fistula thrombosis. Methods Clinical data were retrospectively collected from 1,168 patients who received hemodialysis via arteriovenous fistulas at The Central Hospital of Wuhan between January 2017 and October 2024. A comprehensive analysis of 55 features was conducted utilizing Python. The dataset was partitioned into a training set and a test set, comprising 70% and 30% of the samples, respectively. Six distinct machine learning models—namely, Random Forest, Extreme Gradient Boosting, Decision Tree, Logistic Regression, K-Nearest Neighbors, and Naive Bayes—were constructed to predict the risk of thrombosis in arteriovenous fistulas. The performance of these models was assessed utilizing several metrics, including the F1 score, precision, specificity, accuracy, area under the receiver operating characteristic curve, and recall rate. The contribution of each feature within the most effective model was evaluated using SHAP values, and a specific case was selected to demonstrate the model’s predictive capability. Results The study encompassed a cohort of 974 patients, each characterized by 55 clinical data features. Among the six machine learning models evaluated, the Random Forest model demonstrated superior performance, achieving an AUC-ROC of 0.984. SHAP visualization analysis identified the number of surgeries, stenosis, free fatty acids, platelet count, and C-reactive protein as the five most significant features influencing the risk of arteriovenous fistula thrombosis. Conclusion We developed a Random Forest model based on patients’ clinical data, which effectively predicts the risk of thrombosis in arteriovenous fistulas. SHAP analysis offers the potential to inform personalized and evidence-based nursing interventions for healthcare professionals. |
|---|---|
| ISSN: | 1471-2369 |