Construction of Risk-Prediction Models for Autogenous Arteriovenous Fistula Thrombosis in Patients on Maintenance Hemodialysis.

INTRODUCTION

Autogenous arteriovenous fistula (AVF) is the preferred vascular access in patients undergoing maintenance hemodialysis (MHD). However, complications such as thrombosis may occur. This study aimed to construct and validate a machine learning-based risk-prediction model for AVF thrombosis, hypothesizing that such a model can effectively predict occurrences, providing a foundation for early clinical intervention.

METHODS

The retrospective longitudinal study included a total of 270 patients who underwent MHD at the Hemodialysis Center of the Second Affiliated Hospital of Harbin Medical University between March 2021 and December 2022. During this study, baseline data and scale information of patients between March 2020 and December 2021 were collected. We recorded outcome indicators between March 2021 and December 2022 for subsequent analyses. Five machine learning models were developed (artificial neural network, logistic regression, ridge classification, random forest, and adaptive boosting). The sensitivity (recall), specificity, accuracy, and precision of each model were evaluated. The effect size of each variable was analyzed and ranked. Models were assessed using the area under the receiver-operating characteristic (AUROC) curve.

RESULTS

Among the 270 included patients, 105 had AVF thrombosis (55 male and 50 female patients; age range, 29-79 years; mean age, 56.72 years; standard deviation [SDs], ±13.10 years). Conversely, 165 patients did not have AVF thrombosis (99 male and 66 female patients; age range, 23-79 years; mean age, 53.58 years; SD, ± 13.33 years). During the observation period, approximately 52.6% of patients with AVF experienced long-term complications. The most common complications associated with AVF were thrombosis (105; 38.9%), aneurysm formation (27; 10%), and excessively high output flow (10; 3.7%). Fifty-four (20%) patients with AVF required intervention because of complications associated with vascular access. The AUROC curve of the testing set was between 0.858 and 0.903.

CONCLUSION

In this study, we developed five machine learning models to predict the risk of AVF thrombosis, providing a reference for early clinical intervention.