Role of rivaroxaban in arteriovenous graft thrombosis after endovascular treatment and establishment and evaluation of a nomogram predictive model for postoperative recurrent thrombosis risk

Abstract

Objective

To evaluate the role of rivaroxaban in preventing arteriovenous graft (AVG) thrombosis following endovascular intervention, assess the risk of recurrent thrombosis, and develop a clinically applicable predictive model.

Methods

A retrospective analysis was performed on 108 patients with AVG thrombosis who underwent endovascular treatment at our center between October 2020 and October 2024. Patients were stratified into a rivaroxaban group (n = 71) and a control group (n = 37). Fifteen variables from the two groups were analyzed and summarized. Four machine learning algorithms (random forest, Support Vector Machine – Recursive Feature Elimination, Least Absolute Shrinkage and Selection Operator, and XGboost) were applied to identify risk factors for recurrent thrombosis. A predictive model was constructed using logistic regression and validated with an internal dataset.

Results

Among the 108 patients, 69 were male and 39 were female, with ages ranging from 21 to 89 years. Fibrinogen levels were significantly lower in the rivaroxaban group (4.02 ± 0.87 vs. 4.59 ± 1.03, p < 0.01), More notably, the 12-month postprocedural patency rate was significantly higher in the rivaroxaban group (75%) than in the nonrivaroxaban group (17%), with a statistically significant difference (p < 0.01). No statistically significant differences were observed between the two groups in other aspects (p > 0.05). The model exhibited a C-index of 0.87 (95% CI: 0.78–0.95). The receiver operating characteristic and decision curve analysis curves demonstrated that the multifactor model had superior discriminative ability and net clinical benefit for identifying recurrent AVG thrombosis compared to single factors (AUC_{linear predictor}= 0.87, AUC_{rivaroxaban use}= 0.78, AUC_D−dimer= 0.60). Validation in the internal validation set and the entire cohort confirmed good calibration and efficacy (validation set: AUC = 0.86, entire cohort: AUC = 0.85).

Conclusion

Rivaroxaban may exert a positive effect on maintaining the patency of AVG fistulas after endovascular treatment for thrombosis. The constructed nomogram prediction model can be used to predict the risk of recurrent thrombosis following endovascular treatment of AVG fistula thrombosis.

Keywords

Retrospective analysis arteriovenous graft thrombosis rivaroxaban risk factors nomogram

Introduction

For patients with end-stage renal disease (ESRD), clinical practice guidelines for vascular access issued by the National Kidney Foundation's Kidney Disease Outcomes Quality Initiative indicate that arteriovenous graft (AVG) have become a critical option for dialysis access.¹ The 2-year patency rate of AVGs is approximately 50%.² As the lifeline for ESRD patients, establishing and maintaining a functional dialysis access is fundamental to ensuring effective hemodialysis. Existing studies have identified multiple factors influencing fistula patency, including age, blood glucose, body mass index (BMI), blood lipids, calcium-phosphorus metabolism, peripheral vascular disease, vascular anatomical characteristics, surgical techniques, and puncture practices. Intimal hyperplasia and thrombosis represent the primary causes of failure in most arteriovenous accesses.^3,4 Clinically, management of AVG thrombosis primarily relies on two strategies: open surgical intervention and endovascular therapy. A prior study demonstrated that compared with open surgery, endovascular therapy is associated with lower posttreatment occlusion rates, offers greater minimally invasive benefits, and preserves the original hemodialysis access—attributes that make it the preferred approach in clinical practice.⁵ Notably, patients with AVG thrombosis face an increased risk of recurrent thrombosis following endovascular treatment. Existing limited studies on endovascular treatment of AVG report a primary patency rate of 23–68% at 12 months postprocedure.⁶ Therefore, the optimal use of postoperative prophylactic medications holds significant importance. A review of recent guidelines and literature in the hemodialysis access field reveals inconsistent conclusions regarding the efficacy of prophylactic oral agents (including fish oil, lipid-lowering drugs, antiplatelet agents, and anticoagulants) in maintaining vascular access function. Although several guidelines provide relevant recommendations, their evidence levels and recommendation grades remain relatively low.^7–10 In recent years, our center has implemented rivaroxaban for prophylactic treatment in patients undergoing endovascular therapy for AVG thrombosis, with favorable outcomes observed during follow-up. Building on this experience, we further investigated the risk of recurrent dysfunction after endovascular treatment for AVG thrombosis and developed a predictive model to facilitate early clinical intervention, thereby reducing the risk of recurrent AVG thrombosis.

Materials and methods

General data

A retrospective analysis was performed on 108 patients with AVG thrombosis who underwent endovascular treatment at Fujian Provincial People's Hospital between October 2020 and October 2024. Patients were stratified into the rivaroxaban group (n = 71) and blank control group (n = 37) based on their postoperative treatment regimens. Inclusion criteria: (1) patients with chronic kidney disease receiving maintenance hemodialysis (MHD); (2) patients with first-episode AVG thrombosis; (3) patients without hematological disorders or coagulation dysfunction; (4) patients who provided signed informed consent; (5) patients with AVG located in the forearm. Exclusion criteria: (1) patients who died within 1 year postoperatively; (2) patients with missing baseline data; (3) patients taking antiplatelet agents, anticoagulants, or lipid-lowering drugs other than rivaroxaban postoperatively; (4) patients lost to follow-up. This is an observational study, which adheres to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines and the Helsinki Declaration of 1975 as revised in 2024.¹¹

Treatment methods

All endovascular procedures were performed by the same surgical team. Patients in the rivaroxaban group received long-term oral rivaroxaban (10 mg once daily) starting after the endovascular intervention, administered in the morning with the first bite of food, except on hemodialysis days.

Observation indicators

Clinical data including age, gender, BMI, hypertension, diabetes, prothrombin time (PT), activated partial thromboplastin time (APTT), thrombin time (TT), fibrinogen (FIB), D-dimer, triglycerides, cholesterol, serum calcium, serum phosphorus, 12-month patency following endovascular treatment for AVG thrombosis and Bleeding events (recorded and classified according to the International Society on Thrombosis and Haemostasis criteria: major bleeding included intracranial hemorrhage, massive gastrointestinal bleeding requiring blood transfusion or surgical intervention, retroperitoneal hemorrhage, and others; minor bleeding included gingival bleeding, skin ecchymosis, hematoma, epistaxis lasting < 5 min) were collected to evaluate the role of rivaroxaban in this patient population.

Machine learning

Four machine learning models random forest (RF), Support Vector Machine – Recursive Feature Elimination (SVM-RFE), Least Absolute Shrinkage and Selection Operator (Lasso), and eXtreme Gradient Boosting (XGBoost) were employed to identify risk factors for recurrent thrombosis following endovascular treatment of AVG thrombosis. Random forest, a supervised learning algorithm based on decision trees, enhances prediction accuracy by aggregating multiple classification trees and was implemented using the R package “randomForest” (ntree = 500). The SVM-RFE, a feature selection algorithm based on embedded methods widely applied in pattern recognition and machine learning, was executed using the R packages “e1071,” “kernlab,” and “caret.” Lasso, a regularization method that mitigates overfitting through model simplification, was analyzed via the R package “glmnet.” The XGBoost algorithm integrates multiple low-accuracy tree models into a high-performance predictive model, implemented using the R packages “xgboost,” “tidyverse,” and “caret.” Finally, overlapping risk factors identified by all four models were designated as the key risk factors for recurrent thrombosis after endovascular treatment of AVG thrombosis.^12–15

Identification of risk factors and generalization performance for recurrent thrombosis after endovascular treatment for AVG thrombosis

Patients were randomly divided into a training set and a validation set at a ratio of 7:3 to verify the generalization performance of the model. Univariate and multivariate logistic regression analyses were used to construct and evaluate a model for predicting risk factors for recurrent thrombosis after endovascular treatment for AVG thrombosis. Linear predictors for each sample were calculated based on regression coefficients to derive predicted probabilities, with specific formulas as follows:

Linear_predictors_Intercept + \sum_{i = 1}^{n} {Expression}_{gene_i} \times Logistic_coeffieicen t_{gene_i}

predicted_probability = \frac{e^{Linear_predictors}}{1 + e^{Linear_predictors}}

Model correlation tests were performed in R using the “rms” and “ResourceSelection” packages. Multicollinearity was detected by calculating the variance inflation factor (VIF). Goodness of fit was evaluated using the likelihood ratio test and calibration curves. Predictive ability was measured using the C-index and receiver operating characteristic (ROC) curves. Decision curve analysis (DCA) curves were used as indicators of clinical effectiveness.

Statistical methods

All statistical analyses were performed using R software (version 4.4.1). Measurement data with a normal distribution were described as X ± SD, while non-normally distributed data were described as M (IQR). For intergroup comparisons, independent sample t-tests were used for normally distributed data, and rank-sum tests for non-normally distributed data. Categorical data were described as n (%) and compared using the standard chi-square test when n ≥ 40 and T ≥ 5, the corrected chi-square test when n ≥ 40 and 1 < T < 5, and Fisher's exact test when n < 40 or T ≤ 1. Binary logistic regression analysis was used for regression analysis with binary dependent variables. A p-value <0.05 was considered statistically significant.

Results

Univariate analysis of clinical data

Among the 108 patients, there were 69 males and 39 females, with ages ranging from 21 to 89 years. The mean age was 58.62 ± 14.91 years in the blank control group and59.42 ± 14.24 years in the rivaroxaban group. One patient in the rivaroxaban group developed a pseudoaneurysm. As shown in Table 1, the FIB level in the rivaroxaban group was significantly lower than that in the blank control group, with a statistically significant difference (t = –3.00, p < 0.01). The number of cases with 12-month postoperative patency in the rivaroxaban group was significantly higher than that in the blank control group, and the difference was statistically significant (χ² = 36.03, p < 0.01). The distribution of outcomes between the two groups is illustrated in Figure 1. In addition, there were no statistically significant differences between the two groups in terms of age, gender, BMI, history of underlying diseases, PT, APTT, TT, D-dimer, platelet count, triglycerides, cholesterol, calcium, or phosphorus (all p > 0.05).

Figure 1.

The ratio of patients with dysfunction and nondysfunction between the two groups.

Table 1.

Clinical data of 108 patients.

Characteristics	Without rivaroxaban group	With rivaroxaban group	p value
n	37	71
AVF dysfunction, n (%)			<0.01
Nondysfunction	9/37 (24.32%)	59/71 (83.10%)
Dysfunction	28/37 (75.68%)	12/71 (16.90%)
Genders, n (%)			0.49
Female	15/37 (40.54%)	24/71 (33.80%)
Male	22/37 (59.46%)	47/71 (66.20%)
Hypertension, n (%)			0.98
No	10/37 (27.03%)	19/71 (26.76%)
Yes	27/37 (72.97%)	52/71 (73.24%)
Diabetes, n (%)			0.19
No	22/37 (59.46%)	51/71 (71.83%)
Yes	15/37 (40.54%)	20/71 (28.17%)
Bleeding event, n (%)			1.00
No	37/37(100%)	70/71(99%)
Yes	0/37(0%)	1/71(1%)
Age, mean ± SD	58.62 ± 14.91	59.42 ± 14.25	0.79
PT, median (IQR)	12.80 (12.20, 13.40)	12.80 (12.05, 13.35)	0.72
APTT, median (IQR)	36.90 (33.80, 42.70)	36.30 (33.35, 42.45)	0.75
TT, median (IQR)	17.80 (16.60, 19.00)	18.10 (17.05, 19.05)	0.21
FIB, mean ± SD	4.59 ± 1.03	4.02 ± 0.87	0.01
D-dimer, median (IQR)	0.59 (0.33, 1.56)	0.63 (0.38, 1.10)	0.72
BMI, median (IQR)	22.06 (19.75, 24.81)	23.14 (20.54, 25.28)	0.20
PLT, mean ± SD	216.3 ± 55.78	195.03 ± 55.47	0.06
Triglyceride, median (IQR)	1.7 (1.31, 2.56)	1.59 (1.21, 2.52)	0.35
Cholesterol, mean ± SD	4.02 ± 1.23	4.16 ± 1.23	0.58
Calcium, mean ± SD	2.18 ± 0.25	2.25 ± 0.28	0.21
Phosphorus, median (IQR)	1.78 (1.59, 2.19)	2.02 (1.46, 2.41)	0.35

BMI, body mass index; PT, prothrombin time; APTT, activated partial thromboplastin time; TT, thrombin time; FIB, fibrinogen.

Identification of risk factors for recurrent thrombosis after endovascular treatment of AVG based on machine learning

Four machine learning algorithms were applied to identify risk factors for recurrent thrombosis after endovascular treatment of AVG.For the Lasso algorithm, the model fitting was optimal when λ min was 6 (Figure 2A). The risk factors included in the Lasso model were rivaroxaban, diabetes, APTT, D-dimer, triglycerides, and cholesterol. For the SVM-RFE algorithm, the root mean square error (RMSE) of 10-fold cross-validation was the smallest when the number of feature genes was 13. The screened risk factors included 13 items such as rivaroxaban, D-dimer, APTT, triglycerides, TT, PT, serum calcium, and serum phosphorus (Figure 2B). For the RF algorithm, the number of iterations of the RF classifier was set to 500, resulting in a stable out-of-bag (OOB) error of less than 0.3 (Figure 2C). Using the RF algorithm with a Gini coefficient greater than 2 as the screening criterion, 10 risk factors including rivaroxaban, D-dimer, APTT, triglycerides, serum calcium, and serum phosphorus were identified as specific factors for recurrent thrombosis after endovascular treatment of AVG. For the XGboost algorithm, the histogram of risk factor importance showed that rivaroxaban ranked first, followed by D-dimer, cholesterol, PT, BMI, and APTT (Figure 2D).

Figure 2.

Identification of risk factors for recurrent thrombosis after endovascular treatment of AVG based on machine learning. (A) The LASSO algorithm identified 6 risk factors based on λ min value. (B) The SVM-RFE algorithm identified 13 risk factors based on RMSE. (C) The RF algorithm identified 10 risk factors based on OOB error and MDG. (D) The XGboost algorithm identified 13 risk factors and ranked them by importance. (E) A Venn diagram was plotted by overlapping the results of the above four algorithms, yielding four risk factors. Lasso, least absolute shrinkage and selection operator; SVM-RFE, support vector machine-recursive feature elimination; RMSE, root mean square error; RF, random forest; OOB, out-of-bag; MDG, mean decrease Gini; AVG, arteriovenous graft.

Finally, four overlapping factors were identified: Rivaroxaban, APTT, D-dimer, and triglycerides (Figure 2E). To obtain the optimal model, the 108 patients were randomly divided into a training set and a validation set at a ratio of 7:3, and logistic regression was performed.

Establishment and validation of the risk factor model for recurrent thrombosis after endovascular treatment of AVG thrombosis

Univariate and multivariate logistic regression analyses showed a strong association between rivaroxaban and a reduced risk of recurrent thrombosis (OR = 0.01, 95%CI = 0.03–0.17, p < 0.01); D-dimer was a risk factor, and elevated levels might increase the risk of events (OR = 1.37, 95%CI = 1.01–1.84, p = 0.04); other characteristics such as gender, age, hypertension, diabetes, and most coagulation indicators were not significantly associated with the outcome events (Table S1 and Table 2). A model was established based on these results. The regression coefficients of the risk factors (Table 3) were used to calculate the linear predictors for each sample to identify recurrent AVG thrombosis. The model is shown in Figure 3A, and its performance is detailed in Table 4. The VIF values of all four risk factors included in the model were less than 3, indicating no severe multicollinearity (Rivaroxaban_No = 1.24, APTT = 1.06, D-dimer = 1.02, Triglyceride = 1.19).

Figure 3.

Establishment and validation of the risk factor model for recurrent thrombosis after endovascular treatment of AVG thrombosis. (A) Nomogram for identifying recurrent thrombosis after endovascular treatment of AVG thrombosis, established by logistic regression based on four important risk factors. (B–D) Validation of the efficacy of identifying recurrent thrombosis by calibration curve, ROC curve, and DCA curve in the following datasets: (B) Training set, AUCLinear predictor = 0.865; (C) Validation set, AUCLinear predictor = 0.858; (D) All sets, AUCLinear predictor = 0.853. ROC, receiver operating characteristic; DCA, decision curve analysis; AVG, arteriovenous graft.

Table 2.

Multivariate regression analysis of factors affecting patency at 12 months after surgery.

Characteristics	Total (N)	Multivariate analysis
Characteristics	Total (N)	Odds ratio (95% CI)	p value
Rivaroxaban	108
No	37	Reference
Yes	71	0.06 (0.02–0.17)	<0.01
D-dimer	108	1.41 (1.02–1.95)	0.04

Table 3.

Coefficients of variables in the multivariate model for recurrent AVG thrombosis postendovascular treatment.

Multifactor model variables	Coefficients
Intercept	2.784
Rivaroxaban No	−3.283
APTT	−0.045
D-dimer	−0.455
Triglyceride	0.561

Note: $Linear_predictors_Intercept + \sum_{i = 1}^{n} {Expression}_{gene_i} \times Logistic_coeffieicen t_{gene_i}$

$predicted_probability = \frac{e^{Linear_predictors}}{1 + e^{Linear_predictors}}$

AVG, arteriovenous graft; APTT, activated partial thromboplastin time.

Table 4.

Assessment parameters of the model for the risk of thrombus reformation.

Item	Statistic	p value
Model
Likelihood ratio test	Chi-square value = 33.306	<0.001
C-index	0.865 (0.781–0.950)	<0.001
Hosmer–Lemeshow goodness-of-fit statistic	Chi-square value = 7.2074	0.5144
Item	VIF
Rivaroxaban_No	1.2407
APTT	1.0568
D-dimer	1.0229
Triglyceride	1.1916

VIF: variance inflation factor; APTT, activated partial thromboplastin time.

The chi-square value of the likelihood ratio test was 33.31 (p < 0.01). The Hosmer–Lemeshow goodness-of-fit statistic showed no significant difference (chi-square = 7.21, p = 0.51). The apparent line was similar to the bias-corrected line and close to the ideal line (Figure 3B). The C-index of the model was 0.87 (0.78–0.95). The ROC curve and DCA curve indicated that the model combining multiple risk factors was superior to the identification using a single risk factor in terms of effectiveness and net benefit in identifying recurrent AVG thrombosis (AUC_{linear predictor}= 0.87, AUC_Rivaroxaban= 0.78, AUC_D-dimer= 0.60). The validation results of the validation set and the entire set showed that the model had a good fit and excellent efficacy (Figure 3C and D, validation group: AUC_{linear predictor}= 0.86; all groups: AU_{Clinear predictor}= 0.85).

Discussion

In patients with poor native vascular conditions, AVG have become an alternative to autologous arteriovenous fistulas (AVF). However, with the increasing proportion of elderly and frail patients in the dialysis population, the maturation and patency rates of fistulas have declined accordingly. Thrombosis of the access accounts for 65–85% of permanent access loss.¹⁶

Rivaroxaban is a novel oral anticoagulant that inhibits factor Xa, widely used for the prevention and treatment of venous thromboembolic diseases.¹⁷ Previous studies have shown that rivaroxaban can exert vascular protective effects through various mechanisms.¹⁸ Hara et al. found that rivaroxaban blocks the inflammatory activation of macrophages and smooth muscle cells by inhibiting factor Xa, thereby attenuating neointimal formation after vascular injury.¹⁹ Combined with the data from our center, it was shown that the patency of the fistula was improved in patients treated with rivaroxaban after endovascular therapy for AVG thrombosis. Additionally, studies have reported that rivaroxaban is rapidly absorbed, reaching peak plasma concentrations 2–4 h after administration, with high oral bioavailability of 80–100%. The pharmacokinetic variability is considered moderate (coefficient of variation 30–40%).²⁰ A single-center experience and literature review from Russia reported that among 30 included patients with upper extremity deep vein thrombosis, no recurrent symptomatic venous thromboembolism or asymptomatic upper extremity deep vein thrombosis occurred during 6-month follow-up with rivaroxaban, and no major bleeding episodes were observed. This suggests that rivaroxaban appears to be safe and effective in maintaining vascular patency.²¹ Dakis et al. summarized and analyzed existing studies on the effectiveness of antithrombotic therapy in AVG patients, pointing out that studies on anticoagulant therapy are limited, and evidence regarding the impact of direct oral anticoagulants on AVG patency is extremely scarce, with only supportive reports on the application of apixaban.²² Therefore, there is no clear consensus on whether anticoagulants should be used after AVG thrombus recanalization to prevent recurrent thrombosis. The results of this study, which used rivaroxaban as a postoperative prophylactic agent to maintain AVG patency and the incidence of bleeding events, suggest that rivaroxaban has great potential and positive clinical significance.

The risk of recurrent thrombosis after AVG thrombus recanalization is high, which not only increases patient suffering but also affects the continuity of dialysis treatment. The development of accurate predictive models is crucial for identifying the risk of recurrent thrombosis and implementing early interventions. Nomogram models, with their intuitive visual interface and ability to quickly derive numerical results without complex equations, can help clinicians assess risks and make decisions conveniently, and have been widely used as prognostic tools in multiple disciplines.²³ Applying them to predict the risk of recurrent thrombosis after AVG thrombus recanalization in MHD patients has important potential. Based on existing studies on risk factors for AVF and AVG, we selected 13 risk factors, including age, gender, BMI, hypertension, diabetes, PT, APTT, TT, FIB, D-dimer, triglycerides, cholesterol, serum calcium, serum phosphorus, and the intervention factor rivaroxaban used in this study, for risk prediction analysis in the cases of this study.^24–33 We used four machine learning models (RF, SVM-RFE, Lasso, and XGboost—to further screen these 13 initially selected risk factors. Each risk factor had different weights in the machine learning models. For more precise screening, we constructed a Venn diagram of the risk factors identified by these four machine learning models, and finally obtained four highest-weight risk factors: APTT, D-dimer, triglycerides, and rivaroxaban.

These four risk factors are all closely related to thrombosis. Among them, APTT is closely associated with various coagulation factors, prothrombin, and FIB; a shortened APTT may lead to thrombosis.³⁴ The impact of D-dimer on hemodialysis access remains controversial. This study indicates that elevated peripheral D-dimer is a risk factor for recurrent thrombosis after AVG recanalization. In reviewing previous studies, we found that several researchers have pointed out that D-dimer is an independent risk factor for AVF dysfunction, which may be because its elevated level reflects the degree of thrombosis and a state of debilitation, being associated with an increased risk of access thrombosis.^35–37 Triglycerides are also a high-risk factor for recurrent AVG thrombosis. Multiple previous studies have shown that elevated triglyceride levels can enhance platelet sensitivity to ADP, lower the aggregation threshold, or inhibit the conversion of plasminogen to plasmin by upregulating PAI-1 levels, leading to impaired thrombolysis.^38–40 We converted the selected variables such as rivaroxaban, APTT, D-dimer, and triglycerides into intuitive scores to facilitate rapid clinical estimation of individual risks. In all sets, the linear prediction model integrating multiple indicators showed better discriminative ability than single indicators, indicating that multifactor synergy can more accurately identify populations at risk of recurrent thrombosis. The net benefit curve further verified that the linear prediction model had higher net benefits at different risk thresholds, with significant clinical utility. The calibration curve showed that the predicted probability was well-fitted with the actual probability, demonstrating good model calibration and high reliability. Compared with single indicators (such as rivaroxaban and D-dimer), the multi-indicator model has obvious advantages. As a new oral anticoagulant, rivaroxaban has attracted attention for its role in AVG thrombosis management. The negative association between rivaroxaban and recurrent thrombosis in the model suggests that standardized use of rivaroxaban may reduce the recurrence risk, which is consistent with the importance of anticoagulant therapy in clinical practice. In patients with early recurrent AVG thrombosis, obvious symptoms are often absent, but D-dimer levels rise prior to the onset of symptoms. Regular postoperative D-dimer reexaminations can monitor the anticoagulant effect of rivaroxaban, and a reexamination one week after surgery is recommended. As a classic marker of thrombosis, the predictive value of D-dimer for recurrence risk in the model echoes the mechanism of coagulation-fibrinolysis imbalance in the pathological process of thrombosis. In this study, although preoperative FIB levels showed a statistically significant difference between the two groups, FIB was not included as a risk factor in the predictive model. This may be attributed to the small sample size of the study, which prevented the model from capturing the predictive value of FIB. However, existing studies have confirmed that in dialysis patients with a FIB level > 4.5 g/L, the risk of AVG thrombosis recurrence increases by 2.1–2.8 folds. Therefore, regular postoperative reexamination of FIB levels is of certain necessity, and a reexamination one week after surgery is recommended.⁴¹ However, it should be noted that individual clinical situations are complex, such as patients with multiple underlying diseases and differences in treatment compliance. Although the model shows good performance, its practical application still needs to be comprehensively judged in combination with clinical practice.

The limitations of this study are as follows: (1) This study included a small number of patients and was a single-center study. A small sample size may lead to overestimation of the study results. The model was established with only internal validation rather than external validation, which introduces a certain degree of bias and limits its representativeness in a broader population. (2) The follow-up duration of this study was only 1 year, which is insufficient. Longer-term follow-up should be conducted in future studies. (3) Thrombosis recurrence is a time-dependent event. However, the logistic regression model and the 12-month binary endpoint used in this study failed to fully utilize time-to-event data and could not handle censored data. Although supplementary analysis showed that the Cox model was consistent with our main conclusions, future studies should adopt survival analysis as the primary method to provide more reliable and dynamic risk assessment. (4) The risk factors included in this study were not comprehensive. More comprehensive risk factors should be collected in subsequent studies. (5) This study did not conduct further statistical analysis on factors related to dialysis procedures and surgical operations in patients. Data collection in this aspect should be improved in future studies. (6) Due to the limited number of existing studies on the prophylactic use of rivaroxaban after endovascular treatment of AVG, we only conducted a retrospective analysis of patients who received empirical medication in the past, without adopting randomization. In the future, we will expand the sample size and use the “propensity score matching” method to balance the baseline differences between the two groups, so as to further verify the efficacy of rivaroxaban.

Conclusion

Rivaroxaban demonstrates efficacy in preventing recurrent thrombosis following AVG recanalization. The predictive model established based on risk factors screened by four machine learning algorithms can provide certain theoretical references for clinicians when evaluating the risk of recurrent thrombosis after AVG recanalization, and offer assistance in reducing such risk. In the next step, we will collaborate with other centers to collect more patient data and incorporate additional risk factors. This will help further clarify the efficacy of rivaroxaban, optimize the model development, and enhance the stability and accuracy of the model.

Supplemental Material

sj-docx-1-sci-10.1177_00368504251406564 - Supplemental material for Role of rivaroxaban in arteriovenous graft thrombosis after endovascular treatment and establishment and evaluation of a nomogram predictive model for postoperative recurrent thrombosis risk

Supplemental material, sj-docx-1-sci-10.1177_00368504251406564 for Role of rivaroxaban in arteriovenous graft thrombosis after endovascular treatment and establishment and evaluation of a nomogram predictive model for postoperative recurrent thrombosis risk by Mingwei Wang, Tingrong You, Ran Zhang, Bin Liu, Xiangyu Peng, Qingjin Huang and Shiyao Zheng in Science Progress

Footnotes

Acknowledgements

The authors thank all those who participated in this study.

ORCID iD

Shiyao Zheng

Ethics approval and consent to participate

This study has been reviewed and exempted by the Institutional Review Board (IRB) of Fujian Provincial People's Hospital.

Consent for publication

Informed consent was obtained from all individual participants included in the study.

Author contributions

Mingwei Wang was responsible for study design and writing. Bin Liu and Qingjin Huang were involved in the study design and were responsible for scientific revision. Mingwei Wang, Tingrong You, and Ran Zhang contributed equally to this paper. Mingwei Wang, Tingrong You, Shiyao Zheng, and Xiangyu Peng were responsible for the data collection and analysis. Mingwei Wang and Shiyao Zheng contributed to the image painting. All authors have read and approved the final manuscript.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supplemental material

Supplemental material for this article is available online.

References

Lok

Huber

Lee

, et al. KDOQI Vascular Access Guideline Work Group. KDOQI clinical practice guideline for vascular access: 2019 update. Am J Kidney Dis 2020; 75: S1–S164.

Miller

Carlton

Deierhoi

, et al. Natural history of arteriovenous grafts in hemodialysis patients. Am J Kidney Dis 2000; 36: 68–74.

Brahmbhatt

Remuzzi

Franzoni

, et al. The molecular mechanisms of hemodialysis vascular access failure. Kidney Int 2016; 89: 303–316.

Aitken

Jackson

Kong

, et al. Renal function, uraemia and early arteriovenous fistula failure. BMC Nephrol 2014; 15: 79.

Chan

Wee

Soong

, et al. A systematic review and meta-analysis of surgical versus endovascular thrombectomy of thrombosed arteriovenous grafts in hemodialysis patients. J Vasc Surg 2019; 69: 1976–1988. e7.

Shemesh

Goldin

Berelowitz

, et al. Thrombolysis for early failure of prosthetic arteriovenous access. J Vasc Surg 2008; 47: 585–590.

Irish

Viecelli

Hawley

, et al. Effect of fish oil supplementation and aspirin use on arteriovenous fistula failure in patients requiring hemodialysis: a randomized clinical trial. JAMA Intern Med 2017; 177: 184–193.

Young

Goodkin

Mapes

, et al. The Dialysis Outcomes and Practice Patterns Study (DOPPS): an International Hemodialysis Study. Kidney Int 2000; 57: S74–S81.

Kukita

Ohira

Amano

, et al. Update Japanese Society for Dialysis Therapy guidelines of vascular access construction and repair for chronic hemodialysis. Ther Apher Dial 2011; 2015: 19. (Suppl 1): 1-39.

10.

Crowther

Clase

Margetts

, et al. Low-intensity warfarin is ineffective for the prevention of PTFE graft failure in patients on hemodialysis: a randomized controlled trial. J Am Soc Nephrol 2002; 13: 2331–2337.

11.

Von Elm

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med 2007; 147: 573–577.

12.

Breiman

. Random forests. Mach Learn 2001; 45: 5–32.

13.

Guyon

, et al. Gene selection for cancer classification using support vector machines. Mach Learn 2002; 46: 389–422.

14.

Friedman

Hastie

Tibshirani

. Regularization paths for generalized linear models via coordinate descent. J Stat Softw 2010; 33: 1–22.

15.

Chen

Guestrin

. XGBoost: a scalable tree boosting system.

In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining,

2016, pp.785–794.

16.

Quencer

Friedman

. Declotting the thrombosed access. Tech Vasc Interv Radiol 2017; 20: 38–47.

17.

Ajmal

Friedman

Sipra

QUAR

, et al. Rivaroxaban: expanded role in cardiovascular disease management—A literature review. Cardiovasc Ther 2021: 8886210.

18.

Sanmartín

Bellmunt

Cosín-Sales

, et al. Role of rivaroxaban in the prevention of atherosclerotic events. Expert Rev Clin Pharmacol 2019; 12: 771–780.

19.

Hara

Fukuda

Tanaka

, et al. Inhibition of activated factor X by rivaroxaban attenuates neointima formation after wire-mediated vascular injury. Eur J Pharmacol 2018; 820: 222–228.

20.

Kvasnicka

Malikova

Zenahlikova

, et al. Rivaroxaban—metabolism, pharmacologic properties and drug interactions. Curr Drug Metab 2017; 18: 636–642.

21.

Schastlivtsev

Lobastov

Tsaplin

, et al. Rivaroxaban in the treatment of upper extremity deep vein thrombosis: a single-center experience and review of the literature. Thromb Res 2019; 181: 24–28.

22.

Hedayat

Soltani

Hakiminezhad

, et al. Efficacy and safety of apixaban for the prevention of thrombosis in arteriovenous grafts. Eur J Trans Myol 2024; 34: 12029.

23.

Chen

Qin

, et al. Development and validation of dynamic nomogram of frailty risk for older patients hospitalized with heart failure. Int J Nurs Sci 2023; 10: 142–150.

24.

Sanada

Miyasaka

Kanno

, et al. Efficacy of statin on vascular access patency in diabetic hemodialysis patients. J Vasc Access 2017; 18: 295–300.

25.

Yoshida

Doi

Nakashima

, et al. Different risk factors are associated with vascular access patency after construction and percutaneous transluminal angioplasty in patients starting hemodialysis. J Vasc Access 2021; 22: 707–715.

26.

Wärme

Hadimeri

Nasic

, et al. The association of erythropoietin-stimulating agents and increased risk for AV-fistula dysfunction in hemodialysis patients. A retrospective analysis. BMC Nephrol 2021; 22: 30.

27.

Misskey

Faulds

Sidhu

, et al. An age-based comparison of fistula location, patency, and maturation for elderly renal failure patients. J Vasc Surg 2018; 67: 1491–1500.

28.

Ying

. Causes and care of arteriovenous fistula thrombosis in 102 hemodialysis patients. Chin J Nurs 2020; 55: 607–610.

29.

Jin

, et al. Construction of risk prediction models for autogenous arteriovenous fistula thrombosis in patients on maintenance hemodialysis. Chin Blood Purif 2024; 23: 209–213.

30.

Valerianova

Kudlicka

Chytilova

, et al. Factors influencing dialysis arteriovenous graft survival. J Vasc Access 2017; 18: 139–143.

31.

Montoya

Del Castillo Rix

Polania-Sandoval

, et al. Blood pressure variability and cardiovascular risk assessment using machine learning clustering after arteriovenous fistula creation in hemodialysis patients. J Vasc Access. 2025; 11297298251326313.

32.

Fisher A

Mulaney-Topkar

Sheehan B

, et al. Association between heart failure and arteriovenous access patency in patients with end-stage renal disease on hemodialysis: Presented at 2023 Vascular Annual Meeting of the Society for Vascular Surgery, National Harbor, Maryland, June 14,2023. Eur J Vasc Endovasc Surg 2024; 67: 1.

33.

Jankovic

Damjanovic

Djuric

, et al. Impact of vascular calcifications on arteriovenous fistula survival in hemodialysis patients: a five-year follow-up. Nephron 2015; 129: 247–252.

34.

Shin D

Rhee S

Jeon H

, et al. An increase in mean platelet volume/platelet count ratio is associated with vascular access failure in hemodialysis patients. PLoS One 2017; 12: e0170357.

35.

Wang

Yang

Zhao

. Retrospective analysis of predictive factors for AVF dysfunction in patients undergoing MHD. Medicine (Baltimore) 2024; 103: e37737–82.

36.

Kong

Tang

Liang

, et al. Clinical outcomes following the surgery of new autologous arteriovenous fistulas proximal to the failed ones in end-stage renal disease patients: a retrospective cohort study. Ren Fail 2019; 41: 1036–1044.

37.

Hemmelgarn

Manns

Quan

, et al. Adapting the Charlson Comorbidity Index for use in patients with ESRD. Am J Kidney Dis 2003; 42: 125–132.

38.

Fuhrman

Brook

Aviram

. Increased platelet aggregation during alimentary hyperlipemia in normal and hypertriglyceridemic subjects. Ann Nutr Metab 1986; 30: 250–260.

39.

Eitzman

Westrick

, et al. Hyperlipidemia promotes thrombosis after injury to atherosclerotic vessels in apolipoprotein E-deficient mice. Arterioscl Throm Vas 2000; 20: 1831–1834.

40.

Jiang

Lin

Chen

, et al. ACSL1 aggravates thromboinflammation by LPC/LPA metabolic axis in hyperlipidemia associated myocardial ischemia-reperfusion injury. Adv Sci 2025; 12: e2406359.

41.

Song

Yang

Kim

, et al. Association of plasma fibrinogen concentration with vascular access failure in hemodialysis patients. Nephrol Dial Transpl 1999; 14: 137–141.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB