文献速递血液透析通路

成熟自体动静脉内瘘血栓形成预测模型

Predictive Model for Thrombosis in Mature Autologous Arteriovenous Fistula.
PubMed 文献 · Ann Vasc Surg · 2026-06-28

本研究基于「高风险狭窄-血栓形成」理论,回顾性分析521例维持性血液透析患者临床数据,按7:3随机分为训练集和验证集,构建逻辑回归和XGBoost预测模型并比较性能。结果显示近期内瘘功能障碍、透析中舒张压大幅波动、低血红蛋白、既往2次及以上AVF失败史和区域或扣眼穿刺法是AVF血栓形成的危险因素;验证集中逻辑回归模型AUC为0.955(95%CI:0.924-0.985),XGBoost模型AUC为0.954(95%CI:0.921-0.988),推荐逻辑回归模型作为基线筛查工具。

要点

  • 近期内瘘功能障碍、透析中舒张压大幅波动、低血红蛋白、既往AVF失败史和区域或扣眼穿刺法是AVF血栓形成危险因素。
  • 逻辑回归模型验证集AUC为0.955(95%CI:0.924-0.985),XGBoost模型AUC为0.954(95%CI:0.921-0.988)。
  • 推荐逻辑回归模型作为识别AVF血栓高危患者的基线筛查工具。

核心结论

  • 逻辑回归和XGBoost模型均显示出良好的预测性能,但基于临床适宜性原则,推荐逻辑回归模型作为AVF血栓高危患者的基线筛查工具。

研究设计与样本

  • 数据集按7:3比例随机分为训练集和验证集,使用训练队列构建逻辑回归和XGBoost预测模型,并在验证集中比较预测性能。

主要发现

  • 近期(过去一个月内)内瘘功能障碍、透析中舒张压大幅波动、低血红蛋白水平、既往2次及以上AVF失败史以及区域或扣眼穿刺法被确定为AVF血栓形成的危险因素。
  • 在训练集中,XGBoost模型在所有性能指标上优于逻辑回归模型。
  • 在验证集中,逻辑回归模型的AUC为0.955(95%CI:0.924-0.985),略高于XGBoost模型的AUC 0.954(95%CI:0.921-0.988)。

临床怎么用

  • 对于维持性血液透析患者,可依据近期内瘘功能障碍、透析中舒张压波动、血红蛋白水平、既往AVF失败史和穿刺方式等因素评估AVF血栓风险。
  • 逻辑回归模型可作为临床基线筛查工具,用于识别AVF血栓高危患者。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: Based on the "high-risk stenosis-thrombosis" theory, this study developed a risk prediction model for thrombosis in mature autogenous arteriovenous fistulas (AVFs). METHODS: Clinical data from 521 maintenance hemodialysis patients treated at the vascular access clinic of Central Hospital Affiliated to Shandong First Medical University between January 2023 and December 2023 were retrospectively analyzed. The dataset was randomly divided into a training set and a validation set at a ratio of 7:3. Logistic regression (LR) and extreme gradient boosting (XGBoost) prediction models were constructed using the training cohort, and their predictive performance was subsequently compared in the validation set. RESULTS: The results indicated that recent fistula dysfunction within the past month, large short-term fluctuations in diastolic blood pressure during dialysis, low hemoglobin levels, a history of 2 or more prior AVF failures, and regional or buttonhole cannulation were identified as risk factors for AVF thrombosis. The XGBoost model outperformed the LR model across all performance metrics in the training set. However, in the validation set, the LR model achieved a slightly higher area under the curve (AUC) of 0.955 (95% CI: 0.924-0.985) compared with the XGBoost model's AUC of 0.954 (95% CI: 0.921-0.988). CONCLUSION: Although both the LR and XGBoost models showed favorable predictive performance, the LR model is recommended as a baseline screening tool for identifying patients at high risk of AVF thrombosis, given the clinical principle of favoring an appropriate over a more complex model.

来源:Ann Vasc Surg,查看原文。版权归原作者所有。

引用

  1. Zhang Y, Liu F, Li X, et al. Predictive Model for Thrombosis in Mature Autologous Arteriovenous Fistula. Ann Vasc Surg. 2026. PMID: 42365896. doi:10.1016/j.avsg.2026.06.038

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。