Scholarlink

Introduction <p>Antithrombotic evaluation of postpartum venous thromboembolism (VTE) after anticoagulant therapy is challenging because of the lack of high-quality clinical evidence.</p> Aim <p>To identify and validate a machine learning model to predict thrombus regression in postpartum VTE patients.</p> Method <p>This study constructed three cohorts of postpartum VTE patients receiving low-molecular-weight heparin (LMWH): retrospective (n = 200), prospective (n = 50), and external (n = 29) cohorts. The main endpoint was imaging-based thrombus regression. Based on clinical guidelines and literature, we screened 16 potential predictors. The retrospective dataset was analyzed using eight machine learning models to predict thrombus regression after anticoagulant therapy. The performance of the model was evaluated by the area under the receiver operating characteristic curve (AUC-ROC), residual analysis, accuracy, sensitivity and specificity. The optimal model was selected based on its comprehensive performance and further validated on the prospective and external cohorts. Model interpretability was analyzed using variable importance and partial dependence plots.</p> Results <p>Thrombus regression was significantly associated with anti-Xa activity (<i>p</i> = <i>0.009</i>), antithrombin III levels (<i>p</i> = <i>0.028</i>), and D-dimer levels (<i>p</i> = <i>0.024</i>). Of the eight models<i>,</i> the random forest (RF) model demonstrated the best predictive performance, with the AUC-ROC value of 0.831(95%CI 0.696–0.967), accuracy of 0.77, sensitivity of 0.45, specificity of 0.89, and the highest accuracy in predicting thrombus regression events in the prospective (94.29%) and external (90.00%) postpartum VTE datasets. The key predictive variables were anti-Xa activity, antithrombin III levels, D-dimer levels, and body mass index (BMI).</p> Conclusion <p>Based on data from the three cohorts of patients with postpartum VTE, the RF model was identified as the optimal model for predicting thrombus regression events, with anti-Xa activity, antithrombin III levels, D-dimer levels, and BMI serving as key predictors. This study may help assess changes in the thrombotic state of postpartum VTE patients and guide clinical precision medication.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel predictive model for evaluating thrombus regression in postpartum venous thromboembolism patients

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月7日

查看全文
  • Meixiang Yu,
  • Congcong Jia,
  • Zhenyu Zhou,
  • Qingshan Chen,
  • Xuemei Di,
  • Ning Zhang,
  • Xiaoqing Zhang,
  • Zhendong Xu,
  • Hai Zhang

摘要

Introduction

Antithrombotic evaluation of postpartum venous thromboembolism (VTE) after anticoagulant therapy is challenging because of the lack of high-quality clinical evidence.

Aim

To identify and validate a machine learning model to predict thrombus regression in postpartum VTE patients.

Method

This study constructed three cohorts of postpartum VTE patients receiving low-molecular-weight heparin (LMWH): retrospective (n = 200), prospective (n = 50), and external (n = 29) cohorts. The main endpoint was imaging-based thrombus regression. Based on clinical guidelines and literature, we screened 16 potential predictors. The retrospective dataset was analyzed using eight machine learning models to predict thrombus regression after anticoagulant therapy. The performance of the model was evaluated by the area under the receiver operating characteristic curve (AUC-ROC), residual analysis, accuracy, sensitivity and specificity. The optimal model was selected based on its comprehensive performance and further validated on the prospective and external cohorts. Model interpretability was analyzed using variable importance and partial dependence plots.

Results

Thrombus regression was significantly associated with anti-Xa activity (p = 0.009), antithrombin III levels (p = 0.028), and D-dimer levels (p = 0.024). Of the eight models, the random forest (RF) model demonstrated the best predictive performance, with the AUC-ROC value of 0.831(95%CI 0.696–0.967), accuracy of 0.77, sensitivity of 0.45, specificity of 0.89, and the highest accuracy in predicting thrombus regression events in the prospective (94.29%) and external (90.00%) postpartum VTE datasets. The key predictive variables were anti-Xa activity, antithrombin III levels, D-dimer levels, and body mass index (BMI).

Conclusion

Based on data from the three cohorts of patients with postpartum VTE, the RF model was identified as the optimal model for predicting thrombus regression events, with anti-Xa activity, antithrombin III levels, D-dimer levels, and BMI serving as key predictors. This study may help assess changes in the thrombotic state of postpartum VTE patients and guide clinical precision medication.