通讯作者: 刘娇娇, tzjs85@163.com

DOI：10.12201/bmr.202510.00023

Construction of a Prediction Model for Lower Limb Deep Vein Thrombosis in Patients with Hemorrhagic Stroke Based on Machine Learning

• 摘要：目的:基于机器学习算法构建5种不同的出血性卒中患者下肢深静脉血栓风险预测模型,并比较模型效能,为评估及预防下肢深静脉血栓提供最佳依据。方法: 采用回顾性研究方法,选取2022年2月至2024年2月天津医科大学总医院神经外科重症监护病房(NICU)收治的709例出血性脑卒中患者的临床资料,基于机器学习算法,分别采用使用逻辑回归(LR)、支持向量机(SVM)、随机森林(RF)、决策树(DT)、人工神经网络(ANN)构建5种不同的出血性卒中患者下肢深静脉血栓风险预测模型,采用准确率、精确率、召回率、F1值、ROC曲线下面积(AUC)来评估模型准确度,校准度采用Hosmer-Lemeshow 检验。结果:随机森林算法构建预测模型的预测效能最佳为0.947,校准度检验P＞0.05。运用SHAP对模型作可解释分析:血浆D-二聚体浓度、年龄、左下肢肌力、体质量、镇静镇痛药物应用、医用冰毯机治疗、GCS为主要特征因子。结论:出血性卒中患者下肢DVT发生主要影响因素为:血浆D-二聚体浓度、年龄、左下肢肌力、体质量、镇静镇痛药物应用、医用冰毯机治疗、GCS,基于随机森林算法构建预测模型效能最佳。

  关键词： 机器学习; 出血性卒中; 深静脉血栓; 预测模型

   

  Abstract: Objective: To construct five distinct machine learning-based risk prediction models for deep venous thrombosis (DVT) in lower extremities among hemorrhagic stroke patients, evaluate their comparative performance, and establish optimal evidence for DVT assessment and prevention strategies.Methods: A retrospective analysis was performed utilizing clinical data from 709 patients who experienced hemorrhagic stroke and were admitted to the Neurosurgical Intensive Care Unit (NICU) at Tianjin Medical University General Hospital from February 2022 to February 2024. Five distinct risk prediction models for deep vein thrombosis (DVT) in the lower extremities of hemorrhagic stroke patients were developed through machine learning algorithms, namely Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Artificial Neural Network (ANN). The models accuracy was assessed through metrics including accuracy, precision, recall, F1 score, and the area under the ROC curve (AUC), while the models’ calibration was evaluated using the Hosmer-Lemeshow test.Results: The prediction model developed utilizing the Random Forest algorithm exhibited the most superior predictive performance, achieving an AUC of 0.947, while the calibration test indicated P > 0.05.SHAP was utilized for the interpretable analysis of the model, pinpointing plasma D-dimer concentration, age, muscle strength in the left lower limb, body mass, the use of sedative and analgesic medications, treatment with medical cooling blankets, and GCS as the primary feature factors.Conclusion: The primary factors affecting DVT in the lower extremities of hemorrhagic stroke patients include plasma D-dimer levels, age, muscle strength of the left lower limb, body mass, the administration of sedatives and analgesics, the use of medical cooling blankets, and the GCS score. The prediction model developed using the Random Forest algorithm exhibits the highest performance.

  Key words: Machine Learning; Hemorrhagic Stroke; Deep Vein Thrombosis; Prediction Model

  提交时间：2025-10-14

  版权声明：作者本人独立拥有该论文的版权，预印本系统仅拥有论文的永久保存权利。任何人未经允许不得重复使用。
• html

• 图表

• 李鹏飞, 路鑫, 徐峰. 创伤后深静脉血栓形成的诊治进展. 2024. doi: 10.12201/bmr.202410.00015

  孔祥耀, 吴泽涛. 二种不同腔内血栓清除方式治疗早期下肢深静脉血栓的疗效分析—一个单中心回顾性分析. 2025. doi: 10.12201/bmr.202506.00037

  牛亚丹. 外周血炎症指标对髋部骨折患者术后并发下肢深静脉血栓的诊断价值. 2024. doi: 10.12201/bmr.202408.00016

  成路平, 吴思扬, 路波. 基于机器学习的2型糖尿病发病因素分析与预测. 2025. doi: 10.12201/bmr.202509.00003

  唐诗诗, 李宇轩, 唐圣晟, 刘庆华, 周毅. 基于SARIMA-LSTM模型的肾综合征出血热发病率预测研究. 2024. doi: 10.12201/bmr.202407.00046

  兰雨姗, 郑思, 李姣. 机器学习方法在因果推断中混杂因素控制的应用. 2022. doi: 10.12201/bmr.202203.00015

  康宏宇, 徐晓巍, 郑思, 郝洁, 杨林, 王序文, 侯丽, 李姣. 新医科背景下R语言与机器学习产教研协同教学模式探索. 2025. doi: 10.12201/bmr.202509.00006

  王杰, 王至诚, 娄帅, 董建成, 曹新志. 基于深度学习算法Mask R-CNN的甲状腺结节检测模型研究. 2024. doi: 10.12201/bmr.202411.00085

  苏燕凤, 洪素茹, 陈钰霜, 吴夏阳. 结合机器学习模型的早产儿ICU死亡风险评估与可解释性分析. 2025. doi: 10.12201/bmr.202503.00066

  胡欣. 剖宫产术中寒战风险预测模型及列线图的构建. 2025. doi: 10.12201/bmr.202501.00053

• 序号	提交日期	编号	操作
  1	2025-08-27	bmr.202510.00023V1	下载

• 公开评论  匿名评论  仅发给作者 提交评论