基于巢式病例对照研究构建椎体压缩骨折患者术前深静脉血栓形成的风险预测模型  

Construction of a risk prediction model for preoperative deep vein thrombosis in patients with vertebral compression fractures based on nested case-control studies

作　　者：刘录 司海朋 李春林 任丽 朱玉娇 贺茜 LIU Lu;SI Hai-peng;LI Chun-lin;REN Li;ZHU Yu-jiao;HE Qian(Department of Spine Surgery,Qilu Hospital of Shandong University(Qingdao),Qingdao,Shandong 266000,China)

机构地区：[1]山东大学齐鲁医院(青岛)脊柱外科,山东青岛266000

出　　处：《中华骨质疏松和骨矿盐疾病杂志》2024年第6期580-586,共7页Chinese Journal Of Osteoporosis And Bone Mineral Research

摘　　要：目的分析椎体压缩骨折患者术前深静脉血栓形成(deep vein thrombosis,DVT)的危险因素,构建DVT诊断模型。方法选择2020年1月至2023年10月山东大学齐鲁医院(青岛)脊柱外科收治的骨质疏松性椎体压缩骨折患者作为研究对象,以入院确诊骨质疏松性椎体压缩骨折为起点、出院为终点,进行巢式病例对照研究。采用极端梯度提升算法XGBoost对DVT诊断模型进行训练,采用准确率和受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under the curve,AUC)评价模型性能。结果基于XGBoost训练DVT诊断模型,可视化前20个重要的特征,可以看出凝血酶原时间、甲状旁腺素这两个指标对于DVT的诊断较为重要。在测试集上评估模型,能够取得90.48%的诊断准确率,采用ROC曲线计算DVT诊断模型的AUC为0.9558,差异有统计学意义(P<0.05)。结论基于XGBoost训练的DVT诊断模型用于筛查术前高风险DVT患者有较好的性能及较好的泛化能力,通过对可引起椎体压缩骨折患者术前DVT的前20个危险因素的可视化,方便临床对DVT高危患者及时识别并给予相应的干预措施,避免延误手术。Objective To analyze the risk factors for preoperative deep vein thrombosis(DVT)in patients with vertebral compression fractures and to construct a DVT diagnostic model.Methods Patients with osteoporotic vertebral compression fracture admitted to the Department of Spine Surgery,Qilu Hospital of Shandong University(Qingdao)from January 2020 to October 2023 were selected as the research subjects.The nested case-control study was conducted with admission and diagnosis of osteoporotic vertebral compression fracture as the starting point and discharge as the end point.The extreme gradient lifting algorithm XGBoost was used to train the DVT diagnosis model,and the accuracy and area under receiver operating characteristic(ROC)curve(AUC)were used to evaluate the model performance.Results Based on XGBoost training DVT diagnostic model and visualization of the top 20 important features,it could be seen that the two indicators of prothrombin time parathyroid hormone are more important for the diagnosis of DVT.The diagnostic accuracy of 90.48%could be achieved by evaluating the model on the test set.The AUC of the DVT diagnostic model calculated by ROC curve was 0.9558,and the difference was statistically significant(P<0.05).Conclusion The DVT diagnostic model based on XGBoost training has good performance and generalization ability for screening high-risk DVT patients before surgery.By visualizing the top 20 risk factors that can cause deep vein thrombosis in patients with vertebral compression fracture before surgery,this study facilitates clinical identification of high-risk DVT patients in time and gives corresponding intervention measures to avoid prolonged surgery waiting time.

关 键 词：椎体压缩性骨折 深静脉血栓形成 机器学习算法 巢式病例对照研究 

分 类 号：R683.2[医药卫生—骨科学]