中国实用外科杂志

中国实用外科杂志

• 论蓍 • 上一篇    下一篇

腹股沟疝手术病人静脉血栓栓塞症风险评估决策树模型建立与测试

张    妍1，刘雨辰2，王明刚2，车    焱1   

  1. 1复旦大学生殖与发育研究院上海市计划生育科学研究所   国家卫生健康委员会计划生育药具重点实验室，上海 200032；2首都医科大学附属北京朝阳医院，北京  100020
  • 出版日期:2021-02-01 发布日期:2021-02-05

  • Online:2021-02-01 Published:2021-02-05

摘要： 目的    建立腹股沟疝病人术后静脉血栓栓塞症（VTE）决策树风险评估模型并评价其预测效果。方法 基于住院病历，收集全国58家医院2017年全年14322例成人腹股沟疝手术病人住院期间VTE诊断及手术过程和有关影响因素。用合成少数类过采样技术对数据预处理后，将样本分为训练数据集和验证数据集，训练数据集用于建立决策树风险评估模型，验证数据集用于模型评价。结果    经过数据训练，建立了含血栓相关疾病史、术后切口压迫、病人年龄、体重指数（BMI）、手术时间及术后出血共6个变量在内的腹股沟疝手术病人VTE决策树风险评估模型，其中血栓相关疾病史、术后切口压迫是该模型的关键因素。验证数据集测试结果表明，该模型AUC为0.870（95%CI 0.856-0.885），显著高于Caprini风险评估模型的0.739（95%CI 0.723-0.755）；用原始调查数据验证，两种模型的AUC差异无统计学意义。结论    腹股沟疝病人术后VTE决策树风险评估模型性能不劣于Caprini模型，但所含变量更少，方便应用。然而，由于建模数据的局限性，其性能需要更多临床资料加以验证和改进。

关键词: 静脉血栓栓塞症, 决策树, 风险评估模型, 腹股沟疝, 手术

Abstract: Establishment and validation of a decision tree model for venous thromboembolish risk prediction among inguinal hernia repair patients        ZHANG Yan*, LIU Yu-chen, WANG Ming-gang, et al. *Key Laboratory of Reproduction Regulation of NPFPC, SIPPR, IRD, Fudan University, Shanghai 200032, China
Corresponding authors: CHE Yan, E-mail: yan.che@sippr.org.cn; WANG Ming-gang, E-mail:wmgonly@126.com
ZHANG Yan and LIU Yu-chen are the first authors who contributed equally to the article.
Abstract    Objective    To establish and test a decision tree-based risk stratified venous thromboembolism (VTE) prediction model for patients undergoing inguinal hernia surgery. Methods    A hospital-based survey was conducted at 58 hospitals across China. A total of 14 322 adult inguinal hernia patients admitted to the hospitals for surgeries in 2017 were included. Borderline synthetic minority oversampling technique (Borderline-SMOTE) was used to oversample the data. The dataset was divided into training and testing datasets at a ratio of 3:1. A logistic model-based decision tree model was established using ten-fold cross-validation and recursive partition algorithm. Results    The decision tree showed that VTE-related history and post-operative incision compression were important predictors and would change the effects of age, body mass index, operation duration, and postoperative bleeding on VTE incidence. The AUC of the decision tree model was 0.870 (95%CI 0.856-0.885), which was significantly higher than the 0.739 (95%CI 0.723-0.755) of the Caprini model. The difference of the AUCs between the model-test dataset and the original dataset was not statistically significant. Conclusion    The performance of decision tree model in VTE prediction is not inferior to that of Caprini score model, but decision tree model is simpler and easier to use. Nevertheless, further research is needed to validate the model due to data limitations.

Key words: venous thromboembolism, decision tree, risk prediction model, inguinal hernia, surgery