摘要
目的 构建基于机器学习算法的神经外科静脉血栓栓塞症(venous thromboembolism, VTE)风险预测模型。方法 采用过采样技术对数据进行平衡,应用Boruta算法进行关键特征变量筛选,然后利用AdaBoost、GBDT、LightGBM、SVM四种机器学习方法构建预测模型,并利用准确率、精确度、召回率、F1分数进行模型性能评估,综合对比四个模型的性能,选取最佳模型,并进行模型变量重要性分析。结果 Boruta方法最终将C1(年龄)、C3(体质指数(BMI>25 kg/m^(2)))、C7(下肢肿胀)、C8(卧床的内科患者)、C10(严重肺病,包括肺炎(<1个月))、C16(其他危险因素)、C18(大型开放手术(>45 min))、C20(限制卧床(≥72 h))、C21(恶性肿瘤(现患或既往))、C31(DVT/PTE史)、C33(脑卒中(<1月))、C35(多发性创伤(<1个月))确定为关键特征变量。利用AdaBoost、GBDT、LightGBM、SVM四种机器学习方法构建预测模型,GBDT是表现最佳的模型,准确率是0.967,精确度是0.967,召回率0.954,F1分数是0.954。年龄、下肢肿胀、卧床的内科患者、体质指数(BMI>25 kg/m^(2))、大型开放手术(>45 min)是神经外科VTE最重要的风险因素。结论 基于机器学习和Caprini评估量表构建的风险预测模型具有较高的模型性能,能够有效缩减评估指标,提升VTE评估的准确性和效率。
Objective To construct a risk prediction model of VTE in neurosurgery based on machine learning algorithm.Methods The oversampling technology was used to balance the data,the Boruta algorithm to select the key feature variables,and then the four machine learning methods of AdaBoost,GBDT,LightGBM and SVM to construct the prediction model.And the accuracy,precision,recall and F1 score were used to evaluate the performance of the model.The best model was selected by comparing the performance of the four models,and the importance of the model variables was analyzed.Results Boruta’s method finally compared C1(age),C3(body mass index(BMI>25 kg/m^(2))),C7(lower limb swelling),C8(bedridden medical patients),C10(severe lung disease,Including pneumonia(<1 month),C16(other risk factors),C18(major open surgery(>45 minutes)),C20(limited bed rest(≥72 hours)),C21(malignancy(current or previous)),C31(history of DVT/PTE),C33(stroke(<1 month),C35(multiple trauma(<1 month)were identified as key characteristic variables.AdaBoost,GBDT,LightGBM and SVM are used to construct the prediction model,and GBDT is the best model with the precision of 0.967,the precision of 0.967,the recall of 0.954,and the F1 score of 0.954.Age,lower extremity swelling,medical patients in bed,body mass index(BMI>25 kg/m^(2)),and major open surgery(>45 minutes)were the most important risk factors for neurosurgical VTE.Conclusion The risk prediction model based on machine learning and Caprini assessment scale has high model performance,which can effectively reduce the assessment indicators and improve the accuracy and efficiency of VTE assessment.
作者
胡洋洋
王常娟
刘超
宋仪昕
HU Yang-yang;WANG Chang-juan;LIU Chao;SONG Yi-xin(Neurosurgery Department,The Second Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075100,China;Functional Department,The Second Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075100,China;Medical Department,The Second Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075100,China;Department of Medical Records Management,The Second Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075100,China)
出处
《河北北方学院学报(自然科学版)》
2025年第6期1-5,共5页
Journal of Hebei North University:Natural Science Edition
作者简介
胡洋洋(1988-),男,山东单县人,主治医师,研究方向:神经外科围手术期管理。
