基于机器学习的出血性脑卒中下肢深静脉血栓风险预警模型建立与评价

Development and evaluation of a machine learning-based model for predicting deep vein thrombosis risk in lower limbs following hemorrhagic stroke

目的利用机器学习算法构建出血性脑卒中下肢深静脉血栓风险预警模型,为临床管理决策提供参考。方法选取2019年1月至2023年12月遵义医科大学附属医院351例急性出血性脑卒中住院患者为研究对象,将数据集分为DVT组(n=155)和非DVT组(n=196),通过医院电子病历系统收集风险预测的特征指标,使用最低绝对收缩和选择算子(LASSO)回归进行特征降维筛选,利用6种机器学习算法来构造模型,其中包括线性回归模型(LR)、支持向量机分类器(SVC)、随机森林分类器(RFC)、弹性网络模型(EN)、梯度提升分类器(GBC)和多层感知器分类器(MLPC),使用SHAP对特征属性进行数值表达。结果LASSO回归在训练集中筛选出10个预测因子用于构建模型,包括卧床时间≥72 h、急性感染、红细胞比容、左侧下肢肌力、输血、住院天数、D-二聚体、年龄、血清肌酐、右侧下肢肌力。随机森林算法表现最为良好,构建的模型曲线下面积(AUC)为0.95,SHAP排名前五的特征分别为卧床时间≥72 h、急性感染、D-二聚体、年龄、血清肌酐。结论基于机器学习的出血性脑卒中下肢深静脉血栓风险预警模型建立能有效识别DVT发生风险,为临床预防和干预提供参考。

Objective To develop a predictive model utilizing machine learning algorithms to forecast the risk of lower extremity deep vein thrombosis(DVT)following hemorrhagic stroke,thereby providing a reference for clinical management decisions.Methods A total of 351 hospitalized patients with acute hemorrhagic stroke patients hospitalized at the Affiliated Hospital of Zunyi Medical University between January 2019 and December 2023 were selected as the study population.The dataset was divided into DVT group(n=155)and non-DVT group(n=196).The prognostic features were collected via the hospital’s electronic health record system.Feature reduction was performed using the Least Absolute Shrinkage and Selection Operator(LASSO)regression.Six different machine learning algorithms were deployed to construct the models,including Linear Regression(LR),Support Vector Classification(SVC),Random Forest Classifier(RFC),Elastic Net(EN),Gradient Boosting Classifier(GBC),and Multilayer Perceptron MLPClassifier(MLPC).Model feature importance was numerically expressed using SHapley Additive exPlanations(SHAP).Results LASSO regression identified 10 predictive factors in the training set for model construction,including bed rest time≥72 hours,acute infection(within one month),hematocrit,left lower limb muscle strength,blood transfusion,hospital stay,D-dimer,age,serum creatinine,and right lower limb muscle strength.The model constructed by the random forest algorithm had an area under the curve(AUC)of 0.95,and the top five features ranked by SHAP were bed rest time≥72 hours,acute infection(within one month),D-dimer,age,and serum creatinine.Conclusion The establishment of a risk warning model for lower extremity deep vein thrombosis in hemorrhagic stroke based on machine learning can effectively identify the risk of DVT occurrence,providing a reference for clinical prevention and intervention.