【摘要】 目的 运用机器学习(ML)算法探讨胫骨平台骨折患者术后深静脉血栓(DVT)并发肺栓塞(PE)的危险因素，构建风险预测模型并验证其预测效能。方法 回顾性分析2024年1月至2025年6月我院收治的494例胫骨平台骨折患者的临床资料。按7∶3比例分为训练集(345例)和验证集(149例)。训练集数据用于模型构建，以DVT并发PE为因变量，分别通过随机森林(RF)、极限梯度提升(XGBoost),反向传播神经网络(BPNN)3种机器学习算法筛选特征变量，经过“重叠覆盖”分析取交集变量作为预测变量，根据预测变量构建预测模型。验证集数据用于模型验证。绘制受试者工作特征(ROC)曲线，校准曲线和临床决策曲线(DCA)评价该模型的预测效能和临床应用价值。结果 年龄、术前D-二聚体和术后D-二聚体是胫骨平台骨折患者术后DVT并发PE的独立危险因素。ROC曲线显示，训练集和验证集的曲线下面积(AUC)分别为0.918(95%CI 0.876～0.960)和0.967(95%CI 0.938～0.996)。校准曲线显示，训练集和验证集的Brier评分分别为0.077和0.048。DCA曲线显示训练集和验证集在阈值范围内均具有临床净收益。结论 基于机器学习方法构建的预测模型对预测胫骨平台骨折患者术后DVT并发PE的风险具有较好的区分度、一致性和临床净收益，值得临床推广。更多还原

【Abstract】 Objective To investigate risk factors for deep vein thrombosis(DVT) complicated with pulmonary embolism(PE) postoperatively in patients with tibial plateau fractures using machine learning(ML) algorithms, construct a risk prediction model, and validate its predictive efficacy.Methods From January 2024 to June 2025,a retrospective analysis was conducted on clinical data of 494 patients with tibial plateau fractures admitted to our hospital.Data were divided into a training set(345 cases) and a validation set(149 cases) in a 7∶3 ratio.The training set data were used for model construction, with DVT complicated by PE as the dependent variable.Three machine learning algorithms—random forest(RF),extreme gradient boosting(XGBoost),and backpropagation neural network(BPNN)—were employed to screen feature variables.Overlapping coverage analysis was performed to identify intersecting variables as predictors, and a predictive model was constructed based on these variables.The validation set data were used for model validation.Receiver operating characteristic(ROC) curves, calibration curves, and clinical decision curves(DCA) were plotted to evaluate the model’s predictive efficacy and clinical application value.Results Age, preoperative D-dimer, and postoperative D-dimer levels were identified as independent risk factors for DVT complicated by PE in tibial plateau fracture patients.The ROC curves showed that the area under the curve(AUC) for both training and validation sets was 0.918(95%CI 0.876～0.960) and 0.967(95%CI 0.938～0.996),respectively.The Brier scores for training and validation sets were 0.077 and 0.048,respectively.The DCA curve demonstrated clinical net benefit for both the training set and validation set within the threshold range.Conclusion The predictive model constructed based on machine learning methods exhibits good discrimination, consistency, and clinical net benefit in predicting the risk of postoperative DVT complicated by PE in patients with tibial plateau fractures, making it worthy of clinical promotion.更多还原