高危住院患者深静脉血栓风险管理效果

Risk Management Effect of Deep Vein Thrombosis(DVT)Among High-Risk Hospitalized Patients

目的本研究基于临床电子病历与临床辅助决策系统构建机器学习模型(AI模型),并对机器学习模型与人工Caprini量表对深静脉血栓形成(Deep vein thrombosis, DVT)风险的预测效果进行对比。方法回顾性收集与模型建立一致的住院患者的临床变量、金标准诊断结果以及Caprini评分,比较AI模型与Caprini的受试者特征曲线下面积(AUC)、敏感性与特异性。结果 AI模型对DVT的预测效果(AUC=0.894,95%CI:0.838,0.940)显著优于Caprini量表(AUC=0.716,95%CI:0.626,0.801),P<0.001。机器学习模型的敏感性、特异性也高于Caprini评分。结论机器学习模型预警工具对高危住院患者DVT的预测效果较人工Caprini评分具有明显的优势,或有助于DVT的风险管理,值得在临床上推广应用。

Objective To develop a machine learning model(AI model)based on Clinical Decision Support System(CDSS)and electronic medical records,and to validate its performance in the same hospital by comparing it to Caprini score.Methods This retrospective cohort study included hospitalized patients from Shanghai 10 th People’s Hospital and collected variables used when establishing AI model.The golden standard of DVT diagnosis as well as Caprini score evaluated by health professionals were also recorded,in order to calculate AU-ROC,sensitivity and specificity of AI model and Caprini.Results AUC of AI model(0.894,95%CI:0.838,0.940)was significantly better than that of Caprini(0.716,95%CI:0.626,0.801).Meanwhile,sensitivity and specificity of AI model were also higher than those of Caprini Score.Conclusion Machine learning model,running automatically in CDSS were superior to Caprini score in terms of discriminating high-risk hospitalized patients from others,which may help DVT risk management.