急性A型主动脉夹层发生严重不良事件的随机森林模型建立与验证

Establishment and verification of a random forest model of serious adverse events in acute type A aortic dissection

目的:采用随机森林模型评价急性的Stanford A型主动脉夹层(ATAAD)发生严重不良事件的危险因素。方法:收集武汉大学中南医院2013年1月—2020年12月的179例ATAAD患者,以是否在确诊后30 d内发生死亡、主动脉破裂或即将破裂导致紧急手术分为两组,以随机森林建立模型对纳入的病例进行计算,使用Boruta算法对随机森林模型进行特征可视化,最后用传统的Logistic回归模型对研究结果进行对比验证。结果:随机森林建立了一个敏感度、特异度、AUC分别为0.783、1.000、0.891的高效诊断模型,模型的重要特征为Penn分级、收缩压、心包积血、主动脉周围血肿、C-反应蛋白(CRP)水平、升主动脉直径;传统Logistic回归模型显示Penn分级、收缩压、主动脉周围血肿、CRP水平、升主动脉直径为不良事件的独立预测因素,但是其诊断效能较随机森林模型低,该传统模型的敏感度、特异度、AUC分别为0.674、0.895、0.844。结论:随机森林算法能够建立一个高效的模型对ATAAD确诊后30 d内的严重不良事件进行预测,可为外科个性化治疗方案提供参考。

Objective:To evaluate the risk factors of serious adverse events in acute Stanford type A aortic dissection(ATAAD)using a random forest model.Methods:A total of 179 ATAAD patients were collected in the Zhongnan Hospital of Wuhan University from January 2013 to December 2020,and they were divided into two groups according to the occurrence of death or emergency surgery for aortic rupture or impending rupture within 30 days after diagnosis,and a random forest model was used to establish the model.The included cases were calculated,the Boruta algorithm was used to visualize the characteristics of the random forest model,and finally,the research results were compared and verified with the traditional logistic regression model.Results:We established a high-efficiency diag-nostic model of Random Forest with sensitivity,specificity,and AUC of 0.783,1.000,and 0.891,respectively.Penn classification,ascending aorta diameter,systolic blood pressure,pericardial hem-orrhage,C-reactive protein(CRP)level,and peri-aortic hematoma were the important features of the model.The traditional logistic regression model showed that Penn classification,ascending aorta di-ameter,systolic blood pressure and CRP level,and peri-aortic hematoma were independent predictors of adverse events.However,the sensitivity,specificity,and AUC of the traditional model were 0.674,0.895,and 0.844,respectively,and its diagnostic power was lower than that of the random forest model.Conclusion:The random forest algorithm can establish an efficient model to predict seri-ous adverse events within 30 days after the diagnosis of ATAAD,which can provide a reference for personalized surgical treatment plans.