基于机器学习的可解释模型在急性A型主动脉夹层合并冠状动脉灌注不良术后主要不良心血管事件的预测研究  

Prediction of major adverse cardiovascular events after acute type A aortic dissection combined with coronary malperfusion by machine learning-based interpretable models

作　　者：张浩 贾博 张祚 乔环宇[2] 杨波[2] 杨璟[1] 黑飞龙 侯晓彤[3] 朱俊明[2] 刘永民[2] Zhang Hao;Jia Bo;Zhang Zuo;Qiao Huanyu;Yang Bo;Yang Jing;Hei Feilong;Hou Xiaotong;Zhu Junming;Liu Yongmin(Department of Cardiopulmonary Bypass and Mechanical Circulation Assistance,Beijing Anzhen Hospital,Capital Medical University,Beijing 100029,China;Department of Cardiac Surgery,Beijing Anzhen Hospital,Capital Medical University,Beijing 100029,China;Cardiac Surgery Critical Care Center,Beijing Anzhen Hospital,Capital Medical University,Beijing 100029,China)

机构地区：[1]首都医科大学附属北京安贞医院体外循环与机械循环辅助科,北京100029 [2]首都医科大学附属北京安贞医院心脏外科中心,北京100029 [3]首都医科大学附属北京安贞医院心脏外科危重症中心,北京100029

出　　处：《中华胸心血管外科杂志》2025年第3期129-135,共7页Chinese Journal of Thoracic and Cardiovascular Surgery

摘　　要：目的探索急性A型主动脉夹层(acute type A aortic dissection,ATAAD)患者术后主要不良心血管事件(major adverse cardiovascular events,MACEs)的危险因素并构建模型。方法回顾性分析2018年1月至2022年10月,就诊于北京安贞医院并接受外科手术治疗ATAAD患者的临床资料。以MACEs为终点,利用重采样随机将患者分为训练集(70%)和验证集(30%)。训练集应用LASSO回归探寻关键临床变量。依据曲线下面积,从9种机器学习算法中选择最优预测模型。采用绝对收缩和选择算子技术LASSO阐明预测模型。结果共纳入481例患者,135例(35.6%)发生终点事件。综合训练集和验证集的结果,评估对结果预测准确性最高的单一模型,显示logistics模型(0.774,95%CI:0.717~0.830)综合效果最好,准确度较高(0.743,95%CI:0.720~0.766)。根据Shapley Additive explanations的结果,与术后MACEs最相关的因素是脑血管疾病史、冠状动脉受累、入手术室休克状态、纤维蛋白原降解产物、血小板计数、体外循环、升主动脉阻断、年龄。结论9种机器学习模型预测ATAAD术后MACEs的发生,logistic模型的表现更好。ObjectiveTo explore and model risk factors in patients with major adverse cardiovascular events(MACEs)after acute type A aortic dissection(ATAAD),and to develop and validate a personalized machine learning model to assess risk factors and predict MACEs in these patients.MethodsClinical data of patients who attended Beijing Anzhen Hospital and underwent surgical treatment for ATAAD from January 2018 to October 2022 were retrospectively analyzed.Using MACEs as the endpoint,70%of these patients were randomly divided into the training set and the remaining 30%into the validation set.LASSO regression was applied to explore key clinical variables in the training set.The optimal predictive model was selected from nine machine learning algorithms based on area under the curve.And Shapley Additive explanations was used to elucidate the predictive model.ResultsOf the 481 patients included in this study,135(35.6%)patients experienced an endpoint event.By combining the results of the training and validation sets,when assessing the validity of the single model with the highest predictive accuracy for the outcome,it was shown that the logistic model(0.774,95%CI:0.717-0.830)was the most effective in the combined effect and had a high model accuracy(0.743,95%CI:0.720-0.766).According to the results of the LASSO,the factors most associated with postoperative MACEs were history of cerebrovascular disease,coronary artery involvement,shock status on admission to the operating room,FDP,PLT,CPB,ascending aortic clamping,and age.ConclusionIn this study,nine machine learning models were developed to predict the occurrence of postoperative MACEs in patients with acute type A aortic dissection.The logistic model performed significantly better compared to other algorithms.Our study successfully predicted postoperative MACES and identified the factors most associated with MACEs.

关 键 词：急性A型主动脉夹层 MACEs 心肌保护 预测模型 机器学习 

分 类 号：R654.2[医药卫生—外科学]