基于多种机器学习算法的老年瓣膜性心脏病患者术后院内死亡风险因素分析

Establishment of an In-hospital Mortality Risk Model for Elderly Patients Undergoing Cardiac Valvular Surgery Based on Machine Learning

目的:基于多种机器学习算法建立老年(≥65岁)瓣膜性心脏病患者术后院内全因死亡风险的预测模型,为心脏瓣膜术后患者死亡风险评估提供新的思路。方法:回顾性连续纳入2016年1月至2018年12月中国心血管外科注册登记研究数据库(CCSR)中接受心脏瓣膜手术,年龄≥65岁的患者7163例。2016年1月到2018年6月的患者为训练队列(n=5774),2018年7月到12月患者为测试队列(n=1389)。研究终点为患者术后院内死亡。分析其临床资料,包括基本特征、围术期危险因素以及术后主要结局指标等。采用多种机器学习算法构建老年瓣膜性心脏病患者术后死亡风险预测模型。结果:290例(4.1%)患者术后院内死亡。与未死亡患者比,死亡患者年龄较大,既往脑卒中史、慢性心力衰竭史患者占比较大,吸烟史、高脂血症患者占比较少(P均<0.05)。训练队列中线形判别分析(LDA)、支持向量机分类器(SVC)及逻辑回归(LR)预测模型ROC曲线的AUC均较高,Brier分数均较低,具有较好的区分度及校准度。在测试队列中,LDA、SVC及LR预测模型ROC曲线的AUC分别为0.744、0.744及0.746,均优于新版欧洲心脏手术风险评分系统(EuroSCOREⅡ)模型的0.642(P均<0.05)。结论:老年患者心脏瓣膜术后死亡率较高,LDA、SVC、LR预测模型可以较好地预测老年患者心脏瓣膜术后死亡的发生率。

Objectives:To evaluate and predict the risk for in-hospital mortality in elderly patients receiving cardiac valvular surgery preferably,we developed a new prediction model using machine learning.Methods:Clinical data including baseline characteristics,peri-operative data and primary endpoint of 7163 elderly patients aged 65 years or older undergoing cardiac valvular surgery from January 2016 to December 2018 from 87 hospitals were collected from the Chinese Cardiac Surgery Registry(CCSR).Patients from January 2016 to June 2018 were assigened to the training cohort(n=5774)and patients from July to December 2018 were assigened to the validation cohort(n=1389).The primary endpoint was in-hospital mortality.Machine learning algorithms were used to analyze risk factors and develop prediction model.Results:Overall in-hospital mortality was 4.1%.Linear discriminant analysis(LDA),support vector classification(SVC)and logistic regression(LR)models in the training cohort all have high AUCs and low Brier scores,with good discrimination and calibration.In validation cohort,the AUC of LDA,SVC and LR were 0.744,0.744 and 0.746 respectively,which were significantly better than that of 0.642 using the European System for Cardiac Operative Risk Evaluation II(EuroSCORE II)model(P<0.05).Conclusions:The mortality rate for elderly patients undergoing cardiac valvular surgery is relatively high.LDA,SVC and LR can predict the risk for in-hospital mortality in elderly patients receiving cardiac valvular surgery with high accuracy.