基于多因素回归分析和机器学习算法的冠心病预测模型构建及比较

本文旨在构建基于临床电子病历数据的冠心病预测模型. 回顾性收集了2015年至2020年在宁波大学医学院附属医院住院期间, 接受选择性冠状动脉造影的患者的临床数据, 分别应用决策树、朴素贝叶斯和逻辑回归算法构建冠心病预测模型, 比较3种模型的预测性能. 共收集354例患者数据, 其中冠心病患者140例, 非冠心病患者214例, 根据逻辑回归、朴素贝叶斯、决策树算法构建的3种预测模型的准确性分别为70.6%、89.5%、90.7%; 曲线下面积分别为0.676、0.869、0.921. 所构建的3种预测模型均具备较好的冠心病预测能力, 具有用于冠心病筛查的潜在价值.

Construction and comparison of coronary heart disease prediction models based on multivariate regression analysis and machine learning algorithm

The current study was designed to construct coronary heart disease (CHD) prediction models based on clinical electronic medical record data. The clinical data was collected from the patients who underwent elective coronary angiography during hospitalization at the Affiliated Hospital of Medical School, Ningbo University from 2015 till 2020. Decision tree, naive Bayes, and logistic regression algorithms were applied to construct prediction models for coronary heart disease. The predictive performance of these three models was compared. A total of 354 patients, including 140 CHD patients and 214 non-CHD patients, were recruited. Using logistic regression, naive Bayes, and decision tree algorithms, the accuracies of the three prediction models were 70.6%, 89.5% and 90.7% respectively. The AUCs were 0.676, 0.869, 0.921, respectively. All the three prediction models had good predictive ability for CHD and may have screening potential for CHD.