摘要
目的:探讨多种机器学习预测模型对脑卒中发病风险的评估效果。方法:选取2013年1月1日—2017年12月31日参与“北京健康管理队列”的体检人群作为研究对象,基线人群共计56017例。比较研究对象脑卒中发病与未发病人群之间基本人口学信息、代谢异常相关指标的差异,选用经典决策树模型、多层感知器模型、卷积神经网络模型开展模型构建,并与多因素logistic回归分析模型进行比较。结果:各模型分析结果均显示年龄、收缩压、腰围、身体质量指数为脑卒中发病的影响因素;多因素logistic回归分析模型、经典决策树模型、多层感知器模型、卷积神经网络模型的准确率分别为0.978、0.985、0.988、0.996。结论:代谢异常指标中的腹型肥胖、血压升高、低密度脂蛋白胆固醇降低、血糖升高均是脑卒中发病的潜在危险因素;经典决策树模型、多层感知器模型、卷积神经网络模型3种机器学习模型较多因素logistic回归分析模型预测性能更优,其中卷积神经网络模型的准确率最为良好,多层感知器模型的特异度最为良好。
OBJECTIVE To investigate the effect of multiple machine learning prediction models on stroke risk assessment.METHODS From January 1,2013 to December 31,2013,the Beijing Health Management cohort were selected as the study subjects,with 56017 cases in the baseline population.To compare the differences in basic demographic information and metabolic abnormalities between stroke onset and absence,the classical decision tree,multi-layer perceptron and convolutional neural network were used for model construction,and compared with the multifactor logistic regression analysis model.RESULTS The results of all models showed that age,SBP,waist circumference and BMI were risk factors for stroke.The accuracy rates of multi-factor logistic model,decision tree model,multi-layer perceptron model and convolutional neural network model were 0.978,0.985,0.988 and 0.996,respectively.CONCLUSION Abdominal obesity,elevated blood pressure,low density lipoprotein cholesterol,and elevated blood glucose in metabolism-related components are all potential risk factors for stroke.Compared with logistic regression model,the prediction performance of the three machine learning models is better,among which the convolutional neural network model has the best accuracy and the multi-layer perceptron model has the best specificity.
作者
于淼
刘康
徐鑫鹏
罗艳侠
YU Miao;LIU Kang;XU Xin-peng;LUO Yan-xia(School of Publie Health,Capital Medical University,Beijing,100069,China;不详)
出处
《中国初级卫生保健》
2023年第3期25-28,共4页
Chinese Primary Health Care
基金
国家自然科学基金面上项目(81773512)。
关键词
脑卒中
代谢指标
机器学习
预测模型
影响因素
stroke
metabolie index
mach ine leaming
foreast mooel
infuence factor
