基于贝叶斯网络探讨社区老年人轻度认知障碍的影响因素研究

Study on influencing factors of mild cognitive impairment in community-dwelling elderly individuals based on Bayesian network

中文关键词: 贝叶斯网络模型  社区老年人  轻度认知功能障碍  影响因素  随机森林

英文关键词: Bayesian network model  community-dwelling elderly individual  mild cognitive impairment  influencing factors  random forest

中文摘要:

      目的 分析社区老年人轻度认知障碍(MCI)的影响因素及其网络关系.方法 利用上海老年脑健康队列研究中徐汇区社区居民的基线数据,纳入其人口学因素、生活习惯、闲暇及社交活动、病史体检资料、实验室检查、认知功能、心理状况、日常生活活动能力评定方面等资料.采用随机森林模型对 1 043 名参与者进行初筛,使用爬山算法建立贝叶斯网络(BN),并根据极大似然估计进行参数估算.结果 经随机森林模型筛选后,将甘油三酯-总胆固醇-体重指数(TCBI)、慢性病数量、体重指数、载脂蛋白 B、载脂蛋白 A1、文化程度、空腹血糖、低密度脂蛋白胆固醇、脂蛋白 a(Lpa)、肌酐、饮食习惯、是否经常使用电脑手机上网12 个变量纳入BN模型.以筛选出的12 个变量作为网络节点,构建了一个含有13 个节点和20 条有向边的社区老年人 MCI 相关因素 BN 模型.TCBI、Lpa、慢性病数量和是否经常使用电脑手机上网与 MCI 直接相关.结论 BN模型揭示了社区老年人 MCI 与代谢指标、生活方式及慢性病负担之间的复杂关联模式及局部路径结构.

英文摘要:

      Objective To analyze the influencing factors of mild cognitive impairment(MCI)and their network relationships in community-dwelling elderly individuals.Methods Baseline data from community-dwelling elderly individuals in Xuhui District,Shanghai,were extracted from the Shanghai Aging Brain Health Cohort,including demographics,lifestyle habits,leisure and social activities,medical history and physical examination data,laboratory test results,cognitive function,psychological status,and assessments of activities of daily living.A random forest model was used to preliminarily screen the 1 043 participants.A Bayesian network(BN)was established using the hill-climbing algorithm,and parameters were estimated by the maximum likelihood method.Results After screening with the random forest model,12 variables were included in the BN model:triglyceride-total cholesterol-body mass index(TCBI),number of chronic diseases,body mass index,apolipoprotein B,apolipoprotein A1,education level,fasting blood glucose,low-density lipoprotein cholesterol,lipoprotein(a)[Lp(a)],creatinine,dietary habits,andfrequent use of computers/mobile phones for internet access.Using the 12 selected variables as network nodes,a BN model for factors related to MCI in the community-dwelling elderly individuals was constructed,containing 13 nodes and 20 directed edges.TCBI,Lp(a),number of chronic diseases,and frequent use of computers/mobile phones for internet access were directly associated with MCI.Conclusion The BN model reveals the complex association patterns and local path structures between MCI and metabolic indicators,lifestyle factors,and chronic disease burden in community-dwelling elderly individuals.