纵向研究缺失数据多重填补及混合效应模型分析  

Using multiple imputation and mixed-effects model on missing data: a longitudinal study

作　　者：申宁宁[1] 房瑞玲 高宇钊 李少琼[1] 张军锋[1] 刘桂芬[1] 

机构地区：[1]山西医科大学公共卫生学院卫生统计教研室,太原030001

出　　处：《中国药物与临床》2015年第7期901-905,共5页Chinese Remedies & Clinics

基　　金：国家自然科学基金(81172774)

摘　　要：目的阐明马尔可夫链蒙特卡罗(MCMC)多重填补与重复测量资料混合效应线性模型分析的原理,完成纵向监测数据缺失模型的软件实现。方法根据222例高血压患者纵向监测的完全数据,产生缺失比例为18.92%的随机缺失数据集。应用MCMC多重填补方法,进行缺失值填补的模拟研究以及实例分析,并实现重复测量混合效应线性模型分析。结果模拟研究和实例分析表明,样本例数200,缺失比例20%,MCMC法多重填补5次所得结果最稳健;填补前缺失数据与完全数据的混合效应模型分析结果不同,填补后完整数据与完全数据的混合效应模型分析结果相同。结论 MCMC多重填补可以充分利用缺失资料信息,是处理缺失数据模型分析的有效方法之一;针对出现缺失的重复测量资料,结合应用混合效应模型与MCMC多重填补2种方法,从而得出更为符合客观实际的结果。Objective To investigate the mechanisms of Markov Chain Monte Carlo（MCMC） multiple imputa-tion and mixed-effects linear model analysis of repeated measurement data, and to achieve software implementation of longitudinal monitoring on missing data model. Methods Depending on 222 longitudinal monitored complete data of patients with hypertension, a ratio of 18.92% random missing data set was generated. MCMC multiple imputation was used for simulation study and case study on multiple imputation of the missing value, and to achieve mixed-effects linear model analysis of repeated measurement data. Results The simulation study and case study showed that the findings of five-times MCMC multiple imputation on 200 samples with 20% missing data were the most reliable. The findings of mixed-effects model analysis between the missing data and the complete data were different before imputa-tion, which were same after imputation. Conclusion MCMC multiple imputation can take full advantage of the information of missing data, which is one of the effective methods to deal with model analysis of missing data. For missing repeated measurement data, more objective findings can be obtained by using mixed-effects model and MCMC multiple imputation.

关 键 词：纵向研究 缺失数据 高血压 重复测量 

分 类 号：R544.1[医药卫生—心血管疾病]