Multiple Imputation And Mixed-effects Model Applied In Longitudinal Data With Missing Data

Objective:In the National Community hypertension standardized management projects,according to the requirements of the community management of hypertension level,gather baseline information recorded every three months of follow-up, management to monitor follow-up four times during the year, after the implementation of a standardized management for the hypertensive patients with primary and secondary management,to obtain a longitudinal study of hypertension standardized management information. Due to the time required to obtain the data representing a cross-sectional study of a long and complex factors, resulting in less cooperation of the subjects, mobility or a change of residence, etc., there will be missing data inevitably. If we delete the missing data blindly, only using the complete data to analysis, not only will lose some of the information hidden in the original data, but may lead to biased parameter estimation, and even it is contrary to the objective facts coming to the fallacy. In order to utilize the monitoring data in the absence of longitudinal data contains information fully, provide a solution to resolve Longitudinal data with missing data.Methods:the paper will state mainly Markov Chain Monte Carlo(MCMC) and repeat the measurement principle to fill multiple mixed-effects model analysis of information; and will repeat measures mixed effects linear model with multiple MCMC the combination of these two methods to fill and complete lack of longitudinal data model monitoring software.Results:According to the National Community Hypertension Management Program unified standardized inclusion criteria,the datas were randomly selected from the community project management of hypertension, hypertensive patients with two copies of the information is completely data 222,and 222 cases of incomplete data based on longitudinal monitoring of patients with hypertension, resulting in missing data ratio of18.92% of the random missing data sets were verified instances.Simulation study and Case study shows,when the number of samples is 222 and lack of proportion is18.92%:1、when multiple times is five,the results of MCMC method is most robust;2、mixed-effects model for data with missing is low utilization of information, analysis results of data with missing and complete data are slightly different;3 、useing mixed effects models analysis data that was imputated by MCMC and complete data are consistent.Conclusions:(MCMC) simulation results show that Sample size is fixed, with the lack of increase in the proportion, to fill the number increases, missing proportionan is certain,with the increase the sample size, the need to gradually reduce the number of fill.Therefore, the content of different samples with different proportions to fill the missing number is distinct. Mixed effects models with missing data simulation results show that when sample size is certain,with the increase in the proportion of missing,the utilization of Missing data in units of observation information is worse. Through simulation studies with different proportions in different samples of the missing amount,combining with examples of analysis in hypertensive patients,further proves(MCMC) multiple imputation is widely used in missing data and it is insufficient to use mixed effects models alone in missing data analysis.MCMC can take advantage of multiple imputation missing datas and information,which is one of the effective methods for handling missing data model analysis; for repeated measurements with missing data, combined theapplication of linear mixed effects model repeated measures with MCMC multiple imputation method to derive more objective reality of the results.