多重填补法在任意缺失随访资料中的应用

Multiple Imputation Method Used in Arbitrary Missing Follow-up Data

目的比较任意缺失模式下不同填补方法在随访资料缺失数据中的多重填补效果。方法结合我国外周动脉疾病患者踝臂指数(ankle brachial index,ABI)等基线及六年随访数据,通过SAS9.3/MI过程,分别采用马尔可夫链蒙特卡罗(markov chain monte carlo,MCMC)、回归分析、判别分析(discriminant analysis)和logistic回归等方法,实现生存时间、生存结局变量缺失值的填补,并作综合分析及比较。结果得到不同填补方法、不同填补次数多重填补后的生存时间和结局变量完全数据集,并对总体参数作出估计和统计,计算各次填补效率等综合评价指标。结论对于多次随访资料中的连续性变量生存时间,采用回归分析方法填补效率较高,填补效率随着填补次数增加而增大,对于缺失率小的变量填补效率更高。

Objective To evaluate the multiple imputation effect of different imputation methods in arbitrary missing data of follow-up research. Methods Using different methods including Markov chain Monte Carlo（ MCMC）, Regression, dis- criminant analysis and logistic regression and SAS9. 3/MI process ,to make the comprehensive analysis and comparison for miss- ing values imputation. The real data come from a 6 years follow -up research including peripheral arterial disease patients＇ infor- mation and ankle brachial index （ABI） data. Results Including population parameters estimation and statistics inference of con- tinuous variables, frequency calculation of classified variables, based on different imputation methods and imputation numbers. Conclusion In the continuous variables such as survival time, Regression method has the largest imputation efficiency, and the efficiency increases with the increase of imputation number and decrease of the missing rate.