假想策略下处理缺失数据的参照填补法  

Reference-based Imputation for Hypothetical Strategies to Handle Missing Data

作　　者：黄清浩 甘世林 仲子航 倪森淼 刘文 贺志强 尹健 王媛媛 耿睿 于浩 柏建岭 Huang Qinghao;Gan Shilin;Zhong Zihang(Department of Biostatistics,School of Public Health,Nanjing Medical University(211166),Nanjing)

机构地区：[1]南京医科大学公共卫生学院生物统计学系,211166

出　　处：《中国卫生统计》2023年第3期331-334,340,共5页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(82273738,81773554);江苏高校“青蓝工程”。

摘　　要：目的探讨参照填补法基于假想策略来处理缺失数据的统计学性能,为此策略下缺失数据的处理提供参考。方法通过SAS模拟产生不同缺失率、缺失机制的模拟数据,并采用三种参照填补法(J2R、CIR、CR)、LOCF和MMRM进行处理,比较基于五种方法处理缺失后疗效估计的偏倚、均方误差、一类错误率和检验效能,并应用于一个实际临床试验数据。结果当两组疗效无差异时,参照填补法的偏倚和均方误差较LOCF、MMRM更低,LOCF、MMRM对应的一类错误率相对较大,参照填补法对应的一类错误率更为保守。当两组疗效存在差异时,参照填补法的偏倚和均方误差较LOCF、MMRM均更大,检验效能更低,且估计的治疗效应最保守,其中以J2R最甚。当缺失率较低时(低于10%),参照填补法的检验效能与其他方法相近,当缺失率较高时(高于10%),参照填补法检验效能过低。结论参照填补法是一种较为保守的填补方法。当数据缺失率较低时,可用于假想策略下缺失数据的填补,当缺失率较高时,可作为主估计目标的敏感性分析以考察试验结果对数据缺失机制假设的稳健性。Objective To explore the statistical performance of reference-based imputation under the hypothetical strategies to deal with missing data,so as to provide reference for the processing of missing data.Methods The mean square error(MSE),typeⅠerror rate and power of reference-based imputation(J2R、CIR、CR),LOCF and MMRM were compared through different missing rate SAS simulation data,and applied to the actual data of clinical trials.Results When there was no difference in efficacy between the two groups,the bias and MSE of RBI were lower than those of LOCF and MMRM.LOCF and MMRM overestimated the typeⅠerror rate,while RBI underestimated theⅠerror rate.When there was a difference in efficacy between the two groups,the bias and MSE of RBI were larger than those of LOCF and MMRM,and the power was lower.Moreover,the estimated treatment effect was the most conservative,and J2R was the most severe.When the missing rate is low(less than 10%),the power of RBI was similar to that of other methods,When the missing rate is high(more than 10%),the power of RBI is too low.Conclusion RBI is a relatively conservative imputation method.When the data missing rate is low,it can be used to impute the missing data under the hypothetical strategies.When the missing rate is large,it can be used as a sensitivity analysis of the main estimand to check the robustness of the experimental results to the assumption of data missing mechanism.

关 键 词：参照填补法 假想策略 缺失数据 重复测量混合模型 

分 类 号：R195.1[医药卫生—卫生统计学]