A note on the bias of standard errors when orthogonality of mean and variance parameters is not satisfied in the mixed model for repeated measures analysis
A note on the bias of standard errors when orthogonality of mean and variance parameters is not satisfied in the mixed model for repeated measures analysis
复制标题
关于重复测量分析混合模型均值和方差参数不满足正交性时标准误偏差的说明
DOI：
10.1002/sim.8474
复制
发表时间：
2020
影响因子：
2
通讯作者：
Gosho Masahiko
中科院分区：
文献类型：
--
作者：
Maruo Kazushi;Ishii Ryota;Yamaguchi Yusuke;Doi Masaaki;Gosho Masahiko
在纵向随机临床试验中，重复测量的混合效应模型(MMRM)分析有时被用作主要分析。在MMRM分析中，由于大多数标准统计软件的默认设置，通常通过假设固定效应和方差-协方差参数的正交性来估计治疗效果的SE，这是多变量正态分布的正交性。然而，当错误指定分析模型和/或数据包括具有随机丢失机制的缺失值时，该属性可能会丢失。在这项研究中，我们研究了正交性假设对MMRM分析中SE估计的影响。模拟和案例研究表明，具有正交性假设的SE具有不可忽略的偏差，特别是当应用假设处理组之间存在异方差的分析模型时。在不假定正交性的情况下，我们还介绍了用于MMRM分析的SAS代码。假设MMRM分析中的正交性会导致无效的统计推断，在大多数标准软件中应用MMRM分析时必须谨慎。
The mixed effect models for repeated measures (MMRM) analysis is sometimes used as a primary analysis in longitudinal randomized clinical trials. The SE for the treatment effect in the MMRM analysis is usually estimated by assuming the orthogonality of the fixed effect and variance‐covariance parameters, which is the orthogonality property of a multivariate normal distribution, because of default settings of most standard statistical software. However, this property might be lost when analysis models are misspecified and/or data include missing values with the mechanism of being missing at random. In this study, we investigated the effect of the assumption of the orthogonality property on the estimation of the SE for the MMRM analysis. From simulation and case studies, it was shown that the SE with the assumption of orthogonality property had nonnegligible bias, especially when the analysis models assuming heteroscedasticity between treatment groups were applied. We also introduce the SAS code for the MMRM analysis without assuming the orthogonality property. Assuming the orthogonality property in the MMRM analysis would lead to invalid statistical inference, and it is necessary to be careful when applying the MMRM analysis with most standard software.