Ahmadi A, Baghfalaki T, Ganjali M, Kabir A, Pazouki A
Department of Statistics, Faculty of Mathematical Sciences, Tarbiat Modares University, Tehran, Iran.
Department of Statistics, Faculty of Mathematical Sciences, Shahid Beheshti University, Tehran, Iran.
J Appl Stat. 2021 May 28;49(12):3164-3177. doi: 10.1080/02664763.2021.1931055. eCollection 2022.
In multivariate longitudinal studies, several outcomes are repeatedly measured for each subject over time. The data structure of these studies creates two types of associations which should take into account by the model: association of outcomes at a given time point and association among repeated measurements over time for a specific outcome. In our approach, because of some advantageous arisen from features like flexibility of marginal distributions, a copula-based approach is used for joint modeling of multivariate outcomes at each time points, also a transition model is used for considering the association of longitudinal measurements over time. For the problem of incomplete data, missingness mechanism is assumed to be ignorable. Some simulation results are reported in different scenarios using the Gaussian, and several commonly used copulas of the family of Archimedean copulas. Akaike information criterion (AIC) is used to select the best copula function. The proposed approach is also used for analyzing a real obesity data set.
在多变量纵向研究中，随着时间推移，对每个受试者的多个结局进行反复测量。这些研究的数据结构产生了两种关联，模型应予以考虑：给定时间点结局之间的关联以及特定结局随时间重复测量之间的关联。在我们的方法中，由于边际分布的灵活性等特征带来了一些优势，因此在每个时间点使用基于copula的方法对多变量结局进行联合建模，同时使用转换模型来考虑纵向测量随时间的关联。对于数据不完整的问题，假定缺失机制是可忽略的。使用高斯分布以及阿基米德copula族中的几种常用copula，报告了不同场景下的一些模拟结果。使用赤池信息准则（AIC）来选择最佳的copula函数。所提出的方法还用于分析一个真实的肥胖数据集。
