Statistical Methods for Multilevel Multivariate Functional Studies

抽象的。多发性硬化(MS)是一种免疫介导的中枢神经系统疾病 (CNS)据估计，仅在美国就有40万人受到影响。MS的特征是 通过局灶性脱髓鞘损害，并导致身体和认知障碍。在进行影像研究时- IES广泛应用于临床实践和研究，具有针对性和较强的预测性 基于神经成像的生物标记物很小。因此，我们将重点放在两种很有希望的成像方式上 可能捕捉到关于MS疾病严重性和动态的补充信息：(1)纵向 弥散张量成像(DTI)捕捉到的脑白质完整性的变化-- Losum；(2)多序列磁共振图像中MS病变体素强度的纵向变化 磁共振成像(MRI)。为了解决这些新兴的数据结构，我们提出了现实的生物统计学-- CAL方法，可以放大并产生纵向高维数据的原则性推理。 fi第一个目标是大规模单变量广义线性混合效应模型(MU-GLMM)。 并提出了一种简单的推理方法来处理研究内和研究间的参与者 相关性。第二个目标是关于密集纵向高维度的联合建模。 局部数据(例如，病变体素强度)和存活时间(例如，体素恢复时间)。第三 AIM的目的是量化纵向神经成像和已建立的 MS生物标志物。第四个目标是致力于实现、软件和可重复性。

Abstract. Multiple Sclerosis (MS) is an immune-mediated disease of the central nervous system (CNS) that affects an estimated 400; 000 people in the United States alone. MS is characterized by focal demyelinating lesions and causes physical and cognitive impairment. While imaging stud- ies are widely used in clinical practice and research, the number of targeted and strongly predictive neuroimaging-based biomarkers is small. Thus, we focus on two promising imaging modalities that are likely to capture complementary information on MS disease severity and dynamics: (1) longitudinal changes in white matter integrity captured by Diffusion Tensor Imaging (DTI) across the corpus cal- losum; and (2) longitudinal changes in the voxel intensities of MS lesions in multi-sequence Magnetic Resonance Imaging (MRI). To address these emerging data structures we propose realistic biostatisti- cal methods that can scale up and produce principled inference for longitudinal high dimensional data. The first Aim is focused on massively univariate generalized linear mixed effects models (MU-GLMMs) and proposes a simple inferential approach for dealing with the within- and between-study participant correlation. The second Aim is concerned with the joint modeling of dense longitudinal high dimen- sional data (e.g., lesion voxel intensities) and survival time (e.g., time to voxel recovery). The third Aim is designed to quantify the association between the longitudinal neuroimaging and established MS biomarkers. The fourth Aim is dedicated to implementation, software, and reproducibility.