Abstract:
|
Multivariate longitudinal data across modalities arise in many chronic diseases such as Alzheimer Disease (AD), measuring different aspects of the underlying disease processes. An important scientific question is to locate the biomarkers exerting the earliest disease-related changes. The answer is crucial as it may help identify the target for earliest possible prevention or therapeutic intervention. Another important question is the temporal orderings of changes in these biomarkers during the potentially decades long disease processes. We develop a statistical methodology to address these questions by jointly modeling multiple biomarkers, and define the biomarker ordering by using the estimated rates of changes. Specifically, we choose a random intercept and random slope model for each biomarker, and propose a statistical test for the ordering of changes across biomarkers whose distribution depends on the Bessel functions. Finally, we apply the proposed methodology to the database of a longitudinal biomarker study of AD including cerebrospinal fluid and neuroimaging biomarkers, and infer on their orderings changes during the preclinical stage of AD.
|