Enormous effort has been made to uncover the series of changes in biomarkers along Alzheimer?s disease (AD) pathophysiological pathway and its later clinical manifestations. The most influential hypothetical model proposed by Jack and colleagues has greatly shaped AD research in the past decade, whereas it remains a hypothesis to be validated. The key challenge in the validation is the fact that AD pathophysiological process is not directly observable. The temporal biomarker profile is therefore usually examined against discrete clinical diagnoses, estimated years from clinical symptom onset or test score of cognitive impairment ? neither is a good measure of the AD pathogenic process, but merely clinical consequences that have been shown to vary greatly among individuals and also to be affected by other diseases. In this proposal, we will tackle this topic in the following aspects. (1) We will develop appropriate statistical models that directly address the unobservable nature of the AD pathophysiological process and therefore provide the foundation to operationalize and validate hypothetical AD biomarker models. (2) We will utilize data across multiple AD database to provide data-based evidence on the AD biomarker cascade and its clinical manifestations, as well as inform the link between the newly proposed biological AD definition in the 2018 NIA-AA research guideline and the current syndromic AD definition. (3) We will develop a statistical framework for dynamic prediction of AD pathophysiological progression trajectory and its clinical manifestations based on the history of a patient?s biomarker profiles. (4) We will develop a web-based application that allows for expedited delivery of statistical learning into practice. Although the scientific questions are focused, the proposed statistical model is applicable to many observational studies with longitudinal, multivariate biomarker measures to capture an unobservable structure, such as in aging or mental health studies.
The project develops appropriate statistical methodologies to evaluate temporal biomarker changes along a pathophysiological process that cannot be directly observed, bridge the gap between a rigorous evaluation and the practical constraints using existing data sources, and attempt to provide individualized prediction of a patient?s future progression trajectory and associated clinical manifestations based on the biomarker history of this patient. Methods are motivated by the need to understand the series of changes in biomarkers along the Alzheimer?s disease pathophysiological pathway during the asymptotic stage and its later clinical manifestations, but can be applied to many biomarker studies where the outcome is not directly observable, such as in aging or mental health studies.
