Dynamic Multivariate Normative Comparison and Risk Screening for Alzheimer's Disease Progression

点击翻译按钮获取中文摘要

This project focuses on the development of statistical methods for analyzing data from the Alzheimer Disease (AD) Research Center (ADRC). The first objective is to develop robust procedures to classify cognitive impairment over repeated visits. This has important clinical implications, as current diagnostic methods tend to falsely flag healthy subjects as impaired. The second project focuses on systematic evaluation of high-dimensional risk factors to select promising features that can separate those subjects who will develop AD, from those who might die, and those who will be alive and disease free by a certain time point. The completion of this project will lead to the identification of important risk factors that are predictive of both AD and survival. These newly identified biological, clinical, and genetic markers will guide future studies developing targeted intervention for AD. The proposed methods are relevant for disease diagnosis and risk screening but may also be applied to other areas such as economics, finance and engineering. The project will integrate research and education through the mentoring of graduate students. The first project concerns longitudinal measures of multiple domain scores of cognitive functioning modeled using multivariate mixed-effect models. A longitudinal multivariate normative comparison (MNC) statistic is then computed to measure the distance between a subject's domain scores and the estimated norm of healthy controls. Different thresholding methods are proposed for the longitudinal MNC based on the Chi-square approximation and permutation to identify cognitive impairment from retrospective data. Two familywise-error-rate controlling procedures are developed to dynamically screen for cognitive impairment at each ongoing visit, by comparing the p-values from the longitudinal MNC with adaptive significance levels. In the second project, a recently developed diagnostic measure of the volume under the ROC surface (VUS) is adopted as a model-free screening metric for ordinal competing endpoints. The VUS can be readily estimated as a concordance probability by some weighted U-statistics. The proposed screening procedure based on the U-type estimator of the VUS provides systematic and dynamic evaluation of markers' discriminatory capacity without any model assumptions. As the first screening method developed specifically for ordinal disease progression, the successful completion of the second project will contribute to the broader field of high-dimensional risk screening.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.