Statistical Methods for Integrating Mixed-type Biomarkers and Phenotypes in Neurodegenerative Disease Modeling

点击翻译按钮获取中文摘要

Project Summary: Neurological disorders pose an immense burden on patients, families, and health care systems, thus un- derscoring the urgent need to develop disease-modifying treatment. Research on Alzheimer's disease and other degenerative diseases faces unique challenges, including the fact that these disorders typically have slow pro- gression, the diagnostic criteria rely on clinical symptoms due to a lack of highly sensitive and specific biomarkers in many diseases, and there is substantial disease and subject heterogeneity. Precision medicine initiative has re- cently been launched in diverse populations, providing access to data on hundreds of thousands of individuals and millions of families history records. International consortia of neurological disorders (e.g., National Alzheimer's Coordinating Center, Parkinson's Progression Markers Initiative) were established to unify multi-modality data in several large existing cohorts and provide access to new biomarker data. Digital phenotypes as real-world as- sessments of patients are being tested as digital biomarkers to track disease progression and potentially serve as digital endpoints for clinical trials. The main goal of this proposal is to develop modern analytic methods for these new data types and studies in order to address the emerging challenges in research on personalized disease susceptibility, progression, diagnostics, and treatments. We will build data-generative models that will lever- age complementary contributions from multi-type biomarkers, including genomic measures, brain neuroimaging, biofluids, comprehensive neuropsychiatric assessments, and digital biomarkers. These methods will be applied to carefully selected clinical data collected by the investigative team or available from large consortia in order to guide genetic counseling and risk prediction, assist disease staging and clinical trial design, and optimize person- alized treatment policies. Specifically, in Aim 1 we will jointly analyze co-morbidities from family data to estimate co-heritability and dissect whether shared phenotypic co-variation is due to latent environmental factors, genetic factors, or both. In Aim 2, we will develop an integrated, multi-domain dynamic system for mixed-type biomarkers (e.g., neuroimaging, genomics, fluid, clinical, neuropsychiatric markers) using differential equations modeled on a novel progression scale that uses latent processes. In Aim 3, we will leverage gold-standard neuropathological diagnosis based on postmortem brain autopsy combined with antemortem biomarkers to improve clinical diagno- sis of neurological disorders. In Aim 4, we will model multi-domain digital phenotypes to learn optimal treatment policy from digital biomarkers. In each aim, we will establish theoretical properties using modern empirical pro- cess theory and statistical learning theory. Together, the state-of-the-art analytic methods proposed here will substantially improve analytic accuracy, and our combined statistical and clinical expertise will ensure that our methods are translated directly back to the clinical and translational research community.