PROJECT SUMMARY/ABSTRACT
Advances in early detection of Alzheimer’s disease (AD) pathology via imaging and fluid-based
biomarkers in individuals with little to no evidence of cognitive decline allow for the targeted enrollment
of participants who are most likely to show benefit from early interventions. However, many of these
biomarkers are costly and invasive. As a result, more accessible, performance-based measures of
cognitive function are needed to improve early detection, aid in recruitment for clinical trials, serve as
clinical endpoints for interventions, and improve access to individuals from underserved communities.
In this R01 proposal, we respond to this need: by leveraging existing longitudinal data of over 3,000
speech samples from two large cohorts, we propose validation of a digital speech marker across all
ADRD stages. Connected speech and language (CSL) analysis of digitally recorded speech—the
detailed measurement of everyday spoken language—is a noninvasive digital marker that measures
communication, an activity essential for quality of life. Advances in natural language processing and
Machine Learning have made automated linguistic analysis a rapid and effective means for analyzing
CSL, while the ubiquity of mobile devices makes remote and frequent digital speech collection widely
accessible to more people at risk for AD. Our central objective is to further develop, increase
accessibility, validate, and improve the sensitivity of digital speech markers to AD through new analytic
approaches and remote collection methods. By pursuing the following specific aims on a large-scale
dataset, we investigate early changes in CSL to predict how and when CSL relates to the risk of
developing ADRD: Aim 1: Validate connected speech and language measures from older adults across
multiple stages of ADRD, including prodromal ADRD, mild cognitive impairment, and dementia. Aim 2:
Test the hypothesis that connected speech and language measures are associated with AD
biomarkers, including Aß and tau from PET imaging, levels of tau, Aß, and neurodegeneration (NFL) in
CSF and blood plasma, and global neurodegeneration from MRI. Aim 3: Evaluate the usability and
accessibility of frequent, remote at-home speech collection and validate linguistic metrics collected
remotely against those obtained in person. The results of these three aims will lead to a larger goal of
understanding the CSL features that are sensitive to cognitive decline and AD neuropathology in at-
risk adults, and the linguistic and acoustic markers necessary to develop a widely accessible tool
measuring early cognitive change along the ADRD continuum.