Precision Brain Health Monitoring for Alzheimer's Disease Risk Detection in the Framingham Study

Project Summary
The path to effective treatment and prevention of Alzheimer's disease (AD) depends on disease detection that
occurs before it is too late to reverse progression. Amyloid beta (Aß) is the widely accepted "gold standard"
biomarker of AD and current methods for measuring this biomarker rely on positron emission tomography (PET)
scans and/or analysis of cerebrospinal fluid (CSF). These AD biomarker acquisition methods, however, are
expensive, invasive, and difficult to scale and reliance on these approaches have exacerbated racial and ethnic
disparities in AD research. Digital technologies offer an alternative method for clinical phenotyping that can detect
AD-related changes well before the threshold of clinical symptom severity meets diagnostic criteria. Further,
digital phenotyping makes possible the identification and validation of digital biomarkers by determining digital indices
that correlate highly with more widely-accepted biological biomarkers. Within this context, this application seeks to
capitalize on the opportunistic timing of the Framingham Heart Study (FHS) middle-aged Generation 3 and Omni
Generations 2 cohorts as participants return for their NHLBI-funded 4th health examination. The NHLBI funding,
however, only covers costs associated with about 20% of the health exam components. The remaining 80% of
the health exam will be determined by ancillary studies such as the project proposed here. This project aims to
add two new components to the Gen 3/OmniGen 2 health exam. Aim 1 proposes conducting a novel lens Aβ
eye scan that pairs a topically-applied fluorescent Aβ-binding ligand with a specialized spectroscopic eye
scanner that can detect Aß deposition in the lens of the eye and has demonstrated higher sensitivity and
specificity to detect early AD-related Aβ pathology compared to amyloid-PET brain scans. Aim 2 seeks to use a
smartphone application to collect 3 years of longitudinal cognitive metrics from which to characterize those with
stable cognition versus declining cognition. Proposed analyses across these two aims will test the overall
hypothesis that novel digital cognitive profiles that are unique combinations of digital features (e.g., item-specific
responses, latencies, error rates, acoustic and linguistic measures) can detect those who are lens Aβ positive
and/or at high AD risk (e.g., high cardiovascular risk, ApoE4+, family history of dementia, women, age >60+).
Aim 3 will further apply traditional a priori and novel data-driven machine learning computational tools to construct
multi-marker profiles that are highly predictive (AUC > .85) of stable cognition and cognitive decline. We posit
that machine learning methods will generate more highly predictive models specific to digital cognitive profiles
as compared to a priori methods.

Grant Number: 5R01AG072654-03
NIH Institute/Center: NIH
Principal Investigator: Rhoda Au