A Next Generation Data Infrastructure to Understand Disparities across the Life Course

University Of Southern California, Los Angeles CA

Investigators

Abstract

Alzheimer’s disease and related dementias (ADRD) are an escalating public health crisis that affects >30% of seniors and costs $300-640B annually in the US. ADRD disproportionately harms certain phenotypic subgroups like individuals with diabetes, that experience higher rates and earlier onset. Behavioral interventions may mitigate much physiological and societal ADRD burden, but their development and potential clinical impact face two key obstacles: (1) discerning subtle, abnormal cognitive changes from normal aging that can manifest years prior to clinical ADRD; and, (2) isolating modifiable risk factors related to this change from highly multidimensional, interconnected determinants that manifest in different ways across people and time. Large-scale person-generated health data (PGHD) from digital technologies offer a vital opportunity to develop effective behavioral interventions to mitigate ADRD risk as they enable better detection of subtle cognitive changes and isolation of modifiable risk factors from a complex background of everyday life-course determinants. PGHD are non-invasive, low-burden, real-world, and high-frequency/continuous, reducing errors from intermittent clinical assessments, and allowing targeted, precision health applications through use of AI/ML. However, the field currently lacks “benchmark” PGHD, limiting our ability to train and validate transparent, interpretable, reproducible, and generalizable precision ADRD models. ADRD benchmark PGHD should (1) well-reflect US population characteristics; (2) pair individual passive datastreams with frequently-repeated active measures of cognition and life-course ADRD determinants (biological, psychosocial) using validated instruments sensitive to subtle, real-world change; (3) span longitudinal, life-course designs; (4) scale for AI/ML applied to individual, subgroup, or population levels; and, (5) embody findable, accessible, interoperable, and reusable principles. While no extant dataset fulfills these criteria, our team is uniquely suited to address this gap. The objective of this competitive revision to U01AG077280 is to fill the above criteria to generate benchmark PGHD that enable precision assessments of life-course ADRD risk. We will leverage the aims/infrastructure of the parent U01, which expands the Understanding America Study (UAS), a probability-based Internet panel that collects rich data to study life-course heterogeneities in preventable chronic conditions like ADRD. We will concurrently expand UAS’s American Life in Realtime (ALiR; R01LM013237) substudy, the first PGHD benchmarking pilot that fulfills all criteria except sample size. We will, (1) expand UAS-ALiR’s cohort to ~10,000 followed over 2.5 years to generate PGHD with sufficient statistical power to detect differences in ADRD-related outcomes across phenotypic subgroups, adapting successful UAS-ALiR methods; (2) develop precision models of ADRD risk by applying statistical, AI/ML, and quasi-experimental methods to different combinations of PGHD; and, (3) develop PGHD triggers for future just-in-time methods where meaningful deviations from individual-specific baselines in passive PGHD trigger an automated intervention or assessment.