Genomics-guided sleep biomarker discovery for early Alzheimer's disease: A wearables study

PROJECT SUMMARY This proposal is responsive to NIA solicitation PAR-22-093/NOT-AG-21-048 for projects involving the development of digital technology for the early detection of Alzheimer’s disease. It is motivated by the need for novel low-cost and noninvasive digital biomarkers for Alzheimer’s disease that can flag early changes in at-risk

individuals before cognitive symptoms surface. Changes in sleep patterns have been linked to a future dementia diagnosis in older adults. These changes can be tracked in a passive and unobtrusive way using a wide variety of low-cost consumer wearables. Our overarching objective is to deploy cutting-edge wearables for sleep, heart

rate, and activity monitoring in an older adult cohort with elevated genetic susceptibility to Alzheimer’s disease from the Mass General Brigham Biobank in order to quantitatively evaluate each device, to develop artificial intelligence (AI) approaches for digital phenotyping of Alzheimer’s disease, and to perform hypothesis-driven

statistical modeling for understanding the interplay between Alzheimer’s genetic risk, sleep, and Alzheimer’s pathophysiology. To capture early changes, this project will rely on a unique pool of older adult participants who have elevated polygenic risk of Alzheimer’s disease but are cognitively normal at baseline. We will acquire (i)

electroencephalography (EEG) data using an EEG headband, (ii) heart rate, accelerometry, blood oxygenation, and temperature data using a smartwatch, (iii) heart rate variability data using a smart ring, (iv) plasma biomarker measures for amyloid, tau, and neurodegeneration (ATN), and (v) cognitive scores. The data will be collected

longitudinally at two timepoints with a gap of 2-3-years from N=260 elderly participants at the highest genetic risk for Alzheimer’s disease and 40 low-risk controls based on their polygenic risk scores. We will rigorously validate all three wearables by benchmarking each against polysomnography (PSG), which is the established gold

standard for sleep monitoring, and report their accuracies at measuring sleep and heart rate features. We will develop an AI model known as a transformer to learn composite features from EEG, heart rate, and accelerometry data that predict Alzheimer’s pathophysiology. Finally, we will use this data to test the hypothesis

that pathway-specific polygenic risk for Alzheimer’s disease and sleep disruption associates or interacts with distinct sleep features to predict Alzheimer’s pathophysiological or cognitive endpoints. The project features an interdisciplinary investigative team with expertise in AI, human wearables studies, sleep genetics, sleep EEG

quantitative analysis, preclinical Alzheimer’s neurology, PSG scoring, fluid biomarkers for Alzheimer’s disease, and biostatistics. We do not propose this technology as a substitute for lab-based approaches to detect Alzheimer’s pathophysiology or cognitive change. We envision that the combination of an unobtrusive wearable

device, genetic risk assessment, and AI tools for wearables-based Alzheimer’s digital phenotyping could identify/flag at-risk individuals in the general population for clinical follow-up. Thus, this project could thus have broad clinical impact in the Alzheimer’s field.