Quantifying the Integrity of Sleep-Dependent Memory Processing in Pathological Aging and Alzheimer's Disease: Toward Inexpensive Electroencephalographic Wearable Applications - PROJECT SUMMARY/ABSTRACT Alzheimer's disease is a leading cause of death in the U.S. and there is an urgent and unmet need to develop digital technology for early detection and monitoring of Alzheimer's disease and related dementias. Single- channel sleep electroencephalography (EEG) is inexpensive, easily scalable, and can be recorded in a home setting from currently available “wearable” headband devices. Within sleep EEG, oscillatory events that reflect activity in memory circuits have been associated with aging, cognition, and markers of preclinical Alzheimer's disease. Thus, single-channel sleep EEG provides the framework of practical digital technology that can be used to build a powerful tool for monitoring brain health and for detection of preclinical neurodegenerative disease. Real world application of this technology will require significant work to better understand how oscillatory events in sleep EEG reflect brain aging and neurodegenerative processes, as current knowledge of this neurophysiology remains insufficient to build a robust digital biomarker. Moving to bridge this critical gap, our team has pioneered advanced signal processing methods and generated compelling preliminary data from machine learning approaches that demonstrate convincing predictions in aging, cognitive measures, and biomarkers of Alzheimer's disease pathology. Here we propose to substantially advance this approach by applying a robust machine learning approach that will build the foundational algorithms of a digital biomarker for detecting pathological aging and early stages of neurodegeneration. Our innovative approach will incorporate our team's domain knowledge of sleep's oscillatory events with the power of large-scale machine learning. Utilizing more than 15,000 sleep recordings from nine existing cohorts, we will train and independently validate predictive models as the basis for novel digital biomarkers. Our overall goal is to create digital biomarkers that are sufficiently accurate to monitor personalized brain health via a single channel of sleep EEG. The overarching hypothesis of our proposal is that metrics of sleep EEG capturing the health and consistency of memory processing circuits can be incorporated into machine learning models to provide robust predictions of 1) pathological brain aging, 2) cognitive decline, and 3) neuroimaging and molecular biomarker changes that occur early in Alzheimer's disease pathogenesis. We will interrogate this hypothesis with the following Specific Aims: (Aim 1) Elucidate the oscillatory event features of sleep EEG that best predict brain age and determine the performance of these features in assessing whether an individual is experiencing more “youthful” or “accelerated” brain aging, (Aim 2) Examine the relationship between sleep's memory-processing oscillatory circuit integrity and cognitive decline, (Aim 3) Delineate the relationships between memory-processing oscillatory circuitry integrity and Alzheimer's disease-related neuroimaging and molecular biomarkers. Our proposal is significant because it will seek to build the foundation for a digital biomarker that can monitor brain health and identify pathological aging and early neurodegenerative changes using practical and accessible digital technology.
