Alzheimer’s Disease and Related Dementias (AD/ADRD) are progressive neurological disorders causing cognitive, functional, and motor decline. Current monitoring relies on infrequent specialist evaluations and subjective reports, limiting accuracy. There is a need for objective, continuous monitoring tools to track disease progression effectively.
We propose developing digital biomarkers from commercial wearable devices for monitoring AD/ADRD progression. Wearables enable continuous, real-time, and objective data collection in natural environments, providing personalized insights. Previous studies have demonstrated association of disease progression with accelerometer data (e.g., cadence, daily steps), actigraphy data (e.g., sleep fragmentation, circadian rhythms) and physiological data (heart rate variability). These studies suggest that wearable data can be successfully used to monitor disease progression. However, existing approaches do not integrate multiple data streams with large-scale AI models.
To address this, we propose to develop a foundation model (large-scale task/domain agnostic AI model trained on unlabeled data) using wearables data through self-supervised learning. We will use the UK Biobank accelerometer data to enrich the model’s learning and allow it to accurately identify and generalize across a spectrum of cognitive health-from normal cognition to various stages of cognitive impairment. Subsequently, we will fine-tune the foundation model to classify stage of cognitive impairment using the ICD 10 codes for mild cognitive impairment (MCI) and dementia. Finally, we will evaluate the model using a in-house, deeply-phenotyped dataset that will provide 1-year of longitudinal data on 60 individuals (healthy control n=20, MCI n=20, mild dementia, n=20).
Our overall goal is to develop an AI foundation model using wearable sensor data and assess its feasibility for tracking AD/ADRD progression.