Alzheimer's disease and related dementias (ADRD) is a growing epidemic, and in the absence of effective
treatment, disease burden increases as the population ages. In both ADRD and mild cognitive impairment
(MCI), there is significant temporal variability in disease progression, increasing the difficulty for managing
patient comfort and safety. Early detection of symptomatic states and continuous monitoring are regarded
as effective measures to minimize the impact of the disease as various forms of intervention can provide
opportunities for treatment, compensation and coping. However, current clinic-based cognitive and
behavioral assessments have numerous shortcomings; they are largely non-quantitative and clinicians
often have difficulty determining if there has been significant changes in neurologic condition between
visits. Additionally, assessments are obtained infrequently, and do not objectively account for
disease-related behaviors that could be revealed in daily activities. In this project, we propose to advance
new computational approaches and analytics to identify digital biomarkers for ADRD detection, prediction
and monitoring outside the clinic. This technology-driven approach is based on sensor data passively
acquired from commodity smartphones and wearables, and provides the foundation for a novel embedded
assessment of cognitive status through continuous monitoring.This proposal presents several research
opportunities. Firstly, we will advance passive and continuous data collection methods using multimodal
sensing. Challenges we will address include optimizing battery use for long-term data capture, and
mitigating privacy concerns by performing on-device data and feature pre-processing. Secondly, we will be
building on state-of-the-art research techniques in behavior and context recognition, speech analysis, and
machine learning to identify digital biomarkers of Alzheimer's disease and related disorders. We will
leverage these biomarkers to build computational models for disease stage characterization and
prediction, and individualize them by incorporating race and ethnicity risk factors as priors. Lastly, to
facilitate the use of these models and digital biomarkers in clinical practice, we will advance a novel visual
analytics interface towards helping physicians and health practitioners interact with the acquired sensor
data, validate the digital biomarkers, verify model results, and forecast the progression of disease.
RELEVANCE (See instructions):
A clear and specific clinical need motivates this proposal: improved and continuous understanding,
monitoring, characterization, assessment and prediction of a prevalent neuro-cognitive condition in
naturalistic settings. ADRDs are difficult and costly diseases to treat, affecting millions of people in the U.S
alone. Our approach provides the foundation for a new direction in the early detection and prediction of this
devastating and highly-debilitating condition.