Exploring heterogeneity of cognitive trajectories of preclinical Alzheimer's disease with digital phenotyping

PROJECT SUMMARY The heterogeneous nature of Alzheimer’s disease (AD) manifestation and progression poses a significant challenge to developing effective treatments. As a first step toward personalized medicine for AD, it is crucial to identify clinically relevant subtypes, which will enable the creation of individualized interventions and healthcare strategies. Cognitive function changes as people age, but there is considerable variability both between individuals and across cognitive domains. Accurately characterizing these cognitive trajectories is particularly challenging due to the limited availability of long-term, longitudinal data and the inherent complexity of computational models needed to account for the multiple risk factors that influence cognitive decline. Traditional neuropsychological (NP) tests often fail to capture the subtle and early changes in cognition that precede a clinical diagnosis of AD. This gap underscores the need for more sensitive and continuous methods of tracking cognitive decline over time. Recent advancements in digital voice (dVoice) recordings offer a novel and highly efficient method for capturing cognitive-related measures and characteristics, providing continuous data that may reveal subtle differences in cognitive processing. This project aims to leverage multimodal data, including both traditional NP tests and dVoice recordings, to identify preclinical cognitive trajectories. By incorporating a range of clinical risk factors, and utilizing the rich, longitudinal dataset from the Framingham Heart Study (FHS), we aim to build a comprehensive model of cognitive progression. Cognitive domain scores derived from NP tests will be complemented by composite scores representing different categories of non- semantic acoustic features extracted from dVoice recordings. These scores will then be integrated into group- based multi-trajectory models to map the progression of cognitive change over time. The primary research objectives are: (1) to identify preclinical cognitive trajectories from multimodal measurements; (2) to examine the association between these cognitive trajectories and incident AD, while investigating the specific risk factors associated with membership in different cognitive trajectory groups; and (3) to develop machine learning models that predict cognitive trajectories based on baseline measurements. We hypothesize that the integration of non-semantic acoustic features with traditional NP tests will significantly enhance the precision of cognitive trajectory modeling, offering new insights into the early stages of cognitive decline. By advancing the understanding of cognitive trajectories and identifying key baseline factors that predict cognitive decline, this project directly aligns with the objectives of the NIH R03 grant. It will provide valuable scientific insights that have the potential to improve the prevention and diagnosis of individuals at risk of AD.