Detecting dynamic pathways and risk indicators of resilience, vulnerability and stability in normally aging females and males: Integrating data-driven longitudinal and machine learning analyses

Resilience has become an integral component of theory and research in brain aging. Recent frameworks have provided guidelines for identifying and investigating resilience in the context of diversity (e.g., sex and risk exposure differences). We leverage a multi-factorial longitudinal approach to operationalize and predict pathways of resilience, stability and vulnerability as a function of differential within-person trajectory patterns of hippocampal volume and cognitive change. Participants included cognitively unimpaired (at baseline) older adults (n = 415; M age = 74.9; 59.9% Female) from the Alzheimer's Disease Neuroimaging Initiative with longitudinal imaging and cognition data, plus multimodal risk indicators. We used latent class growth analyses to identify hippocampal and cognitive trajectory classes. Machine learning applications identified risk indicators associated separately by pathway in the entire sample and disaggregated by sex. Based on differential combined class membership, we identified three subgroups: resilience (declining hippocampal/stable cognitive trajectories), stable (stable hippocampal/stable cognitive trajectories), vulnerable (stable hippocampal/declining cognitive trajectories). Leading risk indicators of resilience (e.g., lower plasma Aβ1-42) and vulnerability (e.g., lower education) originated from the biological and demographic domains. Leading risk indicators of stability originated from the biological and genetic domains (e.g., higher CSF Aβ1-42, lower polygenic hazard score). Sex-disaggregated results highlighted indicators unique to the females, such as higher body mass index (stability) and lower geriatric depression score (stability, vulnerability). Pathway patterns of resilience, stability and vulnerability were characterized objectively in initially cognitively unimpaired older adults based on specified hippocampal and cognitive trajectory combinations. Risk factors from a broad range of modalities should be considered in future research.