Physical activity in preclinical Alzheimer disease: current evidence and implications for multimodal longitudinal research

Alzheimer disease (AD) pathology may be detectable years before cognitive symptoms, but biomarker testing of cognitively unimpaired people is not recommended for routine clinical use. Objective physical activity (PA), gait, and rest-activity measures are attractive research signals because they are repeatable and relatively inexpensive; however, none are specific to AD. This narrative review examined evidence linking PA, gait, and rest-activity measures to AD-related outcomes and considered their role in multimodal longitudinal research within biomarker-characterized cohorts. Targeted searches of PubMed, Scopus, and IEEE Xplore were conducted from inception through 25 August 2026, and reference lists of pivotal studies and guidance documents were hand-searched. Priority was given to guidelines, longitudinal cohorts, randomized trials, studies using objective behavior measures and AD biomarkers, and externally evaluated prediction models. Observational evidence links higher activity with lower dementia risk, but reverse causation remains plausible. In biomarker-characterized cohorts, greater baseline step count has been associated with slower amyloid-related tau accumulation and slower cognitive and functional decline, not with lower amyloid burden. Rest-activity fragmentation has shown cross-sectional and prospective associations with amyloid-related measures, whereas a 1-year aerobic-exercise trial did not reduce amyloid accumulation. Recent multimodal AI studies demonstrate that non-PET (positron emission tomography) data can estimate amyloid and tau status. Yet, they do not establish the incremental value of free-living PA in cognitively unimpaired community samples. We therefore outline a framework for longitudinal behavioral research in cohorts whose biomarker status has been characterized independently of PA-based selection. Its primary purpose is to examine associations between behavioral measures and subsequent cognitive, functional, and tau trajectories. Whether PA improves individual prediction beyond clinical information and baseline biomarkers remains a separate, unresolved question. PA is considered a contextual behavioral measure, and the framework does not propose using it to determine eligibility for biomarker testing or research recruitment.