Accelerometry-Measured Physical Activity Predicts Dementia Onset in a National Cohort of Older Adults

Abstract Identifying scalable, objective markers of dementia risk is essential for early prevention and intervention. Physical activity is a key modifiable risk factor, yet evidence from device-based measures in representative populations remains limited. Using nationally representative data from the National Health and Aging Trends Study (NHATS), we examined whether wrist-worn accelerometry predicts incident dementia in a cohort of 562 participants aged ≥ 65 years without dementia at baseline. Over one year of follow-up, higher physical activity levels were associated with lower risk of incident dementia (OR = 0.66, 95% CI [0.55 - 0.79]), even after adjusting for age, gender, and comorbidities (OR = 0.68, 95% CI [0.57 - 0.82]). Receiver operating characteristics (ROC) analyses indicate that accelerometry modestly outperforms self-reported physical capacity in predicting dementia onset. These findings support the use of wearable-derived activity metrics as a potential digital biomarker for dementia risk stratification in population and clinical settings.