JMIR Publications

ABSTRACT

Background:

Early detection of Alzheimer’s disease and related dementias is critical for effective intervention, yet conventional neuropsychological assessments are time-intensive and lack ecological validity, while neuroimaging and blood biomarkers are not always linked to real-world functioning. Digital phenotyping using consumer wearables offers a scalable approach to monitoring physiological and behavioral indicators of cognitive health.

Objective:

To examine the construct validity of smartwatch-derived heart rate variability (HRV), step count, and sleep metrics as indicators of cognitive abilities in older adults with healthy cognition or mild cognitive impairment (MCI). We hypothesized that lower HRV and step count, as well as greater sleep duration variability, would be associated with worse cognition.

Methods:

Participants aged ≥55 years (N=52, Mage=68.6; 42.5% male; 18.5% MCI; 29.6% Black/African American) were recruited from the Philadelphia region. Participants completed baseline neuropsychological testing and provided demographic and clinical information. Participants wore a Garmin Vivosmart 4 smartwatch for 23 hours/day over four weeks, passively recording HRV (SDNN [standard deviation of normal to normal intervals] and RMMSD [root mean square of successive differences]) during sleep and sedentary periods, daily step count, and sleep duration. Step count was derived from accelerometer data. Sleep variability was quantified using the coefficient of variation (CoV) across days. Generalized linear mixed models examined associations between wearable metrics and cognition; linear regression examined sleep duration variability. Models adjusted for relevant demographic, clinical, and device-use covariates,

Results:

Higher daily step count was positively associated with global cognition (b=1708.35, p=.021). HRV metrics were not associated with global cognition. In sensitivity analyses excluding participants with atrial fibrillation or taking medications affecting heart rate, sedentary RMSSD was positively associated with executive function (b=3.42, p=.039). Sleep duration variability was not associated with global cognition (b=-0.79, p=.721).

Conclusions:

Findings provide partial support for the construct validity of passively collected digital measures as markers of cognition in older adulthood. Step count emerged as a robust indicator of global cognitive performance, suggesting that simple mobility-based metrics may capture meaningful variation in real-world cognitive functioning. In contrast, HRV showed limited and context-dependent associations, emerging only for executive function in participants off medications affecting heart rate when measured during sedentary periods. Sleep duration variability was not associated with cognition, suggesting that coarse wearable-derived variability metrics may not adequately capture sleep features relevant to cognitive aging. Overall, results support the promise of select passive mobility metrics for cognitive screening while underscoring the need for more refined, context-sensitive digital phenotyping approaches.