Can smartwatches predict cognitive status in older adults? Original paper

Data from two separate cohorts were collected and included the following:

• Complete 3-day accelerometer recordings with over 10 hours of daily valid wear time
• Cognitive status (defined as cognitive impairment or lower cognitive performance), evaluated using physician diagnoses and/or assessments for things like memory and executive function.

Machine learning was used to identify the most relevant and robust physical activity markers associated with cognitive function. These included measures of repeated movement patterns (like steps and arm swings) and measures of activity intensity transitions (like going from sitting to walking).

Models that used the top 10 physical activity markers common to the dataset, along with demographic characteristics (like age and education level), could reliably predict cognitive status (area under the curve above 0.8).