Background: The development of evidence-based interventions for delaying or preventing cognitive impairment is an important challenge. Most previous studies using self-report questionnaires face problems with reliability and consistency due to recall bias or misclassification among older people. Therefore, objective measurement of lifestyle components is needed to confirm the relationships between lifestyle factors and cognitive function. Aims: The current study examined the relationship between lifestyle factors collected with wearable sensors and cognitive function among community-dwelling older people using machine learning. Methods: In total, 855 participants (mean age: 73.8 years) wore a wristband sensor for 7.8 days on average every 3 months. Various lifestyle parameters were measured, including walking steps, conversation time, total sleep time (TST), sleep efficiency, time awake after sleep onset, awakening count, napping time, and heart rate. Random forest (RF) regression analysis was used to examine the relationships between total daily sensing data and Mini-Mental State Examination (MMSE) scores. Confounding factor analysis was conducted with models that were adjusted and unadjusted for demographic and vascular risk factors, and selected variables were assessed as risk and protective factors using partial dependence plots (PDPs). Results: Lifestyle data were collected for 31.3 ± 7.1 days per year using wristband sensors. RF regression analysis adjusted for age, gender, and education levels selected four variables, including number of walking steps, conversation time, TST, and heart rate. Moreover, walking steps, conversation time, and heart rate remained after RF regression analysis adjusted for demographic and vascular risk factors. Number of walking steps, conversation time, and heart rate were categorized as protective factors, whereas TST was categorized as a risk factor for cognitive function. Although PDPs of number of walking steps and heart rate revealed continuously increased MMSE scores, those of conversation time and TST and revealed that the tendency in the graph was reversed at the boundary of a particular threshold (321.1 min for conversation time, 434.1 min for TST). Conclusions: Lifestyle factors, such as physical activity, sleep, and social activity appear to be associated with cognitive function among older people. Physical activity and appropriate durations of sleep and conversation are important for cognitive function.
Keywords: cognitive function; cross-sectional study; lifestyle factors; mini-mental state examination; random forest regression analysis; wearable sensor.