NIH Research Festival
Background: Previous studies of physical activity (PA) have focused on estimating average levels of PA for people. An alternative perspective is to examine correlates of active days, but little is known about specific factors contributing to an active day and how neighborhoods play a key role. This perspective could inform the development of effective population-level interventions to increase PA.
Methods: A total of 2,625 participants (mean age [SD] = 45.2 [15.38]) from the AmeriSpeak panel, aged 20-75, completed up to two activity recalls over 24 hours using the Activities Completed over Time in 24 Hours (ACT24) instrument in 2019. A physically active day was categorized as "sufficient" (‚â• 1.6 physical activity levels [PAL]) or "insufficient" (<1.6 PAL). A set of 25 variables were analyzed, including demographics (e.g., age, gender, and occupation), health-related factors (BMI), and neighborhood characteristics (segregation and walkability indices, and county-level census variables). Supervised machine learning (ML) algorithms, including random tree-based models, were used to identify the key correlates of a physically active day.
Results: We have identified the top 10 predictors that significantly contribute to a physically active day at individual and neighborhood levels. These predictors were ranked in the following order: population density, age, walkability, education, income, region, race/ethnicity, county-level poverty, marital status, and BMI.
Conclusion: Tree-based ML algorithms suggest that both individual and area-level characteristics are associated with active days. Furthermore, some of the identified correlates of active days include modifiable features of the environment and could inform community-level interventions for at-risk areas.
Scientific Focus Area: Epidemiology
This page was last updated on Monday, September 25, 2023