Walkability and Physical Activity in Nurses’ Health Study 3 with Measurement Error Correction Using Smartphone Mobility Data

Abstract

Background and Aim: Studies of walkability and physical activity often rely on residential measures that omit exposures occurring in other environments, likely leading to exposure misclassification. We measurement-error corrected traditional residential walkability measures using smartphone mobility-based walkability estimates and examined its association with self-reported physical activity in a US-based cohort. Methods: We calculated typical minutes per week spent walking and running from Nurses’ Health Study 3 (NHS3) participants using validated physical activity questionnaires (n=23,983). We spatially joined residential addresses to census tract-level walkability, calculated by summing z-scores of 2018 TIGER/Line road intersection density; 2018 Infogroup business density; and 2015-2019 5-year American Community Survey population density estimates. A subset of NHS3 (n=337) provided smartphone GPS data every 10 minutes across four 7-day sampling periods that captured seasonal variability, which was used to create our ‘gold standard’ walkability exposure. We used standard regression calibration for generalized linear models to produce corrected estimates and 95% confidence intervals for associations of walkability with physical activity in the full NHS3, adjusting for confounders. Results: Participants reported on average 273.15 minutes of physical activity per week (SD = 257.00), had a mean residential walkability score of 0.03 (SD=2.64), and had a mean GPS based walkability score of 0.36 (SD=2.39). Each SD increase in uncorrected residential walkability was associated with a 7.95-minute increase in time spent walking and running per week (95% CI; 3.83, 12.08), whereas each SD increase in measurement error corrected walkability was associated with a 10.80-minute increase in time spent walking and running per week (95% CI: 4.96, 16.64). Conclusions: Our study indicates that traditional residential estimates of walkability may underestimate associations between walkability and physical activity. These findings highlight the impact of exposure misclassification on epidemiological studies of physical activity and the built environment. Keywords: Wearables, Mobility, Measurement Error, Physical Activity, Exposure Correction