HIGH-RESOLUTION DATA ON MOBILITY, THE BUILT ENVIRONMENT, AND PHYSICAL ACTIVITY: Embedding smartphone GPS and consumer wearables within a cohort study to improve environmental epidemiology

Abstract

The ubiquitous nature of smartphones and wearable devices create novel opportunities for epidemiological exposure assessment and measurement error correction within the context of the built and natural environments. Since the foundations of the field, environmental epidemiologists have sought to establish methods to quantify and characterize environmental exposures, and to determine appropriate spatial and temporal scale. This dissertation set out to address the following gaps in the literature: 1) quantification of exposure differences between residential and mobility-based greenness; 2) explore momentary associations between greenness and physical activity through smartphone and wearable device data collection; 3) examine associations of walkability and physical activity at the minute level using GPS data; and 4) understand nondifferential misclassification of walkability exposure and implications for regression calibration on the association between residential walkability and physical activity. The first study (Chapter 2), addressing research gap 1, found residential-based distance buffer estimates of greenness are higher and more variable than mobility-based metrics. These findings contribute to discussions surrounding the choice of an optimal spatial scale for personal greenness exposure assessment. The second study (Chapter 3) sought to undertake the second research gap of momentary associations by utilizing objective physical activity data at fine temporal and spatial scales to present novel estimates of the association between mobility-based greenness and step count. Contrary to our hypotheses, higher greenness exposure was associated non-linearly with lower mean steps per minute after adjusting for confounders. We observed statistically significant effect modification by Census region and season. The third study (Chapter 4) targeted research gap 3 and took a similar approach to the previous chapter. We utilized comprehensive mobility data at fine temporal and spatial scales to present novel estimates on the real time association between walkability and physical activity. We found higher walkability exposure was associated with overall higher mean steps per minute. Associations were non-linear in nature. These findings contribute to discussions surrounding adapting the built environment to increase physical activity, resulting in the potential for improved health outcomes downstream. Lastly, the fourth study (Chapter 5) set out to assess the impact of measurement error found in residential-based walkability measures. We used mobility-based estimates to correct error-prone residence based estimates and then used these error corrected exposures to correct associations between walkability and self-reported physical activity for the error due to the use of residence-based exposures. This chapter highlights residential-based estimates of walkability slightly underestimate associations between walkability and physical activity. These findings highlight the impact of exposure misclassification on epidemiological studies of the built environment and physical activity. GPS data present a feasible solution to correct residential environmental exposures moving forward.

This dissertation contributes to the literature at the intersection of epidemiology, environmental health, and geography on human movement, the built environment, and physical activity. We build off an established U.S.-based nationwide prospective cohort through an internal mobile health (mHealth) substudy with momentary GPS and wearable data for exposure and outcome and in-depth information on individual and area-level covariates. By combining approaches from the fields of epidemiology and geography this mHealth dissertation explores improved exposure assessment using GPS, real-time mechanisms of association utilizing GPS data and integrating error corrections into large preexisting cohorts.