Mechanistic insights into health disparities: leveraging longitudinal wearable data to assess the impact of built environment on physical activity and health outcomes - despite the known benefits of…

The National Heart, Lung, and Blood Institute, a division of the Department of Health and Human Services National Institutes of Health, awarded Duke University $129,399 on January 1, 2026, under the Cardiovascular Diseases Research program (CFDA 93.837) to conduct mechanistic research into health disparities driven by the built environment's impact on physical activity and health outcomes.

Duke University will leverage longitudinal wearable device data from the All of Us Research Program to develop a novel machine learning framework that generates physical activity behavior profiles from minute-level intraday step data. This approach will uncover mechanisms through which built environment factors influence physical activity patterns and, by extension, health outcomes. The research employs high-frequency wearable data to provide greater temporal specificity than conventional summary metrics such as average daily steps or moderate-to-vigorous activity minutes, enabling detailed insights into individuals' daily interactions with the built environment. The work will examine indirect effects of built environment characteristics on health through machine learning incorporating counterfactual analysis.

Work is performed in Durham, North Carolina. The period of performance runs from January 1, 2026, through July 31, 2027. The assistance type is a Project Grant.

Mechanistic insights into health disparities: leveraging longitudinal wearable data to assess the impact of built environment on physical activity and health outcomes - despite the known benefits of regular physical activity (PA), adherence to recommended PA guidelines remains alarmingly low, with marked variation observed across the population. PA behavior is influenced by genetic, behavioral, environmental, and social factors. In particular, the built environment-where people are born, live, learn, work, play, worship, and age-has become an increasingly important focus for interventions aimed at promoting healthy PA behaviors. However, understanding the role of the built environment in shaping PA behaviors has been limited due to incomprehensive measures of PA and the infeasibility of randomized controlled trials. Our project aims to address this GAP by leveraging the large, longitudinal wearable device data from the all of US research program (aourp). We propose a novel approach to represent PA behaviors and a machine learning framework to uncover the mechanisms through which the built environment factors influence PA and, by extension, health outcomes. We will utilize minute-level intraday step data from the aourp to generate PA behavior profiles-a data-driven representation for quantifying daily patterns of PA. This high-frequency data from wearable devices offers detailed insights into each individual's daily interactions with the built environment, offering greater temporal specificity than conventional summary metrics, such as average daily steps or minutes of moderate-to-vigorous activity. We hypothesize that PA will be strongly associated with a variety of factors, including the built environment, and that these associations will be more pronounced using PA behavior profiles than traditional summary measures. We also aim to study the indirect effects of the built environment on health outcomes through a machine learning framework that incorporates counterfactual analysis. Through this framework, we will examine how hypothetical modifications in the built environment could flip health outcomes by altering an individual's PA behavior. In addition, stratified post-hoc analyses will be conducted to explore the differential impacts of environmental changes across various subpopulations. This proposed project will advance our scientific understanding of population-level health outcomes through the development of a new approach for representing PA behaviors from wearable device data. By proposing a novel framework for counterfactual analysis, the project will demonstrate how observational data can be used to understand the underlying etiologic factors and mechanisms that contribute to variation in health outcomes.