Image

Evaluating and Applying Google Location Data for Built Environment and Physical Activity Research

Funding: National Institutes of Environmental Health (NIEHS), R21ES031226 (PI: Hystad)

Environmental exposures are commonly estimated using spatial methods, with most epidemiological studies relying on home addresses. Passively collected smartphone location data, like Google Location History (GLH) data, may present an opportunity to integrate existing long-term time-activity data.

In this grant we collected GLH data from a subset of participants in the Washington State Twin Registry (WSTR) using a citizen science “Bring your own location data” approach. GLH included over 287 million location records from 357 participants. The map below shows GLH location data and how it can be used to assess personal environmental exposures.

GLH data is a feasible and cost-effective method for capturing retrospective time-activity patterns for large populations that presents new opportunities for environmental epidemiology. Cohort studies should consider adding GLH data collection to capture historical time-activity patterns. Privacy and Google policy changes around GLH data retention remains a concern that needs to be carefully managed when using GLH data.

Image

Deep Learning of Street View Imagery to assess Urban Green Space Relationships with Mental Health:  A Twin Study

Funding: National Institutes of Environmental Health (NIEHS), R21ES029722 (PI: Hystad)

We are developing advanced computer vision models to analyze street view imagery, not only to detect and classify elements of urban green space (e.g. trees, grass) but also to predict perceptions of the built environment, such as green space quality. By integrating deep learning techniques, our models can assess both the physical attributes and subjective perceptions of green spaces, provi ding a richer understanding of green space exposure. Applying these models to the WSTR study allows us to control for genetic and familial factors, isolating the urban green space impacts on mental health outcomes.

Image

Built Environment Assessment through Computer visiON (BEACON): Applying Deep Learning to Street-Level and Satellite Images to Estimate Built Environment Effects on Cardiovascular Health

Funding: National Health, Lung, and Blood Institute (NHLBI), R01HL50119 (PI: James)

The BEACON project aims to leverage advanced deep learning techniques to analyze both street-level and satellite images to assess the built environment’s impact on cardiovascular health. By using computer vision models, the project seeks to automatically classify various elements of the built environment—such as green space, walkability, and urban density—as well as urban quality measures—such as green space quality and safety — for the entire United State to link to Nurses Health cohorts.  We have developed urban quality deep learning models for urban green space quality, beauty, attractiveness, relaxing potential, safety from traffic, and safety from crime and have predicted these metrics for 120 million locations for 2008, 2012, 2016, 2020, which can facilitate new studies into how built environment composition and quality influence human health and well-being.

Image

TRANSIT Accountability Study: Impact of Vehicle Emission Regulations on Traffic-Related Air Pollution Exposures and Infant Health

Funding: Health Effects institute (HEI), RFA 18-1 (PI: Hystad)

In the United States, billions of dollars have been spent implementing interventions to reduce traffic-related air pollution (TRAP). These interventions are usually regulatory actions focused on reducing tailpipe emissions. However, they also include local programs to reduce traffic congestion and excess vehicle emissions, such as electronic tolls and roadway capacity improvements. Few health studies have empirically evaluated the direct impact of air pollution exposure reductions from these emission regulations and congestion reduction programs; no studies have examined infant health, an important population health outcome linked to air pollution exposures.

We applied geospatial exposure science methods to Vital Statistics data in Texas from 1996 to 2016 (n = 8.1 million recorded births) to isolate the influence of TRAP, and TRAP changes, on adverse birth outcomes.  We derived diverse environmental exposure measures for residential locations.

Image

We integrated “connected device” data into exposure models to measure vehicle congestion at the street-segment level to determine the additional effects of congestion-related air pollution on health outcomes. This is the first application of connected device data to measure population-level mobility in an environmental epidemiology study. We also developed methods to compare birth outcomes of maternal neighbors living upwind and downwind of the same high-traffic road, reducing the potential for confounding by SES and noise exposures.