Role of sex and ethnicity in the relationships among physical activity, cognition, and AD biomarkers in the HABS-HD cohort.

In recent decades, the proliferation of
high-resolution, geospatial
environmental data setsin combination with the increasing
availability and accessibility of large-scale, real-world health data
that includes geographic identifiers such as residential addresshas
led to a growing interest in leveraging linkages between diverse “big”
data sources to examine impact of physical and built environments
on health and health disparities. Google
Earth Engine and other publicly accessible platforms and data tools
can offer highly spatially and temporally resolved information about
air pollutants, green space, light, heat, and other environmental
factors across broad geographic areas. For environmental epidemiologists,
the spatial coverage of these data can help facilitate subgroup analyses
in large samples and across diverse geographical contexts. In addition, large-scale environmental data can be harnessed to
track changing spatial patterns of climate-sensitive risk factors
and understand how these factors impact health outcomes. The
purpose of this commentary is to unpack potential challenges
associated with linking “big” geospatial data with “big”
health data (particularly real-world health data, including claims
and electronic health record (EHR) databases) in etiologic epidemiologic
research that seeks to shed light on environmentally driven patterns
of population health. In these types of studies, investigators must
grapple with the nuances of multiple overlaid data sources, each with
its own challenges related to validity and representation. We focus
on common challenges associated with each data source, particularly
in the context of data linkages. We then provide a series of guiding
questions and considerations for leveraging these data sources for
environmental epidemiology. While this commentary primarily focuses
on etiological analyses conducted within the U.S., examples and implications
are drawn from global research, highlighting the salience of many
of these challenges across diverse geographical areas. An immediate challenge facing investigators
using many geospatial data sets is the need to achieve an adequate
level of technical expertise in understanding how attributes of physical
or built environments are represented, as well as their potential
sources of bias and uncertainty. Geospatial data may be derived from
multiple sources. Producing spatially and temporally resolved modeled
air pollution exposure estimates, for example, may entail combining
multiple types of data inputs, including information from ground monitors,
satellites, and chemical transport modeling. Efforts
have been made in recent years to increase the accessibility of geospatial
data products for environmental monitoring and analysis, including
NASA’s Applied Remote Sensing Training (ARSET) program and
state-developed environmental screening tools, such as CalEnviroScreen
or Colorado EnviroScreen. Yet, technical barriers may remain
in understanding data inputs, processing data to align with available
spatiotemporal health data, assigning exposure windows that are etiologically
relevant, and characterizing potential sources of bias. The result
is that opportunities for meaningful research may be missed or underutilized
and research output is often insufficiently nuanced in articulating
the limitations of component data sources.