Challenges and Opportunities in Geospatial Research on Pediatric Social Determinants of Health Free

Asthma is an environmentally sensitive condition, with both chronic symptoms and acute exacerbations influenced by various factors found in the environments in which children live, study, and play, such as the physical/built environment, the sociocultural environment, and the health care system.¹  These levels of influence, known as social determinants of health (SDOH), are key factors in the Healthy People 2030 guidelines. Healthy People 2030 is a set of public health priorities established to enhance health and wellbeing in the United States.² 

In this month’s Hospital Pediatrics, Tyris et al conducted a systematic review of geospatially associated SDOH and pediatric asthma emergency department visits and hospitalizations in the United States.³  The authors crafted a robust search strategy across 4 databases and identified 40 articles containing diverse measures of risk factors, outcomes, and geospatial areas. By aggregating and qualitatively synthesizing locally derived findings into the Healthy People 2030 SDOH goals, the authors are able to map what is known about using geographic information system techniques to understand the association between population-level SDOH and asthma-related emergency department visits and hospitalizations in the United States. There are, however, fundamental problems within the underlying literature. In this commentary, we aim to address and highlight some of the key errors commonly made in the emerging field of medical geospatial analysis and offer potential remedies.

This systematic review included 8 distinct geographic units across the 40 studies, with 16 (40%) using either ZIP codes or ZIP Code Tabulation Areas. Although commonly used, ZIP Code-based analyses have several limitations. ZIP Codes are created to optimize mail delivery, not statistical accuracy. This leads to confusing situations, such as a single office building having its own ZIP Code or a single ZIP Code crossing multiple counties and towns. These differences can have a real impact on geospatial-based analysis. For example, the authors of one study⁴  compared the associations of colon cancer incidence rates with area-based socioeconomic measures defined at either the census block group, census tract, or ZIP Code. Approximately 10% of the home addresses were geocoded successfully to census blocks and tracts but failed to successfully map to the decennial census ZIP Codes. This meaningfully impacted the outcomes; when the analysis was conducted at the ZIP Code level, areas with a higher median income had a higher incidence of colon cancer, the opposite of the findings of both census tract-based analysis and all other published literature. This is a prime example of the modifiable areal unit problem, in which statistical bias is created when point-based measurements are aggregated to different area units, such as ZIP Codes, census tracts, or counties.^5,6  Using ZIP Codes because they are conveniently available in the medical record without linking to better geographic boundaries is a common mistake that ought to be addressed in future work.

Because home addresses are considered protected health information (PHI), they deserve protection under the Health Insurance Portability and Accountability Act because patient privacy is critical for ethical medical research. To comply with Health Insurance Portability and Accountability Act regulations, the Department of Health and Human Services suggests 2 methods for safely deidentifying PHI: the “Expert Determination Method” or the “Safe Harbor Method.”⁷  The Expert Determination Method requires someone with experience in appropriate statistical methods to determine that there is a small risk that the intended recipient of the final research data could identify any underlying individual. This method is ideal because it allows for the use of the most granular data available, thus increasing precision. The Safe Harbor Method removes all identifiers so that there are no residual methods to identify underlying individuals in the data. The major problem with the Safe Harbor Method is that it only provides data up to the ZIP Code level, and even worse, it requires the use of only the first 3 digits of the standard 5-digit ZP Code, which necessarily aggregates populations to >20 000 people. Not only is this method susceptible to modifiable areal unit problem-related bias, but it compounds the problem by making ZIP Codes even more heterogeneous than the baseline. Fortunately, there are a number of open-source solutions that enable research teams and institutions to leverage geocoding while protecting PHI. For example, DeGAUSS is an NIH-funded tool created to link “individual-level environmental characteristics while maintaining the privacy of protected health information.”⁸  We hope to see more pediatric studies employing DeGAUSS and other tools to safely leverage granular geospatial data.⁹ 

Time is also an important variable in geospatial analyses. Data sets like the American Communities Survey and the Child Opportunity Index reflect data collected at specific moments in time. American Communities Survey data, for example, are available in 1-year and 5-year estimates, each with their own advantages and disadvantages. The temporal relationship between spatial variables and individuals is complex and can be another confounder in these types of studies. This ties into the larger issue of context; individuals and their health are deeply affected by their environment, but there is still a wide variability between what is true of a geographic area, what is true about any given person in that area, how much time they spend in it, and what protective factors and other modifiers they may experience. Our understanding of these relationships is still evolving, and more research is needed to model these relationships effectively for health care applications. Finally, it is important to note that, although the authors of this systematic review used a robust methodology that included working with a medical librarian, they did not include other databases and the gray literature. Medicine is relatively new to geospatial analysis, and much of the literature is from other fields like informatics, computer science, environmental studies, and urban planning. Some of these disciplines are less manuscript-driven and are more focused on conference papers and proceedings, making them less discoverable to the methods used here.¹⁰  Future scoping reviews should expand the search strategy to include insights from these sources.

Moving forward, it is critical that future research on pediatric asthma and SDOH takes steps to address the unique aspects of geospatial-based analysis. Specifically, researchers should use more granular geographic units, such as census blocks or tracts, and should consider the use of distance decay or other spatially weighted statistics. This will help to ensure that the findings are accurate and actionable and will enable policymakers and practitioners to design more effective interventions to reduce the burden of pediatric asthma.