Community resilience indicators to inform geospatial health analyses: curating the resilience domain of the North Carolina multi-stressors database (NCMSD)

This study curated a geospatial database of tract-level resilience indicators to form the Resilience Domain of the North Carolina Multi-Stressors Database (NCMSD). NCMSD represents a larger effort currently being organized by our team, aimed to characterize multiple facets of vulnerability and resilience to factors affecting the health and well-being of people living in the state of North Carolina (NC). This larger database will be published and made publicly available after individual domains are finalized, such as the current Resilience Domain.Here, we define resilience as the community resources and social structures that collectively strengthen and sustain community health and prevent, manage, or reduce the impact of adverse events. This definition represents a reframing of the classical view that resilience is the individual-level capacity to endure or recover from adversity (1). While resilience has been used in the past to task individual community members with becoming more "resilient" to public health challenges and natural disaster vulnerabilities (2), here, we apply the concept of resilience to describe broader social structures and community resources that could promote public health, as has been more recently proposed (3). We specifically focused on representative indicators across the following categories: (i) health and wellness, (ii) social and economic, (iii) infrastructure, and (iv) political capital, with indicators prioritized that addressed community resources and strengths relevant to human health. These categories are in accordance with recent community resilience frameworks that holistically capture community resources and capacity (1,4,5). While other efforts have sought to compile these indicators, they have been limited by the varying spatial and temporal scales at which data on resilience indicators are available and the patchwork of sources of these data, which include governmental agencies, non-profit organizations, and advocacy groups (6)(7)(8)(9)(10)(11)(12)(13). Through the creation of this database, we seek to enable the integration of measures of community resilience into public health research, starting with NC.To create this database, we established criteria for prioritizing resilience indicators, including data quality, non-missingness, geographic variation, interpretability, geospatial resolution, and lack of redundancy between indicators. When possible, indicators were prioritized that had low missingness (less than 5%) and were available geospatially at the census tract level. Composite measures such as risk indices were generally not included due to limitations in interpretability. We performed basic statistical analyses, including correlations between measures, to identify resilience indicators that captured meaningful variation across the state without being redundant. Indicators with values outside the expected range, such as negative percentages, were excluded due to concerns about data quality and interpretability unless explicitly defined in the source codebook to represent missingness or another condition. This method may be applied in the future to create an updated database of resilience indicators or to other geographies as data becomes increasingly available. Collectively, this process generated a database that can be incorporated into future studies to inform public health burden and potential interventions. 1 reflects the source from which we retrieved the data and may not be the original source or curator of the dataset, which are noted in parentheses when identified. Datasets were downloaded between October 2024 and January 2025.To align timeframes across different indicators, for each indicator, the most recent data release year that used the 2010 census tracts was prioritized when available. When data were available for multiple years that used the 2010 census tracts, values were averaged to minimize the missingness of the reporting for a single year. This led to the inclusion of data for a similar timeframe across indicators, as outlined in Table 1.For indicators that were reported using the 2020 census tracts, 2020 Comparability Relationship File Record Layouts available through the US Census Bureau were used to determine values corresponding to the 2010 census based on weighted averages using the land area, the number of households, or population, depending on the metric (detail provided in Table 1) (9). For example, vehicle access was reported per household, and therefore, average vehicle access was calculated by weighting by the number of households in each 2020 census tract. Similarly, metrics reported with units of percent of the population were weighted by population in each 2020 census tract to calculate 2010 values. While indicators that were available at the census tract level were prioritized to reflect the granularity of community resilience, for indicators such as school performance and voter turnout, it was not possible to measure such characteristics within a census tract. Therefore, data available with county-level resolution was sourced, and the same value for a given indicator was applied to each census tract within a county. This approach does not account for within-county variation across such indicators and thus may not reflect true census tract conditions and potentially mask more localized disparities. For ease of interpretation, the directionality of all indicators was standardized, such that a higher level was indicative of increased resilience.The prioritization of indicators for the Resilience Domain of NCMSD was an iterative process that involved QA/QC from all co-authors. After potential indicators were sourced by AS and EJ, the relevance of indicators was independently reviewed by SM, who identified indicators that may not have a directionality associated with increased resilience or that were difficult to interpret for resilience directionality. For example, unpaid labor, a measure of the percentage of the population working as an unpaid family member, was removed from the considered indicators. While unpaid labor could be a source of support to individuals and societal infrastructure, it may also be disproportionately experienced by some community members and may reflect social inequities (14,15). Therefore, it is challenging to interpret associations between resilience and unpaid labor. Various methods were also used to evaluate potential resilience indicators based on the available data. This process included performing basic descriptive statistics and graphing values by county and census tract to understand the range of values and potential outliers and assess data quality. Data quality concerns and relevant processing were discussed with all co-authors to form a consensus surrounding all methods employed in generating the Resilience Domain of NCMSD. Sourced indicators that were not ultimately incorporated within the Resilience Domain of NCMSD are summarized in Supplemental Table S1.All data manipulation and analysis were performed in R Studio using R version 4.4.2. General packages that were used throughout the analysis included tidyverse (16) and dplyr (17) in addition to baseline R packages. An overview of the data processing workflow is shown in Figure 1. Basic statistics were calculated for all selected resilience indicators, including the mean, standard deviation, median, minimum, maximum, interquartile range, 25 th and 75 th percentiles, and percent of missingness. Some datasets had undergone prior processing before integration into NCMSD. For example, missing values in the US CVI were imputed with the median of nationwide data for some indicators. Therefore, the percentage of census tracts that were imputed in NC was also analyzed. Pearson correlation coefficients were calculated between indicators based on pairwise complete observations using the corrplot package, ordering indicators based on the first principal component. All scripts are publicly available on the UNC-SRP GitHub site (18).Health and wellness indicators were extracted from the ACS, ATSDR EJI, CDC/ATSDR SVI, CDC PLACES, EPA EJ Screen, FEMA CRCI, Healthy Communities NC, and US CVI (6-13). These health and wellness measures are reflective of human capital and health vulnerabilities, which are vital to community resilience because they represent the baseline health of individuals who drive response and the capacity of community resources for recovery efforts (1). Selected indicators included health insurance (see Figure 1B), measured as the percent of adults aged 18-64 who have health insurance (2019) (8), the number of medical practitioners per 1,000 people in a census tract (2020) (6), and the number of hospital beds per 10,000 people in a county (2016) (11). Estimates of the percentage of the civilian noninstitutionalized population with a disability and the percentage of individuals reporting not good mental health were reported by EJI for 2015-2019 (12). Access to food was represented as the inverse of percent of people within a county who were food insecure in 2019 (11). Individual health behaviors that affect wellness, including sleep, physical activity, smoking, and binge drinking, were reported by Healthy Communities NC from PLACES data for 2019-2021 (13). These indicators were defined as the percentage of adults who get on average at least 7 hours of sleep per night, who have participated in physical activities in the past month, who have not smoked more than 100 cigarettes in their lifetime, and who have had fewer than five drinks if male or four drinks in female on an occasion in the past 30 days, respectively (8).Social and economic indicators were extracted from the ACS, FEMA CRCI, Healthy Communities NC, myFutureNC, and RWJF (6,9,13,(19)(20)(21). This domain encompasses the networks of local institutions and interpersonal relationships, along with the financial resources and economic opportunities that together support community well-being and development (1). These indicators included the following: school performance, school attendance, and youth engagement (13), social and civic organizations (11), and religious institutions (11), living near a public library (11), economic stability (6, 13), and community equality (6,13,21). School performance was assessed as the percentage of schools not receiving a state-designated status of low-performing (2018-2019) (13). School attendance, defined here as the lack of chronic absenteeism, is the percentage of public-school students who missed less than 10% of school days in a school year (2018) (13), and youth engagement is the percentage of the population aged 16 to 19 enrolled in school or working (2020-2021) (13). Social and civic organizations and religious organizations were defined as the number of organizations per 1,000 people from 2003-2017 (7,11). Library access, defined as living near a public library, was included as the percentage of a census tract that lives within 3 miles of a public library (2017) (7,11). This serves as a measure of community space and access to resources. Economic stability metrics included the percent of people who are employed (2015-2019) (12) and the percent of the workforce employed not in the dominant sector (2020) (6). Housing affordability for renters and homeowners was defined as the percentage of households with housing costs less than 30% of their income (2020-2022) (13). Lastly, measures of community equality included the Gini Index of income inequality, a measure of income distribution across the population (2020) (6), gender pay equity (see Figure 1C) (13), a ratio of women's median earnings to men's median earnings for all full-time, year-round workers (2020-2021) (13, 21), and the residential segregation index of dissimilarity (2020-2022) (13, 21), a measure of the residential segregation between non-White and White county residents (21).Infrastructure indicators evaluated critical aspects related to the adequacy of housing (6, 13); transportation (7, 13); energy (22); access to greenspace (12); and access to the internet (13), smartphones (6), and computers (23). These indicators were sourced from ATSDR EJI, the DOE LEAD Tool, FEMA CRCI, ResilientNC, and US CVI (6,11,12,22). These indicators were included to represent the accessibility of community resources and connectivity, from physically navigating space to having means of communication and access to information. Housing indicators included the percentage of all housing units that were owner-occupied (2020) (6), the percentage of renter and owner-occupied housing units that were not mobile homes (2020) (6), and the percentage of renter and owner-occupied housing units that were built after 1980 (2015-2019) (12). Transportation was measured as the percentage of households without a vehicle (2020) (6), relative walkability (2015-2019) (12), bikeability scores (2022) (11), and public transit performance scores (2019) (11). The affordability of energy was captured as the average percent of median annual income that households paid for electricity and gas bills (2018-2022) (22). As a proxy for access to parks and green space, we selected the proportion of tract area within a 1-mile buffer of green space (2015-2019) (12). Lastly, smartphone, computer, and internet access were measured as the percentage of households with a smartphone (2020) (6), the estimated percent of households with one or more members of the household owning or using a computer (2018-2022) (23), and the average percent of households with internet access (2013-2017) (11), respectively (see Figure 1D).In the final resilience category, the rate of voter turnout in the 2020 presidential election among those eligible to vote (11) was used as an indicator of political capital reflective of enfranchisement and civic engagement (see Figure 1E) (1). This indicator was selected based on the resilience activation framework proposed by Abramson et al. (1). Other metrics, which were considered for this domain, included access to people in leadership or the effectiveness and equity of government infrastructure; however, relevant data for such indicators was not identified at a census or country level tract for NC.Basic summary statistics are summarized for each indicator in Supplemental Table S2. Correlations between indicators by domain are presented in Figure S1, ordered by the first principal component value. Notably, this ordering did not group indicators by resilience category, indicating that there were correlations both within and across resilience categories. For example, physical activity was strongly positively correlated with non-smoking (r = 0.94), health insurance (r = 0.90), internet access (r = 0.84), mental health (r = 0.84), adequate sleep (r = 0.81), and energy affordability (r = 0.80). Moreover, housing affordability, classified as a social and economic indicator, and owner-occupied homes, an infrastructure indicator, were positively correlated (r = 0.72). Other related indicators, such as computer access and smartphone access, were also positively correlated (r = 0.79), as expected. Several of these indicators were sourced from different data sources, further validating the approach used to identify, process, and combine indicators. In general, resilience indicators were positively correlated with other resilience indicators, except for library access, religious organization, and lack of binge drinking. For instance, there were negative correlations between the lack of binge drinking and physical activity (r = -0.74) and internet access (r = -0.68).This dataset represents an important compilation of indicators that can now be used to inform resilience across the state of NC, now poised for integration into NCMSD. While this iteration of the dataset is novel and informative, it is not without limitations. First, while we prioritized data that was available at the census tract geospatial resolutions, for several indicators, due to data availability, the same value was assigned to each census tract in a county; these values may not accurately reflect exposures within a census tract. Moreover, while the resulting database captures the geospatial variation in community resilience indicators, it does not capture temporal variation. Furthermore, the majority of values came from documented measures from the 2010 census tracts, while pulling data from more recent years when needed to assist in data completeness. As data becomes increasingly updated and available from more recent years, we hope to continue to update this resource to continue to inform resilience indicators across the state. In presenting this method of identifying and organizing relevant sources of resilience data, this effort can now be repeated in the future to reflect changes in these data and downstream interpretations that could assist in public health studies and decision-making. Future efforts will also combine additional indicators to capture community resilience as more data become available in the coming years. Funding Support was provided through the National Institutes of Health (NIH) from the National Institute of Environmental Health Sciences, including grant funds (P42ES031007, T32ES007018). Support was additionally provided through pilot funds from the University of North Carolina Gillings School of Global Public Health. 9 Tables Table 1. Resilience indicators by category. Indicators by resilience category, data source, and data processing steps performed for integration within the curated database. Data sources were accessed from October 2024 to January 2025. Within the Description and Processing columns, explanations are included on whether the indicator was inverted, to standardize each indicator such that a higher level was indicative of increased resilience. Data sources shown in parentheses in the last column represent other identified sources of the same data. Specific citations are also added in the Source column when referring to specific sites used to download subsets of the data contained within the larger parent Source mentioned first. However, the listing of alternate sources is not exhaustive, and data retrieved from those sources may have different values compared to the selected data due to differences in data processing performed by different sources.