Development of a Cumulative Stressors and Resiliency Index to Examine Environmental Health Risk: A South Carolina Assessment

Overview

We designed a CSRI that screens for cumulative risk based on the relationship between chemical and nonchemical stressors and resiliency factors that may buffer against environmental exposures, hence creating a more accurate risk calculation. The primary categories in the index are “environmental stressors” and “resiliency buffers,” while subcategories include environmental hazards, environmental exposures, pathogenic factors, and salutogenic factors (inverse) (Fig. 1).

OPEN IN VIEWER

FIG. 1.

Primary and subcategories of the CSRI. ^aPathogenic factors on the lower spectrum create a buffer effect (i.e., low violent crime promotes resiliency compared with high violent crime). ^bTaking the inverse of salutogenic factors meant assigning lower scores to neighborhood assets on the higher end of the spectrum (i.e., adequate or high access to grocery stores) to represent greater resiliency, and higher scores to assets on the lower end of the spectrum (i.e., low access to grocery stores) to reflect low resiliency. CSRI, cumulative stressors and resiliency index.

Approach

The relationship between stressors and resiliency is embedded in our CSRI where we multiply environmental stressors (environmental exposures+environmental hazards) by resiliency buffers (pathogenic factors+inverse of salutogenic factors) to calculate a risk score for each census tract. We used a multiplicative approach for the calculation since a higher prevalence of stressors may increase community susceptibility to the impacts associated with pollution exposures. ¹⁶ Our calculation was modified based on a risk formula used by CalEPA ¹⁷ in its cumulative impact assessment tool: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*} Risk \ = \ Threat \ \times \ Vulnerability \end{align*} \end{document}

Resiliency buffers are used instead of vulnerability since we focus more on individual-level and community-level assets and take into account the impact of common pathogenic factors that may influence vulnerability. The combination of pathogenic factors on the lower spectrum (i.e., low unemployment) and salutogenic factors on the higher spectrum (i.e., high access to grocery stores) allows the resiliency buffers to become protective against environmental exposures and reduces the overall CSRI score.

Indicators

The 26 indicators selected for the CSRI were informed by interviews with community residents and scientific literature known to have a positive or negative association with adverse health outcomes. We created a potential list of CSRI indicators (Table 1) and used principal component analysis (PCA) to systematically construct the final list using SAS Enterprise Guide version 7.1. PCA does not make assumptions regarding the causal structure of variables, but it is a variable reduction procedure used to capture indicators that explain most of the variance among observed variables. Other studies have documented the use of this variable reduction procedure in creating an index. ¹⁸

We performed a separate PCA for all subcategories and entered each variable into the model based on our findings from a previous study where African American residents in North Charleston, SC, prioritized EJ issues impacting their community. ¹⁹ For example, the highest percentage of residents ranking an indicator as an extremely high priority was entered into the PCA model first, followed by the next highest ranking variable within a respective subcategory. Any indicator accounting for >10% of the total variance among observed variables in a subcategory was a candidate for the final CSRI based on the following equation: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*} Proportion = { \frac { Eigenvalue \ for \ the \ component \ of \ interest } { Total \ eigenvalues \ of \ the \ correlation \ matrix } } \end{align*} \end{document}

where the denominator (total eigenvalues of correlation matrix) is equal to the total number of variables (N = 26) being analyzed. ²⁰ The 26 proposed indicators were reduced to the 20 used in the equation to calculate CSRI scores for each census tract (Table 2).

Indicators were assigned a score from 0 to 1 in increments of 0.25 based on a census tract's percentile range and represented a community's rank for each stressor or resiliency factor (Table 3). For example, a census tract with a high mean concentration of diesel particulate matter (PM) would receive a higher score closer to 1, while a tract with a low mean concentration of diesel PM would be ranked closer to 0. Each census tract could receive a score of 0, 0.25, 0.50, 0.75, and 1 that would correspond with percentiles calculated from state estimates.

The “access to grocery stores” indicator was scored as 1 or 0 to represent a low access or nonlow access tract, respectively. For environmental exposures, environmental hazards, and pathogenic factor subcategories, being in the lower percentile represented lower exposures. Regarding salutogenic factors, being in the lower percentile meant a census tract had greater access to community assets such as higher education, more health insured individuals, and was not designated as a Health Professional Shortage Area (HPSA) or was a HPSA with a lower need for primary health professionals in the area.

The environmental hazard indicator scores corresponded with four groups that were developed based on a census tract's distance to a specific hazard. We decided not to use a weighting method for our environmental hazards indicators previously applied in CalEnviroScreen 3.0 ²¹ since we still need to consider how to appropriately weight the resiliency buffers in a way that counteracts environmental hazards. The CSRI scores could range from 0 to 100 (Table 3), where low scores were indicative of low risk/high resiliency and high scores represented high risk (HR)/low resiliency.

Study area

CSRI scores were calculated for census tracts in SC, which is located in the southeastern region of the United States and spans ∼30,061 square miles of rural and urban landscape. SC may be further divided into four Environmental Affairs (EA) regions that provide local support to communities located within their boundaries: (1) Upstate, (2) Midlands, (3) Pee Dee, and (4) Lowcountry. SC has 1103 census tracts that equate to 3059 block groups and 181,901 blocks. ²² Census tracts (15) assigned to bodies of water and those with little or no residential population were excluded from the analysis, ²³ thereby reducing the total number of census tracts to 1088.

The population consists of ∼4,625,364 residents with 153.9 persons/square mile. ²⁴ The population is mostly non-Hispanic white (66.2%), followed by non-Hispanic black (27.9%) and Hispanic/Latino (5.1%). SC's median household income ($45,483) is roughly $8,400 lower than the median income for the United States ($53,889). ²⁴ The percentage of persons below poverty level in SC (16.6%) exceeds the U.S. average (13.5%).

Analysis

CSRI scores were calculated for the 1088 SC census tracts. Indicator data were obtained from various publicly available websites, requested from an agency, or purchased from a vendor. A crosswalk was used to convert all data to the appropriate scale using SAS Enterprise Guide version 7.1. Data sets were linked with SC census tracts and percentiles were calculated based on the raw scores for each indicator (Table 4). The percentiles for each census tract were assigned a value ranging from 0 to 1 that was incorporated into the final CSRI calculation.

Census tracts were assigned a score of 0 if they had a raw score of 0 for the respective indicator, while census tracts with scores in the <25th percentile range (excluding 0) received a score of 0.25. Census tracts in the >25th to 50th percentile were scored as 0.50, those at the >50th to <75th percentile receive a score of 0.75, and any census tract at the >75th percentile range was scored as 1. The salutogenic factors received an inverse indicator score based on the respective raw score. For example, census tracts with adequate or high access to grocery stores received a low indicator score (0) to represent greater resiliency. Census tracts with low access to grocery stores had a higher indicator score (1.00), which contributed to an increase in the overall CSRI score.

The scores assigned to each percentile were summed for both categories and multiplied to compute the final CSRI score. An example calculation for the CSRI is shown in Table 4. Descriptive statistics for mean (M), standard deviation, minimum, and maximum values of CSRI scores were calculated for the state and four EA regions. We calculated CSRI scores in the 90th percentile for the state and EA regions to represent HR tracts. Any census tract with a CSRI score ≥90th percentile for the state was a HR community with high stress/low resiliency. A one-way analysis of variance (ANOVA) was also performed to assess mean differences in CSRI scores by EA region.

Choropleth maps were created using ArcMap 10.5 to illustrate the spatial relationship between CSRI scores (Fig. 2) and percentage of non-white populations (Fig. 3) across the state. Each map contained five classes using natural breaks that depicted variations between the highest and lowest class of scores. Linear regression was used to assess racial/ethnic disparities in cumulative exposures to environmental stressors and resiliency buffers at a significance level of 0.05 (Table 6). Percent non-white was the independent variable and CSRI score was the dependent variable in the model.

OPEN IN VIEWER

FIG. 2.

Choropleth map of CSRI scores by census tract. Fifteen census tracts (marked in gray) were excluded due to their census assignment to bodies of water (9900s) and have little or no residential population (9800s, 45027960202, and 45061920301). CSRI scores range from low risk to high risk based on a community's combined load of environmental stressors and resiliency buffers. The highest class of CSRI scores (43.8–64.0) represents communities in the 90th percentile for risk or those classified as “high risk.”

OPEN IN VIEWER

FIG. 3.

Choropleth map of percent non-white population by census tract. Fifteen census tracts (marked in gray) were excluded due to their census assignment to bodies of water (9900s) and have little or no residential population (9800s, 45027960202, and 45061920301).

Based on our PCA criteria for selecting the final CSRI model, 17 indicators explained ≥10% of the total variance for their respective subcategory (Table 2). Traffic density (4.8%), toxic releases from toxic release inventory facilities (6.4%), and leaking underground storage tanks (3.6%) did not meet the ≥10% variance criteria; however, they were still included in the index due to their impact on health and presence in other EJ assessment tools (i.e., CalEnviroScreen). Five of eight pathogenic indicators met the inclusion criteria, which eliminated linguistic isolation (8.7%), Gini index (5.1%), and low birth weight (3.9%) from the final CSRI. Five of eight salutogenic indicators also met the inclusion criteria, thereby excluding mental health services (8.5%), access to green space (5.4%), and homeownership (4.7%). CSRI scores for each census tract ranged from 7.4 to 64.0 (M = 28.0) (Table 5).

The one-way ANOVA findings yielded a statistically significant difference in CSRI scores by EA region (p < 0.0001) (Table 6). The Tukey's Honestly Significantly Difference (HSD) test calculated more specific regional comparisons, and results indicated significant differences in CSRI scores only applied to certain regions. There was no statistically significant difference in mean CSRI scores when Lowcountry was compared with Pee Dee since the 95% confidence intervals (CIs) [−1.53 to 2.68] overlapped. All other regional comparisons remained significant in the HSD (α = 0.05). The linear regression model containing percent non-white populations and CSRI scores was also statistically significant (α = 0.05; p < 0.0001), where a one-unit increase in the percentage of non-white populations increased CSRI scores by roughly 6.1% at the census tract level (Table 6).

Results from the choropleth maps (Figs. 2 and 3) show the highest scores in areas containing the highest percentage of non-whites. The red areas on the CSRI map represent tracts close to or within the 90th percentile for CSRI scores (≥43.8) in the state. These HR tracts are primarily located in more densely populated counties such as Charleston, Florence, Greenville, Spartanburg, Anderson, and York. The census tract with the highest CSRI score (64.0) and hence the low resiliency was detected in the Upstate region in Greenville County, SC. The tract with the lowest CSRI score and high resiliency (7.4) was located in Berkeley County, SC, in the Lowcountry region. Many HR census tracts were geographically adjacent to tracts with low stress/high resiliency.