Children’s Environmental Health (CEH) Systems Model
A conceptual system-of-systems modeling approach for assessing children’s environmental health was previously described in Cohen Hubal, et al. [11], based on the approach described in Little, et al. [3]. Briefly, a supreme orienter, or a goal of managing a complex socioenvironmental problem, is defined. Basic orienters, or contributors to achieving the goal, are then defined to help characterize the supreme orienter. Where basic orienters can be abstract and broad, operational orienters represent a range of more specific goals and concepts encompassed by the basic orienters that can be more easily quantified. Finally, indicators, or associated quantifiable data, are identified that can be compared to the operational orienters to assess progress. The system-of-systems model used for assessing children’s environmental health in this demonstration is shown in Fig. 1.
In this model the desired goal, optimal children’s environmental health in North Carolina counties, is embodied in the following basic orienters: a healthy physical environment, a healthy social environment, and a healthy child or group of children. Operational orienters for a healthy physical environment include clean air, clean water, healthy food, and safe products. Quality of a social environment is a function of the social and economic resources available to children. The operational orienter for a healthy child is the realization of their full potential and developing physically and emotionally within a normal range. Key indicators can be identified and measured to characterize the actual state of these operational orienters such as chemical occurrence in the physical environment orienter, income level in the social environment orienter, or presence of disease in the normal developmental orienter. Indicators like these allow evaluation of the impacts of environmental health decisions and actions on the orienters and on the overall complex system governing children’s environmental health. A set of example indicators for each of the basic orienters were selected to demonstrate this approach as described below and as presented again in Fig. S1.
Physical environment
The three indicators selected to represent chemicals in the physical environment in North Carolina counties were: (1) number of Brownfield locations, (2) number of Superfund sites, and (3) percentage of homes built before 1979. Data on Brownfield locations in each county were obtained from the North Carolina Department of Environmental Quality [12] and were manually recorded in an Excel file. Where counties reported duplicate Brownfield locations with the same name and address, only one location was recorded. Data on the number and locations of Superfund sites were obtained from the Environmental Protection Agency’s National Priorities List [13]. Data were filtered to North Carolina locations and addresses were used to manually match locations to the corresponding counties. The number of Brownfield locations and Superfund sites were summed for each county. Data on the percent of homes built before 1979 in each county were downloaded from PolicyMap [14].
Social environment
The three indicators selected to represent the social environment in North Carolina counties were (1) percentage of residents below 18 years of age living in poverty, (2) percentage of residents under the age of 19 living without health insurance, and (3) percentage of heads of household without a high school diploma. These data were all collected from the Kids Count database through the Annie E. Casey Foundation [15]. Data on percentage of heads of household without a high school diploma were downloaded as 5-year averages from 2010 to 2014. A 5-year average was calculated across years 2011 to 2015 for percent uninsured minors in each county and percent of minors in poverty in each county.
Population size for children ages 0-4, ages 5-9, and ages 10-14 for each county were obtained from the 2019 American Community Survey through Social Explorer [16]. This database also provided a total population size for each county including both children and adults of any age. Percentage of children under age 5 in each county was calculated by dividing the population of children ages 0-4 by the total (children and adult) population.
Health outcomes
The three indicators selected to represent children’s health in North Carolina counties were (1) percentage of low birthweight births, (2) percentage of children under age 2 with elevated blood lead levels (≥ 5 µg), and (3) percentage of children under age 15 who were reported as asthma-related hospital discharges. These data were all obtained from the Kids Count database through the Annie E. Casey Foundation [15]. The blood lead data were downloaded as 5-year averages over years 2014 to 2018. Asthma discharge data for each county were downloaded as 5-year averages over years 2010 to 2014 and were divided by the total population of children (under age 19) in each county. Percentage of low birthweights in each county was downloaded for years 2011 to 2015 and calculated as a 5-year average.
ANOVA analysis
Analysis of variance ((ANOVA) tests are commonly used to help determine which predictor variable data best explain variability in the response data. Here we use ANOVA to investigate potential drivers of variability in children’s health outcome data from physical environment factors, social environmental factors, and other health outcomes. The normality assumption for all data described above was confirmed using QQ-plots. One-way ANOVAs were conducted for each health outcome variable against the remaining variables. The p-values were considered significant if less than 0.05. The ANOVA tests returned p-values for the social, physical, and health outcome variables in relation to one children’s health outcome, which indicates how likely the hypotheses of correlation between these different outcomes are. All calculations described in this and previous sections were performed in R (version 4.0.3).
ToxPi Analysis
To demonstrate how integrating disparate data streams and weighing indicators for complex systems in a semiquantitative fashion can inform environmental health decisions, the ToxPi framework was used as a comparative analysis to extend univariate ANOVA results. All physical environment, social environment, and health data outlined above were loaded into the ToxPi Graphical User Interface (GUI) tool [17] and analyzed within the program’s unique statistical framework [18]. This framework provided a dimensionless index score (ToxPi score) for each county in North Carolina that is the cumulative representation of vulnerability based on the collective values of the respective vulnerability metrics. The ToxPi chart for each county is presented as a unit circle separated into different colored slices that represent each data metric. For each slice, distance from the origin is proportional to the normalized value of the component data points comprising that slice, while the width indicates the relative weight of that slice in the overall ToxPi calculation [18]. In this study, all data metrics were weighted equally, as visualized by equivalent radial widths for each slice.
Knowledge-driven exploration of the individual metrics was facilitated through ToxPi’s ranking and visualization of each of the counties. The ToxPi output was imported into R, where K-means cluster plots (Fig. S3) were used to determine the optimal number of groups for the North Carolina counties. A hierarchical clustering analysis within ToxPi (Fig. S4) was used to confirm the decision to use 5 groups of counties and perform the grouping. The grouped county data were exported from ToxPi to be mapped via the ToxPi*GIS web application (https://toxpi.org/gis/webapp/), using the latitude and longitude coordinates for the center of each North Carolina county. To visualize the relationship between the vulnerability indicators (physical environment, social environment, and health outcomes) and the distribution of children in North Carolina, the ToxPi results were plotted over a base map showing the percent of population under age 5 in each county. The base map was created in ArcMap Online, then transferred into ToxPi*GIS. The hierarchical clusters from the ToxPi analysis were then overlayed on the base map to visualize trends in the social environment, physical environment, and health outcomes across the state.