Cumulative Environmental Justice Burden Across U.S. Counties

Framing (non-negotiable). The unit of analysis is the U.S. county (5-digit FIPS), across 3134 counties spanning 50 states + DC (facility, PFAS, and geospatial axes cover the 48 contiguous states + DC, the extent of the spatial backbone). All results are observational county-level associations assembled by integrating independent knowledge graphs on a shared geographic key — hypothesis-generating, not causal or individual-level inference. Evidence types are kept separate and never merged into a single confidence score; the headline ranking is a consensus count of how many independent burden domains flag a county. This framing caveat travels with every downstream claim.

Abbreviations. OKN = Open Knowledge Network; KG = knowledge graph; FIPS = Federal Information Processing Standards county code; S2 = S2 discrete global geospatial grid; CHR = County Health Rankings; SVI = CDC/ATSDR Social Vulnerability Index; SDoH = social determinants of health; PLACES = CDC Population Level Analysis and Community Estimates; EPA FRS = EPA Facility Registry Service; NAICS = North American Industry Classification System; PFAS = per- and polyfluoroalkyl substances; PFOS/PFOA = perfluorooctane sulfonic / octanoic acid; CAS = Chemical Abstracts Service registry number; ToxCast = EPA high-throughput toxicity screening; AOP = Adverse Outcome Pathway; MIE = molecular initiating event; PPAR = peroxisome proliferator-activated receptor; NAFLD = non-alcoholic fatty liver disease; SAMHSA = Substance Abuse and Mental Health Services Administration; ρ = Spearman rank correlation.

1. Executive summary

Integrating eight Proto-OKN knowledge graphs on the shared county-FIPS key, we built a per-county profile of cumulative environmental and social burden for 3134 U.S. counties, spanning six independent burden domains — industrial pollution sources, chemical/PFAS exposure, ambient environmental quality, socioeconomic vulnerability, public safety, and health outcomes — plus a service-access domain. Rather than collapse these into one index, we grade each county by a consensus score: the number of independent domains (0–6) in which it falls in the national worst quintile.

Cumulative burden is spatially concentrated. 25 counties are Tier A (very high; 5–6 corroborating domains) and 470 more are Tier B (high; 3–4 domains) — together 15.8% of counties. The two counties flagged on all six domains are Wayne County, Michigan (Detroit) and Caddo Parish, Louisiana (Shreveport). The heaviest cluster is the Lower Mississippi Valley / Deep South: Louisiana averages 3.17 corroborating domains per county (46/64 counties high-burden) and Mississippi flags 51/82, alongside legacy-industrial cities (St. Louis, Baltimore) and dense-industrial metros in New Jersey and California.

The greatest burden↔service mismatch — high cumulative burden paired with the weakest healthcare access — falls on rural, high-poverty Black Belt counties: Macon County, Georgia, Wilcox County, Alabama, and Quitman County, Mississippi. Socioeconomic vulnerability and health-outcome burden are tightly coupled (ρ = 0.765), whereas raw industrial-facility counts are decoupled from — even inversely related to — social and public-safety burden (ρ = -0.511), a county-scale count-vs-rate signal we flag explicitly.

What this adds: a reproducible, federated synthesis that corroborates place-based burden across genuinely independent data pipelines (EPA facilities, CDC/CHR health-and-social data, SAMHSA services, SCALES federal courts, SAWGraph PFAS, EPA ToxCast/AOP-Wiki toxicology) on one geographic key — surfacing not just where burden is highest but where corroboration is strongest and services are scarcest.

2. Sources used

Every KG below was queried directly (logged SPARQL); each row traces to at least one query in the reproducibility record. Join keys are the verified federation crosswalks.

3. Design & rules

We treat cumulative burden as corroboration across independent data sources, not a single weighted index. Each of 3134 counties was profiled on 31 indicators drawn from the eight KGs and grouped into seven domains (six burden domains D1–D6 plus a service-access domain D7). Indicators that are rates (County Health Rankings / CDC SVI percentages and rates, CDC PLACES disease prevalence) are used as-is; indicators that are counts (EPA facilities, PFAS facilities, federal cases, PFAS samples) are used as absolute-exposure signals and interpreted with a population caveat (§10). Every indicator carries its evidence type, source KG, geographic level, and direction in a separate evidence table (data/evidence_long.csv), preserving evidence types rather than fusing them.

For each domain we standardise its indicators (direction-adjusted z-scores so higher = more burden), average the available ones into a domain index, and flag a county high-burden in that domain if its domain index sits in the national worst quintile (≥80th percentile). The consensus score is the number of the six burden domains (D1–D6) flagged; the service-scarcity flag (D7) is kept separate for the mismatch analysis. Disease prevalence, reported by CDC PLACES at Census-place level, is rolled up to county as a population-weighted mean via spoke-okn's PARTOF_LpL place→county edges. Full replicator specification (exact indicator lists, thresholds, join recipes, and the mismatch formula) is in the reproducibility file.

The federation gives near-complete county coverage on the national axes (spoke-okn, fiokg, scales) and progressively sparser coverage on the regional (sawgraph) and facility-directory (ruralkg) axes — a coverage gradient carried into every downstream flag.

4. Confidence tiers

Counties are tiered by how many independent burden domains corroborate high burden:

Only 2 counties are flagged on all six domains and 23 on five; the tail is genuinely selective. 48 counties are double-burden — flagged in ≥2 environmental domains and ≥2 social/health domains — the counties where environmental and social disadvantage most clearly co-locate.

The distribution is right-skewed: most counties flag 0–2 domains, and the high-burden tail (≥3 domains) is the analytic focus.

5. Findings by axis

5.1 Primary ranking — cumulative-burden counties

Ranking by consensus score (ties broken by the composite burden index) puts Wayne County, Michigan and Caddo Parish, Louisiana at the top — each flagged on all six domains — followed by St. Louis city, Baltimore city, and a dense run of Lower Mississippi Valley parishes and Deep South counties.

Almost every top county flags both environmental and social domains (E and S each ≥2), confirming these are genuine double-burden places rather than counties extreme on a single axis.

5.2 Geographic distribution and hot-spots

Burden is regionally structured. Aggregating to the state level, the Lower Mississippi Valley and Deep South dominate: Louisiana (mean 3.17 domains/county), Mississippi, Arkansas, Oklahoma, and South Carolina carry the highest mean county burden, with secondary clusters in the industrial Northeast (New Jersey) and California's Central Valley / South Coast.

The maps make the concentration visible: a contiguous belt of high-burden counties runs from east Texas and Louisiana up the Mississippi and across the Black Belt of Alabama, Mississippi, and Georgia, punctuated by isolated high-burden metros (Wayne MI, St. Louis, Baltimore, Essex NJ).

5.3 How the domains relate

Because we keep domains separate, we can measure how they co-vary. Socioeconomic vulnerability and adverse health outcomes are strongly coupled (ρ = 0.765), and public-safety burden tracks both (ρ = 0.618 with health). Chemical exposure co-occurs with industrial facilities (ρ = 0.449). Strikingly, industrial-facility density is negatively correlated with public-safety burden (ρ = -0.511) and with service scarcity (ρ = -0.455) — facility counts peak in populous metros that also have more providers, while the worst social/health/safety rates concentrate in poor rural counties (see §7).

The social cluster (D4–D6) is internally coherent and correlates with service scarcity (D7), whereas the environmental-source cluster (D1) sits apart — the core tension this study surfaces.

6. Domain analyses

6.1 Burden↔service mismatch (environmental justice priority)

The environmental-justice question is not only where is burden highest but where is burden highest and help scarcest. We define a mismatch index = (mean burden-domain z) − (service-capacity z), where service capacity aggregates primary-care and mental-health provider availability (CHR) and SAMHSA treatment-provider density (ruralkg). Among counties with consensus ≥ 4, the mismatch peaks in rural, high-poverty counties of the Black Belt and Appalachian fringe.

These counties — Macon County, Georgia, Wilcox County, Alabama, Quitman County, Mississippi, and peers — combine top-quintile cumulative burden with bottom-tier provider access; they are the highest-priority candidates for service investment.

6.2 Chemical exposure → toxicology mechanism (why PFAS matters)

To connect place-based chemical exposure to a biological mechanism, we chained the PFAS measured in SAWGraph water samples to their toxicological profiles. 32 of SAWGraph's measured PFAS carry EPA ToxCast high-throughput bioactivity: PFOS (CAS 1763-23-1) hits 1510 assay endpoints and PFOA (335-67-1) 1396. Those same compounds anchor a curated Adverse Outcome Pathway.

AOP-Wiki AOP 529 spells out the mechanism these bioactivities imply: the molecular initiating event is a stressor binding PPAR isoforms, proceeding through disrupted PPAR nuclear signalling → dysregulated PPAR-network transcription → decreased mitochondrial fatty-acid β-oxidation and triglyceride/fatty-acid accumulation → the adverse outcome, increased liver steatosis. This is a KG-federated exposure→mechanism→outcome bridge: SAWGraph (where measured) → ToxCast (what it perturbs) → AOP-Wiki (what disease it drives). It is a hypothesis-generating link, not evidence that any specific county's contamination caused disease.

6.3 Declared coverage — analyses run vs. deliberately scoped out

To avoid a silent half-completion, we state which candidate axes were run and which were scoped out: Run — industrial facilities (fiokg), PFAS facilities (fiokg), chemical diversity (spoke-okn), measured PFAS (sawgraph), ambient PM2.5 & drinking-water violations (CHR), socioeconomic/SVI (CHR), public safety (CHR + scales), health outcomes (CHR + CDC PLACES), service access (CHR + ruralkg), and the PFAS toxicology chain (biobricks-toxcast + aopwiki). Scoped out with reason — urban flooding (ufokn): the national S2 join exceeds the endpoint's operation budget (verified HTTP 429), so it is a regional-only layer not integrated into the national ranking; water/hydrology (geoconnex, hydrologykg) and soil carbon (sockg): available on the county/S2 hub but regionally scoped (hydrologykg ≈ Illinois, sockg multi-state) and not national; climate (climatemodelskg): joins only ~947 counties (30% coverage), too partial for a national quintile flag; neighborhood justice (nikg): resolves to only 2 counties at the county-FIPS level (a data-model gap), so it cannot contribute a county axis. Each is a declared scope decision, not an omission.

7. Discussion

The federation tells a two-part story. First, cumulative environmental-and-social burden is real, corroborated, and concentrated: 15.8% of counties reach Tier A/B, and the worst — Wayne County, Michigan, Caddo Parish, Louisiana, St. Louis, Baltimore, and the Lower Mississippi Valley belt — are flagged independently by EPA facility data, CDC health-and-social data, and (regionally) measured PFAS, so the ranking does not rest on any single pipeline. Second, the domains that most tightly co-locate are the social ones: socioeconomic vulnerability, adverse health outcomes, and public safety form a coherent cluster (ρ up to 0.765) that also predicts service scarcity — the classic environmental-justice double bind of high need and low capacity, sharpest in the rural Black Belt (§6.1).

The decoupling of industrial-facility counts from social/health burden (ρ = -0.511) deserves care. At the county scale, a raw facility count is partly a population count: large metros host the most EPA-regulated facilities and the most providers, so a count-based pollution-source axis pulls toward well-resourced metros, while rate-based social/health axes pull toward poor rural counties. This is a measurement artifact as much as a finding, and it argues for sub-county (tract/block-group) analysis with population-normalised source density to test whether facility burden truly falls on vulnerable populations — the well-documented pattern our county-count metric cannot resolve. It is the study's central testable prediction.

Actionable implications, flagged by evidence strength: (a) the Tier-A double-burden counties are priorities for coordinated environmental and health intervention; (b) the high-mismatch Black Belt counties are priorities for service investment specifically; (c) the SAWGraph→ToxCast→AOP-Wiki chain identifies PFOS/PFOA-driven hepatic outcomes as a mechanistically-plausible surveillance target where PFAS is measured.

8. Comparison with prior work

According to PubMed, we compared each headline finding against the primary literature. The per-claim record with citations is in Environmental-Justice_literature_comparison.md; every [n] below resolves to §12.

Concordance was assessed against PubMed abstracts and indexed metadata; no claim required full-text retrieval beyond the abstract for the verdict shown. Where the KG evidence diverges from the literature: the only divergence is Claim 6, and it is a scope/measurement difference (county-level counts vs. sub-county population-normalised density), not a graph error — the prior work operates at a finer spatial grain than the county unit this federation joins on. No claim was contradicted.

9. Full ranked results

The complete ranked table of all 3134 counties — every domain index, consensus score, tier, mismatch index, and the contributing source KGs — is in Environmental-Justice_results.xlsx (sheet Ranked results) and data/master_county.csv. The interactive table below is sortable (click a header), filterable (search box + pull-downs for tier, state, and service-scarcity), and paginated; the sources (n) column shows how many federation KGs corroborate each county, with a pill per source (spoke-okn = health/SDoH/exposure, fiokg = facilities, scales = federal caseload, ruralkg = services, sawgraph = measured PFAS).

The ranking shows the Tier-A counties are broadly corroborated (5+ sources) rather than artifacts of a single graph, and lets a reader isolate, e.g., high-burden service-scarce counties in a single state for targeted follow-up.

10. Summary of findings & limitations

Findings recap. Across 3134 U.S. counties, cumulative environmental-and-social burden is concentrated: 25 Tier-A and 470 Tier-B counties (15.8% combined), led by Wayne County, Michigan and Caddo Parish, Louisiana (all six domains) and a Lower Mississippi Valley / Deep South belt (Louisiana 46/64 high-burden). Socioeconomic and health burden are tightly coupled (ρ = 0.765) and predict service scarcity; the sharpest burden↔service mismatch falls on rural Black Belt counties (Macon County, Georgia, Wilcox County, Alabama, Quitman County, Mississippi). A federated exposure→mechanism chain links measured PFAS (PFOS 1510 ToxCast endpoints) through PPAR signalling to liver steatosis (AOP 529).

Limitations.

1. Observational and ecological. All associations are county-level; nothing here supports individual-level or causal claims, and county aggregates mask within-county (tract/neighborhood) disparities that are central to environmental justice.
2. Count vs. rate. Facility, PFAS-facility, federal-case, and provider counts scale with population, biasing count-based domains toward large metros; rate-based domains (CHR/SVI/PLACES) do not. The industrial-source↔social-burden decoupling (§7) is partly this artifact. Federal caseload (scales) is reported as a separate justice-activity indicator and deliberately excluded from the consensus domains for this reason.
3. Uneven coverage. SAWGraph PFAS is Maine-centric (264 counties nationally; top Aroostook County, Maine, 15241 samples), ruralkg providers are a partial SAMHSA directory, and facility/PFAS/geospatial axes cover only the 48 contiguous states + DC. Absence of data is not absence of burden.
4. Threshold sensitivity. The worst-quintile flag and the A/B/C tier cuts are analyst choices; counties near a threshold can move tiers under a different cutoff. The composite burden index is provided for continuous ranking.
5. Provenance heterogeneity. Many indicators derive from the County Health Rankings pipeline within spoke-okn and are therefore correlated by construction; cross-KG independence is strongest between spoke-okn, fiokg, scales, ruralkg, and sawgraph, and that is where "consensus" is most meaningful.
6. Scoped-out layers. Flooding (ufokn), water/hydrology (geoconnex/hydrologykg), soil (sockg), and climate (climatemodelskg) were not integrated into the national ranking (§6.3) for endpoint-scale or coverage reasons; a fuller model would add them regionally.
7. Mechanism is generative. The PFAS→PPAR→steatosis chain (AOP 529) is a plausibility bridge from measured exposure to a documented pathway, not evidence of disease causation in any county.

11. Reproducibility

Everything needed to replicate this analysis — the originating prompt, the replicator specification (indicator lists, thresholds, join recipes, mismatch formula, verified quantities), every supporting SPARQL query verbatim with its row count, pinned KG versions, and timing — is in Environmental-Justice_reproducibility.md, with the analysis scripts in scripts/ and intermediate extracts in data/.

12. References

Retrieved via the PubMed MCP connector.

1. Yang W, et al. PPARα/ACOX1 as a novel target for hepatic lipid metabolism disorders induced by per- and polyfluoroalkyl substances: An integrated approach. Environ Int. 2023. PMID:37572494 · doi:10.1016/j.envint.2023.108138
2. Tompach MC, et al. Comparing the effects of developmental exposure to alpha lipoic acid (ALA) and perfluorooctanesulfonic acid (PFOS) in zebrafish (Danio rerio). Food Chem Toxicol. 2024. PMID:38432440 · doi:10.1016/j.fct.2024.114560
3. Shrestha R, et al. Environmental Health Related Socio-Spatial Inequalities: Identifying "Hotspots" of Environmental Burdens and Social Vulnerability. Int J Environ Res Public Health. 2016. PMID:27409625 · doi:10.3390/ijerph13070691
4. Habran S, et al. Development of a spatial web tool to identify hotspots of environmental burdens in Wallonia (Belgium). Environ Sci Pollut Res Int. 2019. PMID:30725260 · doi:10.1007/s11356-019-04418-5
5. Varshavsky JR, et al. Current practice and recommendations for advancing how human variability and susceptibility are considered in chemical risk assessment. Environ Health. 2023. PMID:36635753 · doi:10.1186/s12940-022-00940-1
6. Wiese LAK, et al. Global rural health disparities in Alzheimer's disease and related dementias: State of the science. Alzheimers Dement. 2023. PMID:37218539 · doi:10.1002/alz.13104
7. Querdibitty CD, et al. Geographic and social economic disparities in the risk of exposure to ambient air respiratory toxicants at Oklahoma licensed early care and education facilities. Environ Res. 2022. PMID:36462693 · doi:10.1016/j.envres.2022.114975