Live Report: View the Interactive Analysis
This analysis examines county-level COVID-19 mortality across 2,904 US counties from March 2020 through September 2022, with a focus on identifying the structural conditions that predicted which communities bore disproportionate and persistent pandemic burden.
The central finding complicates the dominant narrative of the pandemic response: after controlling for age structure, racial composition, poverty, and household density, the proportion of adults without a high school diploma was among the two strongest independent predictors of county COVID death rates, statistically indistinguishable from elderly population share. Each 10 percentage point increase in adults without a high school diploma was associated with approximately a 51% higher death rate (IRR 1.51, 95% CI 1.44–1.58).
Black population share was not a statistically significant independent predictor (IRR 0.993, p = 0.376) once structural conditions were controlled for — indicating that the severe mortality disparities observed in majority Black communities reflect structural inequity rather than demographic risk. This null result held identically across both model specifications tested. —
1. Excess Mortality Is Geographically Concentrated Just 14 counties (0.5%) reached extreme classification exceeding three standard deviations above the national mean. These counties are disproportionately small, rural, majority Black, and located in the Deep South — particularly Georgia, Mississippi, and Louisiana.
2. Pandemic Burden Reorganized Completely After Wave 1 Spearman rank correlation between Wave 1 and Delta county death rate rankings was effectively zero (ρ = -0.003). The communities that bore the greatest initial burden achieved relative protection by Delta through prior exposure immunity, while burden shifted to structurally disadvantaged rural communities with lower prior exposure and less healthcare infrastructure.
3. Cases Became Less Lethal After Omicron The slope of the case rate to death rate relationship declined by 85% between the pre-Omicron and Omicron/post-Omicron periods (0.0089 vs 0.0013), consistent with variant attenuation, accumulated population immunity, and changes in death attribution practices.
4. Structural Conditions, Not Demographics, Predicted Persistent Burden 157 counties remained in the top quartile of death rates across both Wave 1 and the Delta wave. These persistently high burden counties had nearly three times the national average Black population share (22.9% vs 8.4%), substantially higher poverty rates (19.5% vs 15.3%), and lower educational attainment (12.2% vs 8.9% without a high school diploma) — but were not distinguished by age structure (18.4% vs 18.9% age 65+).
5. COVID Mortality Followed Consistent Winter Seasonality STL decomposition reveals a repeating annual seasonal component with winter peaks exceeding summer troughs by approximately 5 deaths per 100k — a magnitude comparable to the underlying trend. Seasonality amplified existing structural disadvantage consistently across all three years of the analysis period.
| Component | Detail |
|---|---|
| Primary Data Source | Google COVID-19 Open Data (covid19_open_data) |
| Demographic Data | US Census Bureau ACS 2019 5-Year Estimates |
| Geographic Unit | US County (FIPS code) |
| Analysis Period | March 2020 – September 2022 |
| Counties Analyzed | 2,904 of approximately 3,143 US counties (92%) |
| ETL Platform | Google BigQuery (SQL) |
| Analysis Platform | R Markdown (POSIT Cloud) |
| Statistical Model | Negative Binomial Regression |
| Seasonal Analysis | STL Decomposition via Loess |
| Outlier Detection | Funnel Plot with 95% and 99.8% Control Limits |
Six states have zero county-level coverage in the Google covid19_open_data pipeline (Alabama, Arkansas, Arizona, California, Colorado, and Connecticut). Demographic comparison of missing vs included counties indicates the coverage gap is less directionally biased than initially characterized — missing counties skew toward small rural white populations rather than the high-poverty majority minority communities driving the core findings. Core findings are assessed as robust to the current data gap. Complete national coverage is a Phase 2 priority.
The analytical data file (county_weekly.csv, 149MB) is excluded from this repository due to GitHub file size limits. It is generated from the BigQuery ETL documented in the R Markdown source file.
structural-health-inequity-covid/
│
├── index.html # Live interactive report
├── covid_county_analysis_final.Rmd # Fully reproducible source
├── about/
│ └── index.md # Project background and methodology
├── .gitignore # Excludes county_weekly.csv
└── README.md # This file
| Priority | Data Source | Analytical Purpose |
|---|---|---|
| 1 | USAFacts + CDC Wonder | Complete national coverage, extend to May 2023 |
| 2 | CDC PLACES | Clinical comorbidity covariates (diabetes, obesity, COPD) |
| 3 | CDC Vaccination Data | Partition prior infection vs vaccination protection |
| 4 | CMS Medicaid/Medicare | Healthcare access as predictor of persistent burden |
| 5 | HRSA Shortage Areas | Pre-pandemic healthcare desert identification |
| 6 | Indian Health Service | Native American community analysis (McKinley County NM) |
SQL / ETL: Google BigQuery Standard SQL
R Packages: tidyverse, bigrquery, MASS, lme4, broom, forecast, tsibble, feasts, ggplot2, plotly, leaflet, tigris, sf, ggridges, patchwork, viridis, kableExtra
Brian E. Ketchens
Data Analytics | Public Health Equity
GitHub
If you use or reference this analysis please cite as:
Ketchens, B.E. (2025). Structural Health Inequity During Pandemics: A County-Level COVID-19 Mortality Analysis. https://brianketchens.github.io/structural-health-inequity-covid/
Analysis period: March 2020 – September 2022
Phase 2 data integration in progress
Last updated: May 2025