Geographic Sensitivity of COVID-19 Outcomes: A Multiscale Analysis of Temporal and Spatial Non-Stationarity in the Contiguous United States
This study demonstrates that county-level COVID-19 disparities in the contiguous United States are driven by temporal and multiscale spatial non-stationarity, particularly a reversal in the urban-rural mortality gradient over time, rather than by fixed socio-environmental vulnerabilities, thereby explaining why cumulative analyses fail to identify consistent predictors.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The geography of disease has long been a central question in public health. For decades, scientists have tried to map why some places suffer more from an illness than others, looking for fixed traits in a community—such as poverty, housing density, or access to healthcare—that might make a location inherently more vulnerable. This approach assumes that a place's risk is a static property, like the color of a house, remaining the same regardless of when an outbreak occurs. However, the COVID-19 pandemic challenged this view. The virus did not strike the entire United States at once; it arrived in different places at different times, and the tools available to fight it changed drastically over three years. Understanding how a location's risk shifts as a crisis evolves, rather than just looking for a single, unchanging cause, is essential for making sense of the pandemic's uneven impact.
A new analysis of the pandemic across the contiguous United States reveals that the story of who got sick and who died is far more dynamic than previous studies suggested. Researchers examined data from 2,754 counties, tracking daily cases and deaths from January 2020 through March 2023. They linked these outcomes to dozens of local factors, including social vulnerability, air quality, and chronic disease rates. The study tested two common assumptions: that a county's characteristics predict its total number of infections and deaths in a stable way, and that these relationships hold true everywhere across the map. The findings overturn the idea that a county carries a fixed level of risk. Instead, the data shows that the factors usually blamed for disparities carried almost no information about the total number of infections a county experienced. When it came to deaths, the only consistent predictor was simply the size of the county's population, but even this relationship flipped completely over the course of the pandemic.
The most striking discovery is how the link between county size and death rates changed direction. In the spring of 2020, when the virus first arrived, larger counties with dense populations faced a much higher risk of death. A county with a population one standard deviation larger than average saw a death rate roughly 29 percent higher than a smaller county. This made sense at the time, as the virus spread rapidly through major cities like New York before vaccines or effective treatments existed. However, as the pandemic progressed through the summer of 2020 and into subsequent waves, this pattern reversed. In every later wave, larger counties actually had lower death rates, ranging from 10 to 24 percent lower than their smaller counterparts. The shift was so complete that the early advantage of small towns and the later advantage of big cities canceled each other out when researchers looked at the entire pandemic as a single block of time. This explains why earlier studies that summed up all the data found weak or confusing results; they were averaging two opposite realities into a single, misleading number.
This reversal was not an artifact of missing data or simple demographics. The researchers adjusted for vaccination rates, which were highly protective in later waves, and they also adjusted for age, since older populations are at higher risk. Even after these adjustments, the pattern held: big cities were hit hardest at the start, and small towns suffered more later on. The study also found that the geographic scale of this effect changed. In the beginning, the high risk in big cities was a very local phenomenon, concentrated in a few coastal metropolitan areas. As the virus became endemic and spread nationwide, the influence of county size expanded to cover the entire country, affecting rural and urban areas alike.
The researchers also tested whether the relationship between local factors and death rates varied across space, a concept known as spatial non-stationarity. They used advanced statistical models to see if the effect of a specific factor, like poverty or air quality, was different in one part of the country than another. They found that most social and environmental factors did not vary by location; their influence was consistent across the nation. The only things that truly varied by place were the baseline risk of death and the effect of population size. This suggests that the idea of a "fixed vulnerability map," where certain types of neighborhoods are always at higher risk, does not fit the reality of the pandemic. Instead, the geography of risk was fluid, shaped by the timing of the virus's arrival and the changing scale of its spread.
Ultimately, this work argues that to understand place-based health risks, we must look at both time and space simultaneously. A single snapshot or a cumulative total cannot capture a phenomenon that reverses itself over three years. The pandemic did not reveal a static landscape of vulnerability; it revealed a dynamic one where the risks associated with being in a large city or a small town changed as the crisis evolved. By recognizing that these relationships are not fixed, public health officials can better anticipate how future outbreaks might move and where they will strike hardest, moving beyond the search for permanent causes to understand the shifting nature of risk itself.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.