Scaling COVID-19 rates with population size in the United States

Introduction

The rapid global spread and persistence of the SARS-CoV-2 virus and disease (COVID-19) pandemic highlights the speed of interaction and magnitude of increasingly connected human populations in the twenty-first century. One of the key challenges to managing COVID-19 has been consistent and coordinated policy that controls the spatial and/or temporal dynamics of the virus in human populations (1, 2). Paramount to understanding how human infectious disease dynamics operate over space and time is an understanding of human population sizes and fluxes with respect to modern cities (3, 4).

Urban Scaling Theory (UST) is a general quantitative framework that provides predictions of how various physical, socio-economic, and biological attributes, including infectious disease, in modern cities change with population size as a power-law function (5, 6). Early analysis on the scaling of infectious disease comes from data on AIDS cases in the United States between 2002 and 2003, which showed a superlinear scaling exponent where larger cities demonstrated disproportionately more cases (6). Subsequent scaling analysis of AIDS cases in Brazil also showed superlinear scaling exponents in the years 2000 and 2010 (7). These results support the broader prediction that cities with larger population sizes are disproportionately more affected by disease than those with smaller population sizes likely due to a higher total number of contacts and speed of social interactions (8).

More recently, UST has been used to study the dynamics of COVID-19. Initial scaling research on COVID-19 in the United States demonstrated superlinear scaling of early pandemic case growth rates with city population size (9). Subsequent work in Brazil (10) and England and Wales (11) showed a temporal change in scaling exponents across the pandemic. At present, and given the United States initially represented a disproportionately large fraction of global COVID-19 cases and deaths (12, 13), we apply UST to COVID-19 cases, deaths, hospital beds (as a proxy for medical infrastructure), and younger and older demographic cohorts in the United States to characterize and predict how scaling parameters change with population size across the first two years (2020–2022) of the pandemic (see Materials and Methods).

Results

Fig. 1 illustrates the time-dependent changes in scaling exponents (β, panel A) and normalization constants (Y₀, panel B) for COVID-19 cases and deaths as a function of county population size within and across each variant. On the whole, β values for cases initially increase to a maximum for each variant (original: 1.3, 95% CI [1.18, 1.43], day 100; delta: 0.95, 95% CI [0.92, 0.98], day 567; omicron: 1.16, 95% CI [1.14, 1.18], day 720), and then slowly decrease thereafter. β values for deaths follow a similar pattern (original: 1.01, 95% CI [0.97, 1.06], day 200; delta: 0.65, 95% CI [0.58, 0.73], day 577; omicron: 0.86, 95% CI [0.82, 0.91], day 755). The analysis demonstrated a sublinear scaling between county population size and the county’s population aged 60 and older (β = 0.93 ± 0.0043 (SE); Fig. 2A), and linear scaling between county population size and the county’s population that is younger than 60 (β = 1.02 ± 0.0013 (SE); Fig. 2B). The scaling relationship between hospital intensive care unit (ICU) beds in a county and its population size was sublinear (β = 0.96 ± 0.014 (SE); Fig. 2C).

• Download figure
• Open in new tab

Fig. 1.

Allometric scaling behavior over time. A) Time-series change of scaling exponent (β) of COVID-19 cases and deaths. Dotted horizontal line indicates isometric scaling. B) Time-series change of normalization constant (Y₀) of COVID-19 cases and deaths. Note: dashed vertical line in each panel indicates the start of vaccinations in the US (day 328).

• Download figure
• Open in new tab

Fig. 2.

Allometric scaling of demographic cohorts and medical infrastructure. A) Smaller population counties have proportionally more older adults (β < 1) than larger counties. B) Younger populations are constant across counties (β ∼ 1). C) Smaller-county populations have disproportionately more IUC beds (β < 1). Frequency distributions and their mean values (dotted line) are shown to the right of each panel (A–C).

Discussion

We show how allometric scaling relationships of COVID-19 cases, deaths, younger and older demographic cohorts, and medical infrastructure change with the United States’ county-level population size over the pandemic’s first, delta, and omicron waves. A key result is that the burden of cases disproportionately affects larger-sized counties, while the burden of deaths disproportionately impacts smaller-sized counties. This is demonstrated in the observation that the scaling exponent for cases is predominantly superlinear (> 1) for each variant, with a trend towards isometry (= 1). In contrast, the scaling exponent for deaths is consistently sublinear (< 1) for each variant. The exception to this observation is that delta cases remain sublinear over the analysis.

This larger pattern of superlinear cases and sublinear deaths may be due, in part, to a higher proportion of older adults who live in smaller-population counties. Policy and management strategies in the United States should consider how to scale medical care (e.g., hospital beds, medical professionals, antiviral medication, vaccinations) to address these differences. For example, differentially distributing medical and non-medical interventions to smaller counties may prevent the deaths of older individuals (14). Additional research may reveal differences in protective health measures in counties of different sizes, which may be related to, for example, past experience of disease burden (15). It is imperative that research in urban scaling delve deeper into the multiple mechanisms influencing epidemiological trajectories. Doing so can significantly amplify its relevance and effectiveness in ongoing and prospective epidemiological research and interventions.

Materials and Methods