Differential contagiousness of respiratory disease across the United States

Abhishek Mallela, Yen Ting Lin, William S. Hlavacek

TL;DR

Researchers at Los Alamos National Laboratory have demonstrated that the initial contagiousness ($R_0$) of COVID-19 was not a single value across the United States, but varied nearly four-fold between different cities. By analyzing 280 metropolitan areas, the study reveals that local population characteristics—not just the virus itself—drastically dictate how fast a respiratory disease spreads.

What Problem Does This Paper Address?

The basic reproduction number, $R_0$, represents the average number of secondary infections produced by a single infected person in a completely susceptible population. While often discussed as a fixed property of a virus (e.g., "the $R_0$ of Measles is 12-18"), $R_0$ is actually a product of both the pathogen's biology and the host population's environment, behavior, and structure.

Most early pandemic models used broad, national-level estimates for $R_0$, which can mask critical local differences. This paper addresses the need for high-resolution, region-specific estimates to understand why some urban areas are inherently more "combustible" for disease spread than others.

Methods

The authors employed a mechanistic compartmental model—a mathematical framework that moves populations through states like Susceptible, Exposed, Infectious, and Recovered. To calibrate these models for 280 different Metropolitan Statistical Areas (MSAs), they used:

  • Bayesian Inference: A statistical method that updates the probability for a hypothesis as more evidence or information becomes available.
  • Adaptive Markov Chain Monte Carlo (MCMC): A sophisticated sampling algorithm used to estimate the "contact rate" parameter for each specific region while keeping other biological parameters (like the incubation period) fixed.
  • Variability Analysis: Before modeling, the team used the Fano factor, Gini coefficient, and Wasserstein-1 distance to prove that MSAs are the most appropriate unit of study. They found that counties within the same MSA are much more homogeneous in their infection risk than counties within the same state.

Datasets

The study relied on publicly available surveillance data from the early stages of the pandemic (January to June 2020) to ensure they were measuring contagiousness in a "naïve" population without significant prior immunity or widespread vaccination.

  • The New York Times COVID-19 Data: Daily county-level reports of new cases.
  • U.S. Census Bureau: Definitions for the 384 Metropolitan Statistical Areas.
  • The COVID Tracking Project: National-level diagnostic testing totals.

Key Insights and Ideas

  • The "Goldilocks" Scale: The researchers found that states are too diverse to model as a single unit, while individual counties produce data that is too "noisy" due to small sample sizes and reporting irregularities. MSAs represent the ideal scale for capturing socio-economic and transmission patterns.
  • Place Matters as Much as the Pathogen: Because the virus was the same across the country, the massive variation in $R_0$ must be attributed to "population properties." These include biological, socio-behavioral, and environmental factors unique to each city.
  • Quantifying Susceptibility: The study provides a "rank order" of susceptibility. This allows public health officials to identify which cities might need more aggressive early interventions in future outbreaks based on their inherent transmission potential.

Key Results

  • Wide Variation: The estimated $R_0$ values ranged from a low of 1.9 (Appleton, WI) to a high of 7.7 (Detroit, MI).
  • Broad Coverage: The 280 MSAs analyzed account for 95% of the U.S. urban population and 82% of the total U.S. population.
  • Model Consistency: The growth rate estimates derived from the model were highly consistent with early surveillance data, validating the Bayesian approach.
  • Detroit as an Outlier: The exceptionally high $R_0$ in Detroit (7.7) suggests that certain urban environments can facilitate explosive spread far beyond what national averages would suggest.

Limitations and Future Work

  • The "Black Box" of Population Features: While the paper proves that population properties cause $R_0$ to vary, it does not pinpoint which specific factors (e.g., population density, public transit usage, housing types, or age demographics) are the primary drivers.
  • Temporal Changes: The authors acknowledge that population properties are not static; behaviors and environmental factors change over time, meaning these $R_0$ "signatures" might evolve.
  • Data Quality: The study is limited by the accuracy of early-pandemic surveillance data, which was often hampered by limited testing availability.
  • Future Direction: The next logical step is to correlate these $R_0$ values with specific census-based socio-economic variables to build a predictive model for future respiratory threats.

Summary updated

Abstract

The initial contagiousness of a communicable disease within a given population is quantified by the basic reproduction number, R<sub>0</sub>. This number depends on both pathogen and population properties. On the basis of compartmental models that reproduce Coronavirus Disease 2019 (COVID-19) surveillance data, we used Bayesian inference and the next-generation matrix approach to estimate region-specific R<sub>0</sub> values for 280 of 384 metropolitan statistical areas (MSAs) in the United States (US), which account for 95% of the US population living in urban areas and 82% of the total population. We focused on MSA populations after finding that these populations were more uniformly impacted by COVID-19 than state populations. Our maximum a posteriori (MAP) estimates for R<sub>0</sub> range from 1.9 to 7.7 and quantify the relative susceptibilities of regional populations to spread of respiratory diseases. ONE-SENTENCE SUMMARY: Initial contagiousness of Coronavirus Disease 2019 varied over a 4-fold range across urban areas of the United States.
BibTeX
@inproceedings{mallela2023differential,
  author = {Abhishek Mallela and Yen Ting Lin and William S. Hlavacek},
  title = {{Differential contagiousness of respiratory disease across the United States}},
  booktitle = {Epidemics},
  year = {2023},
  doi = {10.1016/j.epidem.2023.100718},
}

Loading related content...

Figure