Incorporating county-level NO2 data into a metapopulation model: a case study of the first COVID-19 wave in Oklahoma

Technical Summary: Incorporating County-Level NO2 Data into a Metapopulation Model for Oklahoma's First COVID-19 Wave

Problem Statement
While epidemiological studies have established associations between Nitrogen dioxide (NO2) exposure and increased susceptibility to respiratory infectious diseases, and biological evidence suggests NO2 heightens vulnerability to infection, there is a scarcity of mechanistic epidemic models that explicitly integrate air pollutant data. Existing models often fail to account for how environmental factors directly modulate transmission dynamics alongside human mobility.

Methodology
The authors introduce a novel NO2- and mobility-informed metapopulation model to analyze the first wave of the COVID-19 pandemic in Oklahoma. The methodology integrates two primary data streams:

1. Environmental Data: County-level NO2 concentrations are incorporated directly into the infection rates via a first-order parameterization.
2. Mobility Data: High-resolution mobility patterns are used to capture heterogeneous inter-county connectivity.

The study employs numerical simulations to explore how varying levels of environmental sensitivity affect outbreak trajectories. Furthermore, the model is fitted to county-level COVID-19 incidence data and rigorously compared against two baselines:

• A classic SIR model fitted independently for each county.
• A mobility-only model where NO2 is removed and infection rates are assumed homogeneous across counties.

Performance is evaluated using standard error metrics, specifically normalized Root Mean Square Error (nRMSE), across different epidemic stages: surge, peak, post-peak, and decline.

Key Results

• Simulation Dynamics: Numerical simulations indicate that higher environmental sensitivity leads to earlier and more synchronized outbreaks across counties, whereas lower sensitivity results in delayed and asynchronous trajectories.
• Model Performance vs. SIR: When fitted to historical data, the proposed model outperformed the independent SIR model in over 75% of Oklahoma counties across all standard error metrics.
• Forecasting Accuracy: The proposed model demonstrated superior forecasting capabilities across all epidemic stages. The win rates against the SIR model by nRMSE were:
  • Surge: 59.74%
  • Peak: 87.01%
  • Post-peak: 80.52%
  • Decline: 88.31%
• Contribution of NO2: Isolating the NO2 component revealed that it significantly improved forecasts during periods of high incidence. Specifically, the inclusion of NO2 improved county-level forecasts during the surge (94.81% of counties), post-peak (88.31%), and peak (62.34%) stages compared to the mobility-only model. However, the NO2 term did not yield improvements during the early and decline stages.

Significance and Claims
The paper claims that environmental exposure, specifically NO2, can be directly integrated into the transmission structure of a mechanistic epidemic model. The authors assert that this integration yields measurable forecasting gains, particularly during high-incidence periods, demonstrating that combining environmental data with mobility patterns provides a more robust framework for understanding and predicting epidemic dynamics than mobility data or standard compartmental models alone.