Spread of COVID-19 through Georgia, USA. Near-term projections and impacts of social distancing via a metapopulation model

Abstract

Epidemiological forecasts of COVID-19 spread at the country and/or state level have helped shape public health interventions. However, such models leave a scale-gap between the spatial resolution of actionable information (i.e. the county or city level) and that of modeled viral spread. States and nations are not spatially homogeneous and different areas may vary in disease risk and severity. For example, COVID-19 has age-stratified risk. Similarly, ICU units, PPE and other vital equipment are not equally distributed within states. Here, we implement a county-level epidemiological framework to assess and forecast COVID-19 spread through Georgia, where 1,933 people have died from COVID-19 and 44,638 cases have been documented as of May 27, 2020. We find that county-level forecasts trained on heterogeneity due to clustered events can continue to predict epidemic spread over multi-week periods, potentially serving efforts to prepare medical resources, manage supply chains, and develop targeted public health interventions. We find that the premature removal of physical (social) distancing could lead to rapid increases in cases or the emergence of sustained plateaus of elevated fatalities.

Real-time, interactive website for US-county-level COVID-19 event risk assessment

“…Rapid, population-wide serosurveys are needed to connect case reports to localized estimates of ascertainment bias. Integrating such serosurveys at state levels or improvements in estimates of ascertainment bias using statistical or mechanistic models 30,31 could further refine variation in event risk estimates.…”

Direct estimation of SIR model parameters through second‐order finite differences

“…Runge–Kutta pairs are widely used for addressing IVPs. In consequence are ideal for use in SIR models 2‐4 . Here, we will test the two approaches over some artificially produced data.…”

Runge–Kutta pairs suited for SIR‐type epidemic models

“…Runge–Kutta (RK) pairs are widely used for addressing IVPs. In consequence are ideal for use in SIR models 2‐4 . These pairs are characterized by the extended Butcher tableau: 5,6 …”

“…In consequence are ideal for use in SIR models. [2][3][4] These pairs are characterized by the extended Butcher tableau: 5,6 , with b T ,b T , c ∈ R s , and A ∈ R s×s is strictly lower triangular with Ae = c. For advancing the solution from (t n , y n ) to t n + 1 = t n + h n , we compute two approximations each step y n + 1 ,̂n +1 to y(t n + 1 ) of orders p and p − 1, respectively, given by…”