Charting the spatial dynamics of early SARS-CoV-2 transmission in Washington state

Author summaryGeographic structure in human populations has been widely recognised as a key determinant in the spread of infectious diseases, however the mechanisms underlying it are hard to disentangle. Using two distinct statistical analyses, we sought to address questions surrounding the impact of movement patterns and stochastic events on the spatial spread of SARS-CoV-2 in Washington state. Through our first analysis, which made use of a clustering algorithm, we uncovered a relationship between spatial proximity and the similarity of epidemic trajectories. Clusters of similar trajectories formed a clear geographical pattern. In our second analysis, we performed statistical inference on time series data from the Puget Sound region of Washington state. From our inference, we found that both a high degree of connectivity and atypically fast transmission within the first few weeks of the outbreak were necessary to explain the rapid inter-county spread. Furthermore, this stochastic early spread had a lasting effect on the subsequent epidemic course as non-pharmaceutical interventions were insufficient to temporarily eradicate the virus from the region. The spread of SARS-CoV-2 has been geographically uneven. To understand the drivers of this spatial variation in SARS-CoV-2 transmission, in particular the role of stochasticity, we used the early stages of the SARS-CoV-2 invasion in Washington state as a case study. We analysed spatially-resolved COVID-19 epidemiological data using two distinct statistical analyses. The first analysis involved using hierarchical clustering on the matrix of correlations between county-level case report time series to identify geographical patterns in the spread of SARS-CoV-2 across the state. In the second analysis, we used a stochastic transmission model to perform likelihood-based inference on hospitalised cases from five counties in the Puget Sound region. Our clustering analysis identifies five distinct clusters and clear spatial patterning. Four of the clusters correspond to different geographical regions, with the final cluster spanning the state. Our inferential analysis suggests that a high degree of connectivity across the region is necessary for the model to explain the rapid inter-county spread observed early in the pandemic. In addition, our approach allows us to quantify the impact of stochastic events in determining the subsequent epidemic. We find that atypically rapid transmission during January and February 2020 is necessary to explain the observed epidemic trajectories in King and Snohomish counties, demonstrating a persisting impact of stochastic events. Our results highlight the limited utility of epidemiological measures calculated over broad spatial scales. Furthermore, our results make clear the challenges with predicting epidemic spread within spatially extensive metropolitan areas, and indicate the need for high-resolution mobility and epidemiological data.

Automated summary

Geographic structure and stochastic events have significant impacts on the spatial spread of SARS-CoV-2 in Washington state. Clustering analysis reveals a relationship between spatial proximity and epidemic trajectories, while statistical inference highlights the lasting effect of early stochastic spread. These findings emphasize the geographical variation and predictive challenges in epidemic spread.