Transportation flows and outbreak origins in epidemic spread: Insights from agent-based modeling
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Human mobility is a key driver of early epidemic spread, and restricting travel remains one of the principal non-pharmaceutical interventions. To better understand how infections propagate through real-world mobility networks, it is essential to disaggregate their components and characterize the functional relationships between mobility flows and epidemic metrics.
Here we introduce transCovasim, an agent-based extension of Covasim that explicitly couples parallel city simulations via inter-city traveler exchange, enabling controlled experiments on mobility and disease dynamics. Using transCovasim, we analyze a two-city system with equal or unequal populations and a hub-and-satellite commuting network parameterized to a Moscow-like setting. In paired identical cities, the mean lag between epidemic peaks scales approximately linearly with the logarithm of inter-city traffic, with steeper delays at lower transmissibility; epidemic variability declines as flows increase. With unequal city sizes, mobility primarily redistributes infections between cities; first-order Sobol’ indices show that peak magnitude is largely insensitive to city’s outbound and inbound flows when sizes are comparable, while sensitivity for a smaller city shifts toward inbound flow as the asymmetry increases. In the hub-and-satellite network, reducing commuting flows before the peak significantly lowers peak incidence, and cumulative infections can still be reduced when restrictions are introduced after the peak; early 100-fold cuts outperform 10-fold cuts, but produce similar results when introduced into the late exponential phase. Finally, dynamic time warping applied to surveillance curves identifies the outbreak’s origin: under Moscow-like flows, accuracy reaches ∼85% by day ∼40 with 10% daily testing, and approaches 100% at lower connectivity. These results clarify how specific mobility patterns shape epidemic timing and burden and provide actionable guidance for mobility-targeted non-pharmaceutical interventions and early source attribution.
Author summary
Human mobility governs how epidemics spread between cities, yet policy often treats it as a single lever. We introduce transCovasim, an agent-based extension of Covasim that links parallel city simulations via explicit traveler exchange, allowing controlled tests of pairwise inter-city traffic and hub-and-satellite commuting. In identical cities, the lag between epidemic peaks shrinks approximately linearly with the logarithm of traffic, and stochastic variability decreases as flows rise, especially at lower transmissibility. With unequal sizes, mobility chiefly redistributes infections; sensitivity increases as asymmetry grows. Cutting commuter traffic before the peak reliably reduces peak incidence, while later cuts still lower cumulative burden; early deeper cuts help more. Comparing surveillance curves with time-series alignment can identify the likely source within about a month under moderate testing. These results provide quantitative, network-aware guidance for targeting connections and timing interventions.
