The effect of inter-city travel restrictions on geographical spread of COVID-19: Evidence from Wuhan, China

Billy J Quilty; Charlie Diamond; Yang Liu; Hamish Gibbs; Timothy W Russell; Christopher I Jarvis; Kiesha Prem; Carl A B Pearson; Samuel J Clifford; Stefan Flasche; Petra Klepac; Rosalind M Eggo; Mark Jit

While earlier restrictions on travel from Wuhan may have had a larger impact, in countries with a high-degree of inter-city travel, it may be difficult to implement such highly disruptive travel restrictions at an early stage of the epidemic, before local transmission has already occurred in other cities. We find that local transmission in the four cities we studied (here defined as the probability of sustained transmission exceeding a 95% threshold) was most likely established between 2 January and 8 January; it was only on 8 January that the etiology of the "mystery pneumonia" (which was not yet confirmed to spread from person-to-person (25)) was determined as a novel coronavirus, and the first death occurred (26) . It is difficult to see how the cordon sanitaire could have been justified any earlier, as almost every aspect of COVID-19 virology and epidemiology was unknown. Hence, it is likely that the sustained decline in COVID-19 incidence in other cities of China several months into the outbreak is primarily due to other public health measures to reduce the disease transmissibility, i.e, to reduce the reproduction number to 1 or below (8, 18, 23) . The cordon sanitaire may have been more efficacious in delaying outbreaks internationally, as the relative number of travellers is orders of magnitude lower (7, 27) ; the same may also apply to lower-traffic destinations from Wuhan within China, such as small cities geographically distant from Wuhan, as observed in Tian et al. 2020 (7) . We found a mixed picture in these cities, where the cordon sanitaire may have been more efficacious at delaying or preventing outbreaks ( Figure S1 ). However, COVID-19 transmission dynamics may differ in comparison to big cities, and as such we chose to focus on the effect of travel restrictions in large cities with large volumes of travel from Wuhan, where data on R and k from the early outbreak in Wuhan are likely generalisable. Furthermore, these destinations with low traffic from Wuhan are more likely to be seeded by outbreaks in other, comparatively closer, large cities first. Hence, our assumption of a single outbreak source would have been much less realistic.