Temporal relationship between outbound traffic from Wuhan and the 2019 coronavirus disease (COVID-19) incidence in China

Zaixing Shi; Ya Fang

The optimal ARIMAX models for each province cluster are summarized in Table 1 . The ARIMA (2,1,0) was optimal for provinces with <1 week of lag, and the ARIMA (1,1,0) was optimal for provinces with ≥1 week of lag. The model-predicted cumulative incidence based on actual traffic data demonstrated good fit with the observed incidences ( Figure 4A -4C, Table 1 ). The predicted cumulative incidences would be higher in all three clusters of provinces based on 2019 traffic data ( Figure 4D-4F) , and the differences were projected to grow larger moving forward. It is estimated that there would be approximately 2,123 more cases in provinces with <1 week lag, 4,363 more in provinces with 1 week of lag, and 13,282 more in provinces with 2-3 weeks of lag by the end of February, although the predicted value was not significantly different from the observed in provinces with <1 week of lag (Table 1) . We estimated that the travel ban may have reduced a total of 19,768 (95% CI: 13,589, 25,946) cases outside of Wuhan by the end of February.