Identification of superspreading environment under COVID-19 through human mobility data

Abstract

COVID-19 reaffirms the vital role of superspreaders in a pandemic. We propose to broaden the research on superspreaders through integrating human mobility data and geographical factors to identify superspreading environment. Six types of popular public facilities were selected: bars, shopping centres, karaoke/cinemas, mega shopping malls, public libraries, and sports centres. A historical dataset on mobility was used to calculate the generalized activity space and space-time prism of individuals during a pre-pandemic period. Analysis of geographic interconnections of public facilities yielded locations by different classes of potential spatial risk. These risk surfaces were weighed and integrated into a "risk map of superspreading environment" (SE-risk map) at the city level. Overall, the proposed method can estimate empirical hot spots of superspreading environment with statistical accuracy. The SE-risk map of Hong Kong can pre-identify areas that overlap with the actual disease clusters of bar-related transmission. Our study presents first-of-its-kind research that combines data on facility location and human mobility to identify superspreading environment. The resultant SE-risk map steers the investigation away from pure human focus to include geographic environment, thereby enabling more differentiated non-pharmaceutical interventions and exit strategies to target some places more than others when complete city lockdown is not practicable.

Figure 1

Transmission patterns under the non-superspreading and superspreading environment.

Figure 2

An illustration of the space–time concept and spatial agglomeration.

Figure 3

Public facilities by potential spatial risk (PSR) (a) and risk map of superspreading environment (SE-risk map) (b). (a) Symbol locations correspond to four PSR classes as shown in the map legend. (b) Top ten locations of empirical local cluster infection (numbered 1–6) plotted against the SE-risk map, a generalized surface of potential superspreading risks with darker shadings indicating higher risks of infection. The empirical local cluster infection cases of COVID-19 shown above included the top 10 infected clusters with the highest number of cases recorded between February and July 2020. (Generated by ArcGIS 10.5, URL: http://www.esri.com/software/arcgis/arcgis-for-desktop).

Figure 4

Activity space (AS) based on (a) travel characteristic survey versus (b) empirical data. No significant difference in the size of AS for both (a) and (b). Two-sided Mann–Whitney U test indicates that the AS of individuals who visited street blocks of bars in class 1 PSR (n = 2,463; mean rank = 1,248.97) and the empirically infected cases from bars (n = 35; mean rank = 1,286.74) in class 1 PSR were NOT significantly different in size (p = 0.758). Extreme outliers were removed. (a) AS of individuals who visited bars in class 1 PSR based on TCS-2011. Two-sided Mann–Whitney U test indicates the AS of individuals visiting bars in class 1 PSR (n = 2463; mean rank = 3437.18) is higher than those visiting bars in class 3 PSR (n = 3887; mean rank = 3009.68) (Z = − 9.06; p = 0.000). Extreme outliers were removed. (b) AS of individuals infected with COVID-19 and had visited bars in class 1 PSR. Two-sided Mann–Whitney U test indicated the AS of individuals visiting bars in class 1 PSR (n = 35; mean rank = 23.77) is higher than those visiting bars in class 3 PSR (n = 8; mean rank = 14.25) (Z = − 1.94; p = 0.026). Extreme outliers were removed. (Generated by ArcGIS 10.5, URL: http://www.esri.com/software/arcgis/arcgis-for-desktop).