Transmission of COVID-19 in 3 time periods
By March 18, a total of 32,682 cases were identified from the national infectious disease surveillance system (Table S1). Estimates of the effective reproduction number Rt through the whole epidemic period was shown in Figure 1. The Rt varied in the period 1 with a peak of 3.86 on Jan. 23, and declined in the period 2 and 3. The Rt fell below 1.0 on Feb. 8, 2020 and further decreased to below 0.1 on Mar. 15, 2020. As shown in Table 1, the number of onset cases in three periods were 6,981, 18,381 and 7,320, respectively. Average daily new cases in three periods were 166.2, 1,225.4 and 209.1, respectively. The median of double time elevated from 3.6 days in period 1 to 103.9 days in periods 3, but the median of interval from disease onset to diagnosis decreased form 20.0 day in period 1 to 3.0 days in period 3.
The spatiotemporal distribution of COVID-19 cases in Wuhan
To understand the early transmission of COVID-19 in Wuhan, we made an epidemic map of Wuhan till Dec.25 (Figure 2). We noted that the case first appeared in the center of Wuhan, then extended to the north part after 4 days. On Dec.23, 11 days later, the southern part of Wuhan city also began to report cases, and soon, after 2 days, the east and west parts reported the first case. By Dec.25, 17 days after first reported case, COVID-19 cases had been reported in all 13 districts of Wuhan.
A total of 179 streets in Wuhan city were included in the present analysis and COVID-19 cases were reported from 177 of them. Global spatial trends in whole epidemic and 3 time periods were visualized in Figure 3. The trend lines suggested COVID-19 cases aggregated in central urban area during all periods, but such overall trend of aggregation reduced clearly in period 3. Global spatial autocorrelations in whole epidemic and different periods were examined by Moran's I (Figure 4). In all Moran scatter plots, bubbles mainly aggregated in the first, second and third quadrants, suggested that the spatial distribution form of COVID-19 onset cases in all period were mainly composed of three main patterns: high-high, low-high and low-low. Moran's I in all periods was more than 0, but dropped from 0.31 in period 1 to 0.12 in period 3. Significance tests of Moran's I performed by Monte-Carlo method with 999-time simulations indicated significant (pseudo p value <0.05) global autocorrelation existed in all periods (Figure S1).
In order to have a more intuitive view of spatial distribution of COVID-19 onset cases in different periods, LISA cluster map was employed to graphically demonstrate local autocorrelation of COVID-19 onset cases in street-level (Figure 5). From the perspective of the whole epidemic, the main models of onset cases clustering from the central urban area to the marginal urban area were high-high, high-low or low-high, and low-low, successively. As shown in Table 2, the number of streets which did not present significant clustering elevated obviously from 18 in period 1 to 54 in period 3. Closer inspection of the Table 2 showed such trend of reduction was due to a decrease in both high-high and low-low aggregation patterns.
Analysis of spatial differentiation drivers
To explore the driving factors of COVID-19 cases spatial differentiation, we performed a tertile analysis of the street according to the population density or the number of public facilities in each street (Table S2). The results suggested that all COVID-19 indictors (including cumulative number of case, average prevalence, doubling time and daily new cases were monotonic increase across tertiles of population density (all Ptrend < 0.05). The number of daily new cases in three periods, as well as the average prevalence and the cumulative cases of COVID-19 (all Ptrend < 0.05) elevated significantly with the increase in the number of hospitals. We didn’t observe any one-way variation trend between shopping center (except number of average daily new cases) and other COVID-19 related indicators, or between the number of traffic station and COVID-19 indicators.
To further validate such potential associations, a spatial error model was constructed to detect the association of the number of COVID-19 onset cases with population density, ratio of the elderly population and number of public facilities in street-level. As shown in Table 3, population density and the number of hospitals were significantly associated with the number of onset cases at street-level (both P <0.05) rather than ratio of elderly population and the number of other public facilities throughout the whole epidemic. When stratified into three periods, significant associations of onset cases with population density and the number of hospitals were observed in period 1 and 2. In addition, the number of traffic stations was positively associated with onset cases with a coefficient of 4.437 in period 2. Strikingly, no significant association between population density and onset cases was found in period 3. Nonetheless, the number of hospitals was still positive associated with onset cases elevation in period 3, but the coefficient was lower than that in period 2 (6.809 vs 13.559).