Temporal trends and patterns
As of Mar. 4, 31 provinces (100% of mainland China) reported 80,409 COVID-19 cases, with the number of incident cases ranging from 1 to 15,153 per day. The average incidence rate was 5.76 infections per 100,000 persons (range: 0.03-114.02) during the selected period of the COVID-19 epidemic. Outside of the Hubei Province epicenter, Beijing and Shanghai were among the first case-reported provinces for COVID-19 on Jan. 20, 2019. Compared with COVID-19, SARS had a less widely influential area but a longer epidemic duration, and only 24 provinces (77% of mainland China) reported 3,571 SARS cases as of Aug. 3, 2003, with an average incidence rate of 0.41 per 100,000 (range: 0.00-16.72). (Figure. 1).
To illustrate the spread of the two diseases nationally, we plotted the temporal changes in COVID-19 and SARS in 31 provinces in mainland China (Figure. 2, ordered by administrative area code). In most provinces except Hubei, the rate of increase in the number of cases for COVID-19 was fast for the first two weeks and reached a peak at the end of January. On the other hand, the incidence trend for SARS was mostly flat, except in Beijing, Tianjin, Hebei, Shanxi and Inner Mongolia. Notably, compared to SARS, there was an obvious increasing trend for COVID-19 in terms of the number of new cases in 12 provinces, such as Heilongjiang, Shanghai, Jiangsu, Zhejiang, Anhui, Jiangxi, Shandong, Henan, Hubei, Hunan, Chongqing and Sichuan. On the other hand, several provinces in western China, such as Guangxi, Yunnan, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang and Tibet, had a much lower prevalence for both COVID-19 and SARS.
Identification of spatiotemporal clusters
Through spatiotemporal clustering analysis, we identified 4 high-risk clusters for COVID-19 within 4 cluster time frames (Figure. 3a). The most likely cluster was the epicenter, Hubei, with an RR of 135 compared with the neighboring provinces and the longest high-risk period of 22 days (P < 0.001). Two significant secondary clusters were identified in Zhejiang (from Jan. 28 to Jan. 30, 2020, P < 0.001) and Shandong (in Feb. 20, 2020, P < 0.001), with similar RRs of 1.64 and 1.56, respectively. Another possible cluster was identified in Jiangxi (from Feb. 3 to Feb. 4, 2020, P = 0.982).
When considering the measure of quarantine in Hubei, the RR of 223 in stage 2 (from Feb. 7 to Mar. 4, 2020) was largely increased compared to the RR of 69 in stage 1 (Jan. 20 to Feb. 6, 2020) (Additional file 1). There were different spatial behaviors and temporal features between the two stages. When excluding cases in Hubei, the high-risk clusters were centered on the areas around Hubei and in Beijing, Shanghai, and Heilongjiang in stage 1, whereas the high-risk clusters were only restricted within the neighborhood areas of Hubei in stage 2. Moreover, the RRs in both stages were significantly decreased for the most likely cluster, with RRs of 3.56 in stage 1 and 5.31 in stage 2 (Additional file 2).
Different from COVID-19, the most likely cluster of SARS was centered on Beijing (Figure. 3b), lasting from Apr. 21 to May. 24, 2003, with the highest RR of 423 and a longest period of 34 days (P < 0.001). Three significant secondary clusters were identified in Shanxi and Hebei (from Apr. 21 to May. 14, 2003, P < 0.001), Guangdong (from Apr. 21 to May. 8, 2003, P < 0.001), and provinces of Jilin, Liaoning, Heilongjiang and Tianjin (from Apr. 27 to May. 11, 2003, P < 0.001), respectively.
The global Moran’s I values for COVID-19 and SARS were − 0.022 and 0.073, respectively (both P > 0.05), which indicated that the case distribution may have been due to chance rather than global autocorrelation (Figure. 4).
The LISA cluster map showed the significant locations color coded by the type of spatial autocorrelation. For COVID-19, the high-low spatial clustering was in Hubei Province. In addition, we identified 4 significant clusters at P < 0.01 and 5 significant clusters at P < 0.05. Specifically, Liaoning, Inner Mongolia, and most western provinces had significantly low-low spatial clustering, whereas Anhui, Hunan and Jiangxi of Central China had significantly low-high spatial clustering. For SARS, two significant high-high (Beijing and Tianjin) and low-high (Hebei) clusters were detected. Sichuan, Tibet and Anhui showed significant low-low clustering (Figure. 4).