Summary of the COVID-19 epidemic in China: When and how to launch an emergency response


 In December 2019, Coronavirus Disease 2019 (COVID-19) was first detected in Hubei Province and spread rapidly around the world. Summarizing the development of COVID-19 and assessing the effect of control measures are very critical to China and other countries. A heatmap was used to find the highest concentration of the COVID-19 outbreak and the areas with initial imported cases. A logistic growth curve model was employed to compare the development of COVID-19 before and after the emergency response took effect. We found that the number of confirmed cases peaked 9-14 days after the first detection of an imported case, but there was a peak lag in the province where the outbreak was concentrated. The average growth rate of cumulative confirmed cases decreased by approximately 50% after the emergency response began. Areas with frequent population migration have a high risk of outbreak. The emergency response taken by the Chinese government was able to effectively control the COVID-19 outbreak. Our study provides references for other countries and regions to control the COVID-19 outbreak.


Introduction
On December 31, 2019, China noti ed the World Health Organization of unknown pneumonia cases in Wuhan, Hubei Province [1]. This pneumonia came with persistent fever, cough, and dyspnea [2] and was then named Coronavirus Disease 2019 . The disease spread rapidly from Hubei Province to other provinces in China within two weeks [3,4]. Beginning January 15, 2020, the Chinese government launched an emergency response at all levels. On the one hand, in the outbreak area, Hubei Province implemented tra c control. On the other hand, the whole nation was required to wear masks and to avoid going out and having close contact with other people in order to reduce the exposure to susceptible people. By March 20, 2020, a total of 81,416 con rmed cases and 3,261 deaths had been reported in China, of which less than seventeen percent of the cases and less than four percent of the deaths occurred outside Hubei Province. Since January 13, 2020, rst Thailand [5], then more than 200 countries, including Japan, Korea [6], the United States [7] and the United Kingdom [8], have reported imported COVID-19 cases. Due to the speed and scale of transmission, the WHO described COVID-19 as a pandemic on March 12, 2020, o cially declaring that COVID-19 entered the global epidemic phase.
By March 31, the total number of con rmed COVID-19 cases and deaths outside China reached 776,506 and 38,841, respectively, and were increasing. There have been concentrated outbreaks in Europe [9] and America [10]. Many imported cases have been reported in the Middle East [11] and in African countries [12]. These countries are currently in different stages of COVID-19 development. However, there is no relevant research or evidence on how to assess the development of COVID-19 [13]. As the earliest occurrence area, Hubei Province, China has been through the process of case accumulation -outbreak detection -isolation and control. The rest of China has been through a complete process of case imports -detected transmission -isolation and control. Therefore, summarizing COVID-19 development in Hubei Province and other regions of China can help us to explore the epidemic characteristics of COVID-19 and provide a reference for other countries to assess the stages of the COVID-19 epidemic.
The course of COVID-19 includes incubation, disease, and recovery or death [2,14]. This course is characterized at the population level, as the number of cumulative con rmed cases experience a period of delay before exponential growth, then present a period of maximum increasing density, and nally enter a stable stage. The entire process presents an S-shaped development trend. A logistic growth curve model [15] is often used to describe such ecological processes [16,17]. Both the average growth rate and the maximum value of the growth curve have clear epidemiological signi cance and are of great reference value in the eld of public health. Therefore, this study intends to use the logistic growth curve model to t the development of COVID-19 into two periods and then summarize the development of a concentrated COVID-19 outbreak in Hubei Province and in high-risk areas with imported cases. In addition, this study will extract historical data to simulate a shortterm dynamic prediction and discuss the application of the growth curve model in the assessment of COVID-19 in order to provide a reference for other countries.

Data sources
Con rmed COVID-19 case data were obtained from the Chinese Center for Disease Control and Prevention [18]. All cases were con rmed by laboratory and clinical diagnosis and met the de nition of con rmed cases according to the National Health Commission of China [19]. Baidu is the most widely used search engine in China, and we extracted population migration data from the Baidu Qianxi to nd areas with early imported cases [20]. Considering that in the early stages of the COVID-19 outbreak, the situation reports may have underreported cases, we used con rmed cases from January 22 to March 4, 2020 to ensure the reliability of the data.

Statistical analysis
This study conducted a spatiotemporal distribution analysis of cumulative con rmed COVID-19 cases in China on a provincial level. We selected Hubei Province as the concentrated outbreak area for analysis, and we selected other provinces with early reported cases as representative provinces facing the risk of outbreak.
We tted the growth curve model for the cumulative con rmed cases in Hubei and other provinces facing the risk of outbreak. The formula for the model is as follows: where N t represents the cumulative con rmed COVID-19 cases at time t, N 0 represents the cumulative con rmed cases at the initial time, K represents the maximum cumulative con rmed cases within the analysis period, and r is the average increase rate of the cumulative con rmed cases.
After the outbreak of COVID-19, the rst level of the Chinese public health emergency response [21] (later referred to as "emergency response") was gradually implemented in each province. To assess the impact of the emergency response implemented in each province, we tted the growth curves at two different periods, using an average incubation period of seven days [14,19] after the emergency response time as the cut-off point (for details of the time period, see Supplementary Table S1). The rst time period was used to assess the situation before the emergency response was implemented. The second time period, from the end of period one to March 4, 2020, was used to assess the situation after the emergency response had taken effect. The coe cient of determination (R 2 ) was used to evaluate the goodness of t. The average growth rates of periods one and two in each province were compared to assess the impact of the prevention and control measures.
To evaluate the prediction capacity of the logistic growth curve model, we used the cumulative con rmed cases from January 22 to February 4, 2020 to simulate the short-term dynamic prediction. The step lengths of the dynamic predictions were set as one, three and seven days, referred to as the 1, 3, or 7 out-of-sample prediction. In the 1 out-of-sample prediction, the cumulative con rmed cases from January 22 to February 4 were selected as the training set, and one day after, February 5, was selected as the test set. Then, the model was updated with actual observations from February 5, and the cumulative con rmed cases on February 6 were predicted by the updated model until all the predicted cumulative con rmed cases from February 5 to March 4 were obtained. The average absolute error (MAE) and average absolute percentage error (MAPE) were then calculated for each dynamic prediction with different step lengths to measure the model prediction accuracy.

General characteristics of COVID-19 in China
Wuhan, Hubei Province shut down outward tra c beginning January 23, 2020, followed by the rest of Hubei Province. To nd high-risk areas caused by imported cases, we drew a heatmap of the migration out of Hubei on January 22, 2020 ( Fig. 1a), which indicated that people mainly migrated to Henan, Hunan, Chongqing, Jiangxi, Guangdong, Anhui, Sichuan, Jiangsu, Zhejiang, Beijing and Shanghai. A heatmap of the cumulative con rmed cases in Chinese provinces from January 22 to March 4, 2020 highlights similar provinces (Fig. 1b). Hubei Province was the location of the concentrated COVID-19 outbreak, followed by its neighbors (Henan, Anhui, Jiangxi, Hunan and Chongqing) and some economically developed and densely populated provinces (Guangdong, Zhejiang, Jiangsu, Shandong, Sichuan, Shanghai and Beijing). Thus, Sichuan, Guangdong, Beijing, Shandong, Chongqing, Zhejiang, Jiangxi, Anhui, Jiangsu, Hunan, Shanghai and Henan were selected as highrisk areas with imported cases for further analysis. In addition, since Hubei Province had the most severe epidemic, we also analyzed national data excluding Hubei Province to present the average epidemic in other provinces. Table 1 shows the peak number of con rmed COVID-19 cases, the corresponding peak date and the cumulative number of con rmed cases in China. Fig. 2 shows a time series of con rmed COVID-19 cases in the identi ed provinces. The con rmed COVID-19 cases in Hubei Province and nationwide showed a rapid increase before February 4, followed by a decline, and gradually stabilized after February 18, 2020. In highrisk provinces with imported cases, the peak of con rmed cases was around January 30, 2020 in Sichuan, Guangdong, Zhejiang and Shanghai, and around February 2, 2020 in Beijing, Chongqing, Jiangxi, Anhui, Jiangsu, Hunan and Henan. Two outliers occurred in China and Hubei Province on February 12 and 13, as the National Health Commission of the PRC revised the de nition of COVID-19 con rmed cases in Hubei Province on February 12, adding "clinical case" to "con rmed case," and left the other provinces unchanged [22]. Another outlier was found in Shandong Province on February 20, corresponding to an outbreak at a prison with 200 con rmed cases [23]. The overall trend of con rmed cases in the other provinces increased rst and then decreased.

Impact evaluation of emergency response
We tted the growth curves at two different periods to assess the impact of the emergency response implemented in each province. Fig. 3 shows the growth curves of each area. The coe cients of the logistic growth curve models in two periods are referred in Supplementary Table   S2 and S3. The tted cumulative con rmed cases were close to the actual observed cases, and the R 2 of all models was above 0.95.
The average growth rates of the two periods in Hubei Province, China and twelve high-risk provinces are presented in Table 2 and Fig. 4. The average growth rate decreased by 44.42% nationally and by 32.5% outside Hubei Province. The average growth rate in each province decreased signi cantly after the emergency response. The average growth rate in the twelve high-risk areas decreased by 29.8%, which was lower than that outside Hubei Province. Before the emergency response, the provinces with the highest average growth rates were ranked from highest to lowest as follows: Hunan, Hubei, Zhejiang, Shandong, Jiangxi, Jiangsu, Guangdong, Sichuan, Anhui, Henan, Chongqing, Beijing and Shanghai. Hubei, Shandong, Zhejiang, Jiangxi and Hunan had growth rates higher than the national average. After the emergency response, the average growth rate of each province from highest to lowest was Zhejiang, Hunan, Anhui, Shanghai, Jiangxi, Jiangsu, Hunan, Guangdong, Hubei, Chongqing, Beijing, Sichuan and Shandong. The growth rates of Guangdong, Zhejiang, Jiangxi, Anhui, Jiangsu, Hunan, Shanghai and Henan were higher than the national average. Prediction capacity evaluation of logistic growth curve models We used cumulative con rmed case data, from January 22 to February 4, 2020, to simulate a short-term dynamic prediction. Table 3 shows the MAE and MAPE of the logistic growth curve model in each province. Fig. 5 shows the 1-step dynamic prediction of the logistic growth curve model in Hubei Province, China and twelve high-risk provinces. The 1-step dynamic prediction outperformed the rest, with a MAPE of 1.16%-5.45% in different areas. Except for the models for China, Hubei and Shandong provinces, which were affected by the three outliers mentioned above, the models showed predictions close to the observations.

Discussion
On January 30, 2020, the WHO declared COVID-19 to be a public health emergency of international concern (PHEIC), and later, it was described as a pandemic. COVID-19 is threatening the public health security of every country [24,25]. As the country with the initial COVID-19 outbreak, China issued a series of policies and regulations to control the outbreak [26,27], including cross-regional tra c control and suspending the operations of restaurants, entertainment, and cultural tourism areas. The government encouraged citizens to stay at home, stop gathering, wear masks and wash hands frequently. Summarizing the development of COVID-19 in China and assessing the effect of control measures can provide a reference for other countries to deal with the outbreak. In this paper, we summarized the development of COVID-19 in Hubei Province, China and twelve other provinces with a high incidence of COVID-19. We also compared the characteristics of the epidemic before and after the emergency response to assess the impact of the prevention and control measures.
Before the tra c leaving Wuhan, Hubei was shut down, and people from Hubei Province mainly migrated to Henan, Hunan, Chongqing, Jiangxi, Guangdong, Anhui, Sichuan, Jiangsu, Zhejiang, Beijing and Shanghai, which was consistent with provinces later had high incidences of COVID-19. Other studies have shown that population density can directly affect the spread of such diseases [28]. It has been suggested that blocking migration from severe outbreak areas would be of great importance to prevent the disease from spreading to other areas, especially during early stages. Tian H et al. showed that this suggestion worked [29].
The peak outbreak occurred from February 1 to February 4, 2020, which could be related to the population migration and the incubation of COVID-19. As January 25 th was the traditional Chinese New Year, most people were returning to their hometowns to reunite with their families.
Therefore, the densi ed migration in the week before the traditional Chinese New Year led to the rapid spread of COVID-19. The incubation of COVID-19 is estimated to be 3-7 days. Each province experienced 9-14 days from the rst detection of imported cases to the peak of con rmed cases, which was consistent with the sum of the migration peak and the incubation period. However, in the region with the most severe outbreak, Hubei Province, the peak of con rmed cases was delayed due to the long accumulation of con rmed cases and inadequate testing capacity, which is consistent with the ndings of Kaiyuan Sun et al. [30]. Therefore, 9-14 days after the detection of imported cases is the critical period for preventing further transmission. In this period, screening tests and the quarantine of COVID-19 patients should be carried out to identify the infection source and protect susceptible populations.
We tted the logistic growth curves of cumulative COVID-19 cases before and after the implementation of an emergency response in each study province and found an approximate 50% reduction in the average growth rate after the emergency response. As all the emergency responses were launched within one week after the rst con rmed case, the reduction in the average growth rate suggested that rapid growth of the epidemic can be slowed by a timely emergency response after the early detection of imported cases within the critical period of 9-14 days.
The average growth rate in Zhejiang, Jiangsu, Anhui, Jiangxi, Hunan, Shanghai, and Henan provinces remained higher than the national average growth rate after the implementation of the emergency response. Among them, the economically developed provinces and laborexporting provinces with frequent population migration, such as Zhejiang, Hunan and Anhui provinces, had the highest growth rates, indicating a high outbreak risk. Therefore, the control measures should be particularly strengthened to prevent COVID-19 outbreaks in these regions. Although the emergency response reduced the average growth rate, in the outbreak center, Hubei Province, the peak in con rmed cases was delayed. This suggests that if the outbreak was not detected in time, the critical control period would be missed. This would lead to a lag in the implementation of prevention and control measures in response to the outbreak. Therefore, for concentrated COVID-19 outbreak areas, the growth of the epidemic would not be easily controlled within the standard critical period of 9-14 days. The lagged peak of con rmed cases should be fully considered, and the duration of control measures should be extended for further development of the epidemic.
In the 1-step dynamic prediction of the cumulative con rmed COVID-19 cases in the early stage of the epidemic, the MAPE between the predicted and actual cumulative cases was 1.16%-5.45%. Despite the increase due to the change in diagnostic criteria on February 13 th in Hubei Province, the values predicted by the logistic growth curve model were very close to the actual observed values. Thus, the logistic growth curve model can be used to assess the short-term development of COVID-19 and aid in the short-term adjustment of prevention and control measures.
In conclusion, the logistic growth curve model can accurately assess and predict the short-term development of the COVID-19 epidemic. Timely detection of imported cases and blocking migration from the epidemic area are important for controlling the spread of COVID-19. Nine to 14 days after the rst detection of imported cases is the critical period for epidemic prevention and control. In areas where the epidemic is severe, we need to consider the peak lag and extend prevention measures. Areas with frequent migration have a high risk of COVID-19 outbreak, so the prevention and control measures should be strengthened. The emergency responses launched in China e ciently reduced the spread and further development of the epidemic, which provides a reference for other countries and regions.
This study is based on the existing surveillance data, and the detection capacity of COVID-19 varies between different regions and countries. Insu cient detection capacity will lead to an underestimated occurrence, and the outbreak re ected by the surveillance data may be delayed. Each region should consider local detection capacity when formulating prevention and control measures.

Declarations Data availability
Chinese Center for Disease Control and Prevention has published the COVID-19 situation since Jan 16th. Everyone can obtain the daily con rmed COVID-19 cases from http://2019ncov.chinacdc.cn/2019-nCoV/. This research has been conducted using the con rmed COVID-19 cases from 22th January 2020 to 4th March.
F.Y. and Y.M. designed the study, collected data, and contributed to data analysis. J. T. contributed to the literature search, data analysis, data interpretation, gures, and writing. C.L., J.H., and T.Z. contributed to data interpretation. All authors contributed to writing the manuscript and revising the nal version. Figure 1 a Percentage of the migration population moving from Hubei province to other provinces on January 22, 2020. b The cumulative con rmed COVID-19 cases in Chinese provinces from January 22 to March 4, 2020. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.