key: cord-0589258-qcemuh20
authors: Wang, Peixiao; Ren, Hui; Hu, Tao; Fu, Xiaokang; Liu, Hongqiang; Zhu, Xinyan
title: Spatiotemporal Characteristics and Factor Analysis of SARS-CoV-2 Infections among Healthcare Workers in Wuhan, China
date: 2020-12-16
journal: nan
DOI: nan
sha: 043c1382bad70fdc4c60325fdfd91cebf0c135c2
doc_id: 589258
cord_uid: qcemuh20

Studying the spatiotemporal distribution of SARS-CoV-2 infections among healthcare workers (HCWs) can aid in protecting them from exposure. Existing studies related to HCW infections have emphasized infection rates and protective measures. However, the spatiotemporal patterns and related external environmental factors of HCW infections remain unclear. To fill this gap, an open-source dataset of HCW diagnoses was provided, and the spatiotemporal distributions of SARS-CoV-2 infections among HCWs in Wuhan, China were explored. A geographical detector technique was then used to investigate the impacts of hospital level, type, distance from the infection source, and other external indicators of HCW infections. The results showed that the number of daily HCW infections over time in Wuhan followed a log-normal distribution, with and its mean observed on January 23, 2020 and a standard deviation of 10.8 days. The implementation of high-impact measures, such as the lockdown of the city, may have increased the probability of HCW infections in the short term, especially for HCWs in the outer ring of Wuhan. The infection of HCWs Wuhan exhibited clear spatial heterogeneity. The number of HCW infections was higher in the central city and lower in the outer city. Moreover, HCW infections displayed significant spatial autocorrelation and dependence. Factors analyses revealed that hospital level and type had an even greater impact on HCW infections; third-class and general hospitals closer to infection sources were correlated with especially high risks of infection. These findings can aid national epidemic prevention and control departments to understand the spatiotemporal distributions of viral transmission resulting in HCW infections, as well as external influencing factors, which can facilitate the protection of HCWs in China.

At the end of 2019, SARS-CoV-2 was discovered and it spread rapidly around the world. By October of 2020, more than 200 countries had been infected, and there were more than 30 million confirmed cases and more than 900,000 deaths attributed to the virus (Organization, 2020) . At present, the management of and response to SARS-CoV-2 are common challenges facing all of humanity (Fang, Yi, Wu, Lai, Sun, Zhong and Liu, 2020; Guan, Ni, Hu, Liang, Ou, He, Liu, Shan, Lei, Hui et al., 2020; Huang, Wang, Li, Ren, Zhao, Hu, Zhang, Fan, Xu, Gu et al., 2020) . During the fight against the epidemic, healthcare workers (HCWs) have always been on the front line. While helping patients combat the virus, HCWs are also exposing themselves to high levels of the virus at close range (Wańkowicz, Szylińska and Rotter; Xiang, Jin, Wang, Zhang and Cheung, 2020) . Thus, how to better protect HCWs is a key issue at both the city and national levels.

Since the outbreak of SARS-CoV-2 began, different degrees of HCW infections have occurred in many countries and regions (Jones, Rivett, Sparkes, Forrest, Sridhar, Young, Pereira-Dias, Cormie, Gill, Reynolds et al., 2020; Rivett, Sridhar, Sparkes, Routledge, Jones, Forrest, Young, Pereira-Dias, Hamilton, Ferris et al., 2020) . For example, the number of HCW infections in China accounts for 4% of all infections nationwide (Gao, Sanna, Tsai and Wen, 2020; Novel et al., 2020) , while that in Italy accounts for 10%. Studies have shown that the infection rate of HCWs is significantly higher than that of non-HCWs (Zheng, Wang, Zhou, Liu, Li, Sun, Wang, Zhou and Wang, 2020) . In order to deal with HCW infections, researchers in China and elsewhere have explored the causes of HCW infections (A, B and A, 2020) , revealed the transmission routes leading to HCW infections (Ferioli, Cisternino, Leo, Pisani and Nava, 2020) , and formulated allocation schemes to address resource limitations (Siow, Liew, Shrestha, Muchtar and See, 2020) and protective measures (Tan, 2020; Tan, Kovoor, Williamson, Tivey, Babidge, Collinson, Hewett, Hugh, Padbury, Langley et al., 2020) .

Spatiotemporal analysis involves the use of statistics and geographical and time-series data to study the patterns and mechanisms underlying a given phenomenon over time and space (Yu, Wong, Chiu, Lee and Li, 2005; Vernaz, Sax, Pittet, Bonnabry, Schrenzel and Harbarth, 2008) . In the field of public health, even the introduction of more simplified analytical methods to evaluate spatial and temporal relationships could be beneficial (Davis, Sevdalis and Drumright, 2014) . Thus, researchers used geographic information systems (GIS) (She, Hu, Zhu and Bao, 2019; Zhu, Hu, Ye, Guo, Huang, Xiong, Ai, She, Xiong and Duan, 2019) to understand the spatiotemporal patterns of viral transmission among reported cases in the early stages of the epidemic (Wang, Liu, Struthers and Lian, 2020b; Jia, Lu, Yuan, Xu, Jia and Christakis, 2020; Liu, Sha, Liu, Houser, Zhang, Hou, Lan, Flynn, Lu, Hu et al., 2020) . However, due to the lack of a dataset on HCW diagnoses, few existing HCW infection-related studies have been able to reveal the spatiotemporal characteristics of HCW infections and the external environmental factors influencing infections from a geographic perspective.

In this study, we first reviewed the existing research related to the spatiotemporal distributions of SARS-CoV-2 transmission, as well as that related to SARS-CoV-2 infections among HCWs. Studies on the spatiotemporal distributions of SARS-COV-2 have mainly focused on reported case data and have explored the patterns and movement of the epidemic, so as to provide scientific basis for relevant measures, such as isolation and the restriction of human activities. For example, Novel et al. (2020) analyzed the epidemiological characteristics of SARS-COV-2 in February of 2020 based on reported case data to analyze the spatial distributions of SARS-CoV-2 infections in China. Wang et al. (2020b) used scanning statistics to detect the hotspots of new cases each week based on the confirmed cases of SARS-COV-2 at the county level in the United States, thereby characterizing the infection rates during the epidemic. Zhang, Li, Yang, Zheng and Chen (2020a) employed Geodetector (http://www.geodetector.cn) and a decision tree to detect driving factors in low-risk and high-risk areas based on the reported case data in Wuhan communities. Further, Zhang, Zhang and Wang (2020b) examined the factors influencing the number of imported cases from Wuhan and the rate and pattern of viral transmission by combining national flight and high-speed rail data. Based on mobile phone and confirmed patient data, Jia et al. (2020) developed a spatiotemporal "risk source" model to determine the geographic distribution and growth trends of SARS-CoV-2 infections to quickly and accurately assess related risks. Loske (2020) explored the relationship between SARS-CoV-2 transmission and transport volumes in food retail logistics by combining transport volume data and confirmed patient data. However, these studies were mostly based on reported case data and did not explicitly account for the spatiotemporal characteristics or factors influencing SARS-CoV-2 infections among HCWs.

Existing studies on HCW infections have mainly focused on the causes and spread of HCW infections and the distribution of medical supplies, so as to provide scientific support for alleviating the HCW infections. For example, A et al. (2020) analyzed the causes of HCW infections from four aspects: insufficient awareness of protective protocols, excessive work intensity, insufficient medical supplies, and a lack of professional knowledge. Wang, Ferro, Zhou, Hashimoto and Bhatt (2020a) assessed the association of hospital masking policies with the SARS-CoV-2 infection rate among HCWs, while Moscola, Sembajwe, Jarrett, Farber, Chang, McGinn, Davidson, Consortium et al. (2020) explored the prevalence of SARS-CoV-2 antibodies among HCWs in New York. Siow et al. (2020) proposed a plan for distributing medical supplies, including personal protective equipment, in response to shortages, in order to maximize conventional medical assets. Tan (2020) also put forward suggestions for minimizing HCW infections.

Few studies have explored the spatiotemporal characteristics of HCW infections from a geographic perspective. For example, Gao et al. (2020) revealed the spatial distribution of 1,688 HCWs infected with SARS-CoV-2 at the national scale based on secondary data. However, due to the limited information contained in secondary data, they did not explore the spatiotemporal characteristics of HCW infections and related influencing factors. In general, the spatiotemporal patterns and related external factors of HCW infections remain unclear. To fill this gap, two types of HCW infection inventories were created to analyze the spatiotemporal patterns of HCW infections and external influencing factors, with the goal of aiding the epidemic prevention and control department in China to understand the spatiotemporal trends of HCW infections, which may be used to improve upon the existing protections for HCWs.

We first studied the spatial distributions of HCW infections in Wuhan using partitioning statistics, distribution fitting, and spatial autocorrelation techniques (Moran, 1950; Anselin, 1995) . A geographical detector was then used to explore the impacts of external factors, such as hospital level, type, and distance from the infection source, on HCW infections (Wang, Li, Christakos, Liao, Zhang, Gu and Zheng, 2010) . Our findings can provide valuable insights into the spatiotemporal trends of SARS-CoV-2 infections among HCWs and the external factors influencing such infections, thus providing a foundation for improving the protection of HCWs. The primary contributions of this work include:

(1)A new method for generating infection inventories;

(2)Two HCW infection inventories -the Grid-Level Healthcare Worker Infection Inventory and the Hospital-Level Healthcare Worker Infection Inventory -based on HCW diagnoses in Wuhan;

(3)Clarification of the spatiotemporal characteristics and external influencing factors of HCW infections from a geographic perspective using the new HCW infection inventories;

(4)An open-source dataset of HCW diagnoses, which both ensures the reproducibility of the study and provides the data needed to support related research on HCW infections.

Wuhan is not only the first city in China where the SARS-CoV-2 was discovered, but also the city with the most serious infections among HCWs in China. Since the outbreak began, more than 3,000 HCWs have been infected in Wuhan, accounting for 82.5% of infections nationwide. As shown in Figure 1 , Wuhan is located in Central China. It is composed of the central city and the outer city. The central city includes the districts of Jiangan, Jianghan, Qiaokou, Hanyang, Wuchang, Qingshan and Hongshan, while the outer city includes Huangpi, Xinzhou, Dongxihu, Jiangxia, Caidian, and Hannan. 

The data used in this study can mainly be divided into two types: (1) hospital data from Wuhan and (2) confirmed data on HCW infections in Wuhan. Hospital data from Wuhan were mainly taken from the Health Commission of Hubei Province (http://wjw.hubei.gov.cn/). We used web crawler technology to search for medical institutions in Wuhan and parsed the returned data to obtain hospital information. After parsing, a total of 285 pieces of hospital information were obtained; the data format is shown in Table 1 . Each record contained the unique identity, name, rank, and type of the hospital. Among them, the ranks mainly includes first-class, second-class, and third-class hospitals, and the type mainly included community, specialized, and general hospitals. Data on HCW diagnoses in Wuhan were mainly derived from the Chinese Red Cross Foundation (https://www.crcf.org.cn/), which distributes relief funds to every confirmed healthcare worker (Foundation, 2020) . As of September 11, 2020, 83 groups of HCWs had received foundation assistance. We used crawler technology to obtain 3,743 publications that elucidated the conditions of HCWs that suffered from SARS-CoV-2; the format of these data is shown in Table 2 . Each record contains the province, city, and hospital name of confirmed HCW infections. 

The crawled hospital data only contained the attribute information of the hospitals, not their geographic coordinates. Therefore, we used an application programming interface (Baidu, Inc., China) to match the longitudinal and latitudinal information for each hospital. To study the impact of the distance between a hospital and the infection source on HCW infections, we further calculated the straight-line distance between each hospital and the South China Seafood Market (Kang and Xu, 2020); the calculated results are represented by the "distance" throughout this work.

In addition to the hospital data, that on confirmed HCW infections also required preprocessing. The information on confirmed HCW infections obtained via retrospective analyses were mainly reported to the Chinese Red Cross Foundation in two ways (Foundation, 2020) : (1) directly by individuals to the foundation and (2) collected by their respective hospitals, which then relayed the information to the foundation. After examination and approval by the Chinese Red Cross Foundation, the relevant information was published on their website, after which time the hospital could review it. Those who failed to pass the review did not qualify for aid. Therefore, data records of unqualified persons were from the original data before analysis in this study. As the Chinese Red Cross Foundation not only aided infected HCWs, but also infected or diseased staff during the epidemic, such data on non-HCWs also required deletion. Finally, the "hospital name" field of confirmed HCW infections needed to be linked with the same field in the hospital data. As shown in Figure 2a , after data preprocessing, a total of 3,703 confirmed cases of HCW infections remained, including 3,655 from Hubei Province and 3,058 from Wuhan City. Compared with the real-time data on the confirmed number of SARS-CoV-2 infections among HCWs in the same period ( Figure 2b ) (Gao et al., 2020; Novel et al., 2020) , the corresponding data obtained via retrospective analyses yielded an obvious quantitative advantage.

The main objectives of this study were to clarify the spatiotemporal characteristics of SARS-CoV-2 infections among HCWs in Wuhan and to explore the interactions between related factors and the incidence of infection. The methodological framework of this study is shown in Figure 3 . First, two types of HCW infection inventories, based on the crawled HCW infection data, were constructed: the Grid-Level Healthcare Worker Infection Inventory and the Hospital-Level Healthcare Worker Infection Inventory. Using the Grid-Level Healthcare Worker Infection Inventory, we then explored the spatiotemporal distributions and spatial autocorrelation of HCW infections. Finally, based on the Hospital-Level Healthcare Worker Infection Inventory, a geographical detector technique was used to explore the impacts of hospital level, type, distance from the infection source (i.e., South China Seafood Market), and other environmental indicators on HCW infections. 

Although the data on confirmed infections among HCWs after cleaning recorded the basic information of HCW diagnoses, they did not include the number of HCWs who were diagnosed every day. Due to the incubation period of SARS-CoV-2, the date on which HCW was confirmed to be infected does not represent the date of infection. Therefore, it is difficult to visually describe the spatiotemporal distribution of HCW infections and external factors using these data alone. In view of this limitation, we constructed two HCW infection inventories. The Grid-Level Healthcare Worker Infection Inventory was used to explore the spatiotemporal distribution of infections among HCWs, while the Hospital-Level Healthcare Worker Infection Inventory was used to mine the relevant factors influencing HCW infections. The data structure for these inventories is shown in Figure 4 ; here,

( , ) is a spatiotemporal dataset, where S and T are the spatial and temporal domains, respectively, and

is the total number of spatial objects, and is the total number of timestamps. The construction of the infection inventories can be divided into two main steps: (1) generation of a "confirmed inventory," based on confirmed HCW infections, and (2) calculation of an "infection inventory" based on the "confirmed inventory." Taking the Grid-Level Healthcare Worker Infection Inventory as an example, the number of HCWs confirmed in each grid per day was first calculated using partitioning statistics, such that the number of confirmed HCWs in grid j on day 0 was calculated as:

where represents the total number of confirmed HCWs in China, represents the information for a specific HCW suffering from SARS-CoV-2, within represents a function to determine whether or not a hospital where the confirmed HCW was located was inside of grid j, and _ represents the Grid-Level Confirmed Healthcare Worker Inventory, which can reveal the changes in confirmed HCW infections in every district in Wuhan over time. The corresponding Hospital-Level Confirmed Healthcare Worker Inventory can be expressed as _ . To calculate the Grid-Level Healthcare Worker Infection Inventory, we assumed that the incubation time, , of SARS-CoV-2 exhibited a log-normal distribution, as seen in other acute respiratory viral infections (Lessler, Reich, Brookmeyer, Perl, Nelson and Cummings, 2009 ). Lauer, Grantz, Bi, Jones, Zheng, Meredith, Azman, Reich and Lessler (2020) found that the mean and standard deviation of the random variable, ln(X), were 1.621 and 0.418, respectively (i.e., ( ) ∼ (1.621, 0.418 2 )). Based on the probability distribution of X, the number of HCWs infected each day could be calculated. For example, the number of infected HCWs on day 0 in grid could be calculated as: 

where represents the probability that the incubation time is days; represents the maximum incubation time; since the incubation time of 98.7% patients is within 13 days, is set as 13; represents the mean of lognormal distribution, i.e. 1.621; represents the standard deviation of lognormal distribution, i.e. 0.418; and _ represents the Grid-Level Healthcare Worker Infection Inventory. Similarly, the Hospital-Level Healthcare Worker Infection Inventory can be expressed as _ .

The in ( , ) was essentially a collection of time-series data, which recorded the temporal characteristics of SARS-CoV-2 infections among HCWs. Previous studies have shown that the counts of more frequent infections may be assumed to be approximately normal. Therefore, we employed normal, gamma, and log-normal distributions to describe the distributional characteristics of infections among HCWs. The probability density functions of normal, log-normal, and gamma distributions are shown in Equations (3), (4), and (5), respectively:

(3)

where and represent the hyper parameters of normal, and represent the hyper parameters of lognormal, and represent the hyper parameters of gamma, and Γ( ) represents the gamma function. In this study, the maximum likelihood estimation was used to fit the hyper parameters of the three distributions. As gamma and lognormal are skew distribution, the median of the distribution was used to approximate the peak time of infection.

Trend analysis was used to measure the changes in HCW infections over time. In this study, the slope, , of the time series was used to quantitatively describe the growth rate of HCW infections. For example, the trend of infections in grid over the interval [ , ] was calculated as:

where _ ( , ) represents the number of HCWs infected on day in grid , _ ( , ) represents the number of HCWs infected on day in grid , and k represents the slope of (i.e., the changes in) infections among HCW. When > 0, the rate at which HCWs were infected increased within the interval. When < 0, the rate at which HCWs were infected declined within the interval.

There are two means of analyzing spatial autocorrelations -Global Moran's I and Local Moran's I. Here, Global Moran's I was used to measure the characteristics of spatial autocorrelation across an entire region, such that, within the study period, an index was assigned to the infection inventory to identify the overall spatial distribution of HCW infections. The significance of Global Moran's I was tested using standard statistics. For example, the Global Moran's I and over the interval [ , ] for the Grid-Level Healthcare Worker Infection Inventory were computed as:

where represents the Global Moran's I, and represent the cumulative number of HCWs infected in grids i and j over the interval [ , ] ; represents the mean number of HCWs infected in all grids within the interval [ , ]; represents the number of grids in the infection inventory, is the spatial weight matrix, where = 1 if two grids have a common edge or common vertex, otherwise, = 0, and [ ] and [ ] represent the mean and variance of the Global Moran's I, respectively. When > 1.65 or < -1.65, this indicates that the number of HCWs infected presents either an aggregated or discrete distribution in space, respectively. When -1.65< <1.65, the spatial relationship of HCW infections is not significant.

The Local Moran's I was used to measure the spatial autocorrelation characteristics of local areas. It allowed the spatial autocorrelation patterns of HCW infections to be calculated for each grid, so as to obtain the specific location of spatial aggregations or discrete phenomena. The Local Moran's I, denoted as over the interval [ , ] of grid , is shown in Equation (9).

Here, represents the Local Moran index of the -th grid, , , and carry the same meanings as in Equation (7), and can be divided into two parts: a descriptive variable, , and a spatial lag variable, ∑ =1, ≠ . When > 0 and ∑ =1, ≠ > 0, > 0, there is a positive correlation, indicating that grid i and the neighborhood have high distributions, consistent with a high-high cluster (HH). When > 0 and ∑ =1, ≠ < 0, < 0, there is a negative correlation, indicating that grid i and the neighborhood have a low distribution, consistent with a high-low outlier (HL) and when < 0 and ∑ =1, ≠ > 0, < 0, there is a negative correlation, indicating that grid i and the neighborhood have a high distribution, consistent with a low-high outlier (LH). Finally, when < 0 and ∑ =1, ≠ < 0, > 0, there is a positive correlation, indicating that grid i and the neighborhood have low distributions, consistent with a low-low cluster (LL).

Geographical detectors are a group of statistical methods used to detect the linear and nonlinear correlations between independent and dependent variables within a geospatial context. The advantage of these techniques is that they do not include linear hypothesis. Essentially, if an independent variable has a significant impact on a dependent variable, the spatial distribution of the two variables will tend to be the same. In this study, we mainly used factor detection and interaction detection to analyze the influencing factors of HCW infections.

Factor detection uses statistics with clear physical meanings to express the strength of a correlation between an independent and dependent variable (i.e., how much a given factor explains the dependent variables). The formula for calculating the q statistics is as follows:

where ℎ = 1, . . . , is the index of the layer (i.e., the classification or partition of the dependent variable, , or independent variable, ), ℎ and represent the number of elements in the layer ℎ and in the entire region, respectively, and 2 ℎ and 2 represent the variances of the dependent variable in layer ℎ and in the entire region, respectively. The range of values is [0, 1]. The larger the value of , the stronger the explanatory power of an independent variable is in comparison with that of the dependent variables. In an extreme cases, when = 1, this means that X completely controls the spatial distribution of and when = 0, it means that is not related to .

Interaction detection is used to evaluate the interactions between different factors on the basis of factor detection, that is, to evaluate whether or not factors 1 and 2 work together to increase or decrease the explanatory power of the dependent variable, . First, the values of the two factors, 1 and 2 to ( ( 1 ) and ( 2 )) must be calculated and then the value when the factors 1 and 2 interact ( 1 ∩ 2 ) must be determined before finally comparing

( 1 ), ( 2 ), and ( 1 ∩ 2 ). In the geographical detector employed here, the independent variables were necessarily categorical variables. Therefore, continuous variables were first converted into categorical variables. Natural discontinuities were used to divide the continuous variables into three categories. The results of this classification are shown in Table 3 . 

In order to assess the temporal characteristics of HCW infections, we first analyzed the statistical distributions and changes in HCW infections over time. Figure 5 shows the results fitted to normal, log-normal, and gamma distributions for HCW infections in Wuhan. The mean square error (MSE) was used to evaluate the fit of each distribution. The results showed that the MSE using a normal distribution was 2.34e-05, that using a gamma distribution was 3.46e-06, and that using a log-normal distribution was 2.04e-05, which indicated that a log-normal distribution was the most suitable to describe the evolution of HCW infections in Wuhan. Table 4 shows the statistical characteristics of HCW infections, for which 95% confidence intervals were calculated via bootstrap resampling. Based on the log-normal distribution, the mean day of infections in Wuhan was estimated to be January 23, 2020, according to the data on daily cases of HCW infections, and the estimated standard deviation was 11.8 days. Jan 23 Jan 22∼Jan 24 11.8 days 11.2∼12.3 days 2.04e-06 Gamma Jan 23 Jan 23∼Jan 24 11.6 days 11.1∼12.1 days 3.46e-06

We analyzed the changes in the number of daily HCW infections with time and the results are shown in Figure  6 . Overall, HCW infections in Wuhan mainly occurred between January 1 and February 29 of 2020. Within that period, infections rose from January 1-19, while from January 19 to February 29, they declined. Moreover, from January 23-28, the downward trend of HCW infections gradually decreased, and the infection rate curve showed an upward trend again between January 25 and January 28, when HCW infections fell again. There are two main reasons for this change. First, lockdown measures largely contained the spread of the SARS-CoV-2, but in a short period of time, it caused a surge in the number of patients infected, which increased the intensity of the work of medical staff, leading to increased infection rates among them. Secondly, as of January 28th, there were more than 6,000 HCWs in Hubei, which greatly eased the workload, thus decreasing the rates of infection. Figure 6b clearly shows the changes in the trend of HCW infections over time. The decreasing trend of HCW infections began to slow on January 21st. This time node is highly consistent with that determined by human-to-human transmission. In general, the temporal characteristics of HCW infections were highly correlated with the release times of news or policies. In particular, the release of isolation measures (e.g., lockdowns) may exacerbate HCW infections in the short term. Therefore, before the implementation of such policies, it is particularly important to strengthen the measures in place to protect HCWs from infection. 

As shown in Figure 6 , the infection of HCWs can be roughly divided into four stages: (1) January 1-19, (2) January 20-23, (3) January 24-28, and (4) January 29-February 29. Here, we show the spatial distributions and trends, and analyze the spatial autocorrelations during these four stages.

The spatial distributions of HCW infections are shown in Figure 7 . We found that the spatial distribution of HCW infections during the four stages showed the same trend, in that the rates of infection in the central urban area were high than those in the outer urban area. This indicates that infections among HCWs in Wuhan were extremely unbalanced and spatially heterogeneous. The areas with severe HCW infections were mainly located near the South China Seafood Market in the central urban area, as this market was a key source of the SARS-CoV-2 epidemic in Wuhan. However, there were slight differences in the spatial distributions of HCW infections among the four stages. Compared with the distribution of HCW infections from January 1-19, the rate of HCW infections in the outer city of Wuhan from January 20 to February 29 gradually increased.

The spatial trends of HCW infections are shown in Figure 8 . From January 1-19, the overall rate of HCW infection in Wuhan increased rapidly, with the fastest growth rate in the areas near the South China Seafood Market. From January 20-23, HCW infections exhibited a downward trend for the first time, among which the rate of decline was relatively fast in Wuchang and Jiangan. However, across Wuhan, the infection of HCWs continued to increase. From January 24-28, the rate of HCW infections in Wuhan showed a declining trend overall, while the rates of infection in Dongxihu, Caidian, Hannan, and Jiangxia showed an upward trend again. Combined with Figure 6 , these results demonstrate that intense measures, such as the lockdown of Wuhan, may suddenly increase the work load of HCWs in mild and non-infected areas, and thus increased the probability of HCW infections in outer city areas. From January 29 to February 29, the overall trend of HCW infection decreased, but there were still some areas with slight increases in rates of infection near the South China Seafood Market, indicating that HCWs near the infection source were always more susceptible to infection. 

The Global Moran's I was used to explore the global spatial dependency of the four stages of HCW infections, as shown in Table 5 . In each study interval, there was a significant spatial dependence of HCW infections, and with increases in time, this dependence gradually increased. From January 24-28, the spatial dependency was the strongest, where the Global Moran's I = 0.475, Z > 2.58, and p < 0.001, indicating that HCW infections in Wuhan were significantly and positively correlated at the grid level. In other words, grids with high numbers of infections among HCWs had higher numbers of HCW infections across the grid, while grids with low numbers of HCW infections had lower numbers of HCW infections across the grid. We used the Local Moran's I to explore the local spatial dependency of the four stages of HCW infections, as shown in Figure 9 . Three clustering patterns were observed in the four stages, namely HH, HL, and LH abnormal clustering. High-high clustering occurred in the infection hot spot area, which was mainly concentrated around the South China Seafood Market in the central city. High-low clustering was mainly distributed near the HH clustering, indicating that HCW infections were essentially controlled within a certain range. Abnormal LH clustering was mainly distributed in the districts of the outer city in a discrete manner, and the LH areas gradually increased over time, indicating that some areas of the outer city also had a high incidence of HCW infections. Combined, the Global and Local Moran's I values reveal that HCW infections exhibited regional agglomeration.

In this study, we used geographic detectors to explore the impacts of external environmental factors, such as hospital level, type, and distance from the infection source, on HCW infections in the four stages previously identified. The results are shown in Figure 10 . From a single-factor perspective, hospital level and type had greater impacts on HCW infections, with q-values were mainly falling between 0.18 and 0.35. Meanwhile, distance had a lesser impact on HCW infections, with -values ranging from 0.03-0.04. With respect to the interactions between two factors, hospital and hospital type worked together to explain HCW infections to the greatest extent. At the same time, while the distance had a small impact on HCW infections, the combination of distance and either hospital level or type could cause a significant increase in the -values, which indicates that, combined, these factors could explain HCW infections to a great degree. Additionally, with increases in time, the explanatory power of hospital level or type on HCW infections gradually increased, while the explanatory power of distance changed little.

To further explore the influences of hospital level, type, and distance from the infection source on HCW infections, the percentages of daily HCW infections to total HCW infections for each factor were computed ( Table 6 ). The results show that HCWs in third-class and general hospitals were more likely to be infected. While distance had little explanatory power, HCWs who were close to the infection source were still more vulnerable to infection. As shown in Figure 10 , HCWs at third-class and general hospitals closer to the infection source were at the greatest risk of infection. Table 6 The percentage of daily HCWs infections to total HCWs infections in each factor.

Jan 1st∼Jan 19th Jan 20th∼Jan 23rd Jan 24th∼Jan 28th Jan 29th∼Feb 29th 

As the first line of defense in the fight against the SARS-CoV-2 epidemic, HCWs help patients combat the virus, while exposing themselves to high concentrations at close range. It is important to study the spatiotemporal distributions and factors influencing SARS-CoV-2 infections among HCWs to better protect them. Existing studies related to HCW infections have largely emphasized infection rates and protective measures, while the spatiotemporal patterns and related external factors related to HCW infections have been unclear. To fill this gap in the existing knowledge, we first studied the spatiotemporal distributions infections of HCWs in Wuhan using statistical partitioning, curve-fitting, and spatial autocorrelation. The results showed that the lowest MSE was obtained using a log-normal distribution, which indicates that this distribution was the most suitable to describe the evolution of HCW infections over time. The mean date of infection was January 23rd, with a standard deviation of 10.8 days. The implementation of substantial measures, such as the lockdown of Wuhan, may have contributed to short-term increases in the probability of infection among HCWs, especially in outer city areas. This indicates that it is especially important to strengthen the protective measures for HCWs before such policies are implemented. The rate at which HCWs were infected in Wuhan displayed obvious spatial heterogeneity. The number of HCW infections was higher in the central city and lower in the outer city and the likelihood of infection exhibited significant spatial autocorrelation and spatial dependency.

To better understand the temporal and spatial distributions of HCW infections, geographical detector techniques were used to explore the impacts of external environmental factors, such as hospital level, type, and distance from the infection source on HCW infections. The results showed that hospital level and type significantly impacted the probability of HCW infections, among which third-class and general hospitals carried the greater risks of infection. This conclusion is similar to that of Zheng et al. (2020) , which independently validates the accuracy of our results to a certain extent. Moreover, although the distance from the infection source had little impact on HCW infections, the combined effect of distance and hospital level or type was shown to increase the explanatory power of HCW infections, that is, third-class and general hospitals closer to the infection source had the greatest risks of HCW infection. Therefore, investing more medical supplies into hospitals with higher risks of HCW infection is of vital importance.

This study had four primary limitations. First, the selection of influencing factors excluded factors that are known to contribute to the incidence of HCW infections, such as access to sufficient medical supplies, sufficient knowledge of protective protocols, and patient contact, among others. However, due to the difficulty of data acquisition, only three external factors -hospital rank, type and distance from the infection source -were considered. Secondly, the coverage of HCW infection data was narrow. Only data on HCW infections in China were used to explore related spatiotemporal characteristics and influencing factors; data from other countries were not included, which may limit the generalizability of our results. Third, the infection data for HCWs was not comprehensive. As the information on HCW infections is published on the official website of the Chinese Red Cross Foundation in batches, additional information may continue to be published in the future that could not be included here. Finally, we used the number of HCW infections instead of the infection rate, ignoring the heterogeneity of HCWs in Wuhan. In light to these limitations, future studies should focus on collecting additional domestic and foreign HCW infection data to more accurately and comprehensively analyze the spatiotemporal characteristics and factors influencing SARS-CoV-2 infections among HCWs.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Reasons for healthcare workers becoming infected with novel coronavirus disease 2019 (covid-19) in china

Local indicators of spatial association-lisa

Spatial and temporal analyses to investigate infectious disease transmission within healthcare settings

Clinical characteristics of coronavirus disease 2019 (covid-19): An updated systematic review

Protecting healthcare workers from sars-cov-2 infection: practical indications

Management rules of chinese red cross foundation bytedance healthcare workers humanitarian relief fund

Geo-temporal distribution of 1,688 chinese healthcare workers infected with covid-19 in severe conditions-a secondary data analysis

Clinical characteristics of coronavirus disease 2019 in china

Clinical features of patients infected with 2019 novel coronavirus in wuhan, china

Population flow drives spatio-temporal distribution of covid-19 in china

Effective control of sars-cov-2 transmission between healthcare workers during a period of diminished community prevalence of covid-19

The incubation period of coronavirus disease 2019 (covid-19) from publicly reported confirmed cases: estimation and application

Incubation periods of acute respiratory viral infections: a systematic review

Spatiotemporal patterns of covid-19 impact on human activities and environment in mainland china using nighttime light and air quality data

The impact of covid-19 on transport volume and freight capacity dynamics: An empirical analysis in german food retail logistics

Notes on continuous stochastic phenomena

Prevalence of sars-cov-2 antibodies in health care personnel in the new york city area

The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (covid-19) in china. Zhonghua liu xing bing xue za zhi= Zhonghua liuxingbingxue zazhi 41

Screening of healthcare workers for sars-cov-2 highlights the role of asymptomatic carriage in covid-19 transmission

Managing covid-19 in resource-limited settings: critical care considerations. Tan, L., 2020. Preventing the transmission of covid-19 amongst healthcare workers

Personal protective equipment and evidence-based advice for surgical departments during covid-19

Temporal effects of antibiotic use and hand rub consumption on the incidence of mrsa and clostridium difficile

Geographical detectors-based health risk assessment and its application in the neural tube defects study of the heshun region, china

Association between universal masking in a health care system and sars-cov-2 positivity among health care workers

Spatiotemporal characteristics of covid-19 epidemic in the united states

Assessment of mental health factors among health professionals depending on their contact with covid-19 patients

Tribute to health workers in china: A group of respectable population during the outbreak of the covid-19

Temporal-spatial analysis of severe acute respiratory syndrome among hospital inpatients

Risk assessment of covid-19 based on multisource data from a geographical viewpoint

Exploring the roles of high-speed train, air and coach services in the spread of covid-19 in china

Analysis of the infection status of the health care workers in wuhan during the covid-19 outbreak: A cross-sectional study

Development and implementation of a dynamic and 4d gis based on semantic location model