Advertisement

Spatial and temporal analysis of human infection with the avian influenza A (H7N9) virus in China and research on a risk assessment agent-based model

  • Dongqing Huang
    Affiliations
    School of Information Science and Technology, Yunnan Normal University, Kunming, 650500, China

    GIS Technology Engineering Research Centre for West-China Resources and Environment of Educational Ministry, Yunnan Normal University, Kunming, 650500, China
    Search for articles by this author
  • Wen Dong
    Correspondence
    Corresponding author at: Faculty Of Geography, Yunnan Normal University, Kunming, 650500, China. GIS Technology Engineering Research Centre for West-China Resources and Environment of Educational Ministry, Yunnan Normal University, Kunming, 650500, China.
    Affiliations
    GIS Technology Engineering Research Centre for West-China Resources and Environment of Educational Ministry, Yunnan Normal University, Kunming, 650500, China

    Faculty Of Geography, Yunnan Normal University, Kunming, 650500, China
    Search for articles by this author
  • Qian Wang
    Affiliations
    School of Information Science and Technology, Yunnan Normal University, Kunming, 650500, China

    GIS Technology Engineering Research Centre for West-China Resources and Environment of Educational Ministry, Yunnan Normal University, Kunming, 650500, China
    Search for articles by this author
Open AccessPublished:April 12, 2021DOI:https://doi.org/10.1016/j.ijid.2021.04.030

      Highlights

      • The H7N9 virus epidemic was mainly concentrated in the Yangtze River Delta and Pearl River Delta.
      • The epidemic had obvious seasonality: mostly in winter and spring.
      • The epidemic had a significant correlation with temperature and precipitation.
      • The agent-based model combined meteorological factors, time, and environmental factors.

      Abstract

      Objectives

      From 2013 to 2017, the avian influenza A (H7N9) virus frequently infected people in China, which seriously affected the public health of society. This study aimed to analyze the spatial characteristics of human infection with the H7N9 virus in China and assess the risk areas of the epidemic.

      Methods

      Using kernel density estimation, standard deviation ellipse analysis, spatial and temporal scanning cluster analysis, and Pearson correlation analysis, the spatial characteristics and possible risk factors of the epidemic were studied. Meteorological factors, time (month), and environmental factors were combined to establish an epidemic risk assessment proxy model to assess the risk range of an epidemic.

      Results

      The epidemic situation was significantly correlated with atmospheric pressure, temperature, and daily precipitation (P < 0.05), and there were six temporal and spatial clusters. The fitting accuracy of the epidemic risk assessment agent-based model for lower-risk, low-risk, medium-risk, and high-risk was 0.795, 0.672, 0.853, 0.825, respectively.

      Conclusions

      This H7N9 epidemic was found to have more outbreaks in winter and spring. It gradually spread to the inland areas of China. This model reflects the risk areas of human infection with the H7N9 virus.

      Keywords

      Introduction

      In February 2013, the first case of human infection with the avian influenza A (H7N9) virus occurred in Shanghai, China. This is an infectious disease caused by the avian influenza virus. Human infection with the H7N9 virus has a high mortality rate (
      • Chen Y.
      • Wen Y.
      Spatiotemporal distributions and dynamics of human infections with the A H7N9 avian influenza virus.
      ) and a high-risk assessment (
      • Iuliano A.D.
      • Jang Y.
      • Jones J.
      • Davis C.T.
      • Wentworth D.E.
      • Uyeki T.M.
      • et al.
      Increase in human infections with avian influenza A(H7N9) virus during the fifth epidemic - China, October 2016-February 2017.
      ), which has caused widespread concern (
      • Horby P.
      H7N9 is a virus worth worrying about.
      ,
      • Uyeki T.M.
      • Cox N.J.
      Global concerns regarding novel influenza A (H7N9) virus infections.
      ). Due to the characteristics of human-to-human transmission of the H7N9 virus, the epidemic has a risk of continuous emergence (
      • Li Z.
      • Fu J.
      • Lin G.
      • Jiang D.
      Spatiotemporal variation and hotspot detection of the avian influenza A(H7N9) virus in China, 2013–2017.
      ,
      • Tanner W.D.
      • Toth D.J.A.
      • Gundlapalli A.V.
      The pandemic potential of avian influenza A (H7N9) virus: a review.
      ,
      • Xiang N.
      • Li X.
      • Ren R.
      • Wang D.
      • Zhou S.
      • Greene C.M.
      • et al.
      Assessing change in avian influenza A (H7N9) virus infections during the fourth epidemic—China, September 2015–August 2016.
      ,
      • Xiang N.
      • Iuliano A.D.
      • Zhang Y.
      • Ren R.
      • Geng X.
      • Ye B.
      • et al.
      Comparison of the first three waves of avian influenza A (H7N9) virus circulation in the mainland of the People’s Republic of China.
      ). Until November 2017, human infections with the H7N9 virus frequently occurred in mainland China. Since the H7N9 virus spreads in poultry, and studies have shown that most people are infected due to close contact with live poultry (
      • Virlogeux V.
      • Feng L.
      • Tsang T.K.
      • Jiang H.
      • Fang V.J.
      • Qin Y.
      • et al.
      Evaluation of animal-to-human and human-to-human transmission of influenza A (H7N9) virus in China, 2013–15.
      ,
      • Wang D.
      • Yang L.
      • Zhu W.
      • Zhang Y.
      • Zou S.
      • Bo H.
      • et al.
      Two outbreak sources of influenza A (H7N9) viruses have been established in China.
      ,
      • Wu P.
      • Peng Z.
      • Fang V.J.
      • Feng L.
      • Tsang T.K.
      • Jiang H.
      • et al.
      Human infection with influenza A (H7N9) virus during 3 major epidemic waves, China, 2013–2015.
      ), the density of a live poultry market has a strong correlation with the H7N9 virus infection rate (
      • Gilbert M.
      • Golding N.
      • Zhou H.
      • Wint G.R.W.
      • Robinson T.P.
      • Tatem A.J.
      • et al.
      Predicting the risk of avian influenza A H7N9 infection in live-poultry markets across Asia.
      ). Some cities affected by the H7N9 virus permanently closed all live poultry markets in 2014 (
      • Wu P.
      • Peng Z.
      • Fang V.J.
      • Feng L.
      • Tsang T.K.
      • Jiang H.
      • et al.
      Human infection with influenza A (H7N9) virus during 3 major epidemic waves, China, 2013–2015.
      ). It shows that the H7N9 virus not only affects the development of the poultry industry but also brings great harm to the public health of society. It is particularly important to take effective measures to control the epidemic, especially in some potentially risky areas (
      • Shan X.
      • Lai S.
      • Liao H.
      • Li Z.
      • Lan Y.
      • Yang W.
      The epidemic potential of avian influenza A (H7N9) virus in humans in mainland China: a two-stage risk analysis.
      ).
      Because the H7N9 virus has seriously affected the public health of society, many studies have conducted research on the risk of H7N9 virus transmission in populations. Research has demonstrated that the spread of human infection with the H7N9 virus is importantly related to temperature and precipitation (
      • Fang L.-Q.
      • Li X.-L.
      • Liu K.
      • Li Y.-J.
      • Yao H.-W.
      • Liang S.
      • et al.
      Mapping spread and risk of avian influenza A (H7N9) in China.
      ,
      • Hu W.
      • Zhang W.
      • Huang X.
      • Clements A.
      • Mengersen K.
      • Tong S.
      Weather variability and influenza A (H7N9) transmission in Shanghai, China: a Bayesian spatial analysis.
      ,
      • Li X.-L.
      • Yang Y.
      • Sun Y.
      • Chen W.-J.
      • Sun R.-X.
      • Liu K.
      • et al.
      Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China.
      ). Boosted regression tree (BRT) (
      • Li X.-L.
      • Yang Y.
      • Sun Y.
      • Chen W.-J.
      • Sun R.-X.
      • Liu K.
      • et al.
      Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China.
      ), species distribution models (SDMs) (
      • Bui C.M.
      • Gardner L.
      • MacIntyre R.
      • Sarkar S.
      Influenza A H5N1 and H7N9 in China: a spatial risk analysis.
      ), and spatial stratified heterogeneity analyses (
      • Ge E.
      • Zhang R.
      • Li D.
      • Wei X.
      • Wang X.
      • Lai P.-C.
      Estimating risks of inapparent avian exposure for human infection: avian influenza virus A (H7N9) in Zhejiang Province, China.
      ) are used to spatially predict the main risk areas of human infection with the H7N9 virus. Some researchers have proposed a dynamic mathematical model between wild and domestic birds and from birds to humans (
      • Zhang X.
      • Zou L.
      • Chen J.
      • Fang Y.
      • Huang J.
      • Zhang J.
      • et al.
      Avian influenza A H7N9 virus has been established in China.
      ), and H7N9 avian influenza virus transmission dynamics between humans and poultry (
      • Liu Z.
      • Fang C.-T.
      A modeling study of human infections with avian influenza A H7N9 virus in mainland China.
      ). These models explain the transmission mode of human infection with the H7N9 virus.
      The purpose of this research was to conduct statistical analysis and spatial and temporal analysis of the H7N9 virus case data of people in China from February 2013 to November 2017, and propose a risk assessment agent model. The model used meteorological (such as air pressure, temperature, daily precipitation, mean water pressure, sunshine hours, maximum daily precipitation) and environmental factors (population, gross output value of farming, poultry output, number of slaughter poultry, distance to water, NDVI) to identify the risk areas for human infection with the H7N9 virus in mainland China. The research results could provide decision-making assistance for defense and monitoring of the epidemic.

      Materials and methods

      Data collection

      Data on human infections with the H7N9 virus in China, from February 2013 to November 2017, were retrieved from the Chinese Center for Disease Control and Prevention (http://www.chinacdc.cn). Meteorological data were provided by the China Meteorological Network (http://data.cma.cn/), from February 2013 to November 2017, including: minimum air pressure, minimum temperature, daily precipitation, average relative humidity, maximum wind speed, and sunshine hours. Data on the population, gross output value of farming, poultry output, and the number of slaughtered poultry in China from 2013 to 2017 were provided by the National Bureau of Statistics of China (http://www.stats.gov.cn). Distance to water was calculated as the distance to the nearest water based on the location of the case point on the vector map of lakes and rivers in China (from Google Earth satellite images). The Normalized Difference Vegetation Index (NDVI) data came from the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences (http://www.resdc.cn).

      Kernel density estimation

      Kernel density estimation is a non-parametric statistical interpolation method that converts point events into density surfaces (
      • Hu Y.
      • Wang F.
      • Guin C.
      • Zhu H.
      A spatio-temporal kernel density estimation framework for predictive crime hotspot mapping and evaluation.
      ). The observed data points are then fitted by the kernel function, and the local density around the data points is studied (
      • Bielecka E.
      • Pokonieczny K.
      • Borkowska S.
      GIScience theory based assessment of spatial disparity of geodetic control points location.
      ,
      • Gomes EC de S
      • Mesquitta MC da S
      • Wanderley L.B.
      • de Melo FL
      • Guimarães RJ de PS
      • Barbosa C.S.
      • et al.
      Spatial risk analysis on occurrences and dispersal of Biomphalaria straminea in and endemic area for schistosomiasis.
      ). Finally, a simulated actual probability distribution curve is obtained. In this study, the kernel density estimation method in 10.8 ArcGIS spatial analyses was used to simulate the risk distribution of human infection with the H7N9 virus in China. The number of cases of human infection with the H7N9 virus was used as the input of the kernel density estimation method to obtain a preliminary risk map of the H7N9 virus in China, which was used as the basis for subsequent research.

      Analysis of the temporal and spatial trend of the epidemic

      Standard deviation ellipse analysis was used to: analyze the trend of a group of points or regions; calculate the standard distance in the x and y directions; determine the major axis and minor axis of the ellipse; and analyze the geographical features. This study used the standard deviation ellipse analysis method in ArcGIS 10.8 to explore the temporal and spatial trends of human infection with the H7N9 virus in China, and analyze the spatial characteristics of the epidemic: central tendency, dispersion, and directional trends.

      Spatial and temporal scanning cluster analysis

      To detect possible statistically significant spatiotemporal gathering areas in the epidemic, this study used SaTScan 9.6 software to perform spatiotemporal scanning statistical analysis on human cases of the H7N9 virus in China. It was able to obtain the spatiotemporal gathering area (P < 0.05) of human infection with the H7N9 virus in mainland China.

      Agent-based model for epidemic risk assessment

      The epidemic risk assessment agent-based model consisted of three parts (Figure 1). The Meteorological Risk Module (MRM) evaluated the epidemic risk based on meteorological factors. In this study, meteorological factors with significant correlation with the epidemic situation were selected through correlation analysis as the input of the MRM analysis. The Environmental Risk Module (ERM) evaluated epidemic risks based on environmental factors. In this study, six impact factors – population, the gross output value of farming, poultry output, number of slaughter poultry, distance to water, and NDVI – were selected to construct an environmental risk module. The risk decision module (RDM) interacted with the MRM and ERM modules, perceived the factors in the module, and established epidemic risk decision rules through the multilayer perceptron (MLP) neural network, eventually dynamically reflecting changes in the epidemic risk. Based on the Repast Simphony 2.8 (
      • North M.J.
      • Collier N.T.
      • Ozik J.
      • Tatara E.R.
      • Macal C.M.
      • Bragen M.
      • et al.
      Complex adaptive systems modeling with Repast Simphony.
      ) platform, this study established an agent-based model to assess the risk assessment of human infection with the H7N9 virus in mainland China.
      Figure 1
      Figure 1Structure diagrams of the epidemic risk assessment agent-based model.

      Results

      General trend of human infections with the H7N9 virus

      Monthly statistics on human infection with the H7N9 virus in various provinces and autonomous regions in China were gathered from February 2013 to November 2017 (Figure 2). The results showed that 27 provinces and autonomous regions had an epidemic, with a total of 1490 cases. Among them were: 155 (7.72%, 95% CI 7.33%–8.11%) cases in 2013; 326 (21.88%, 95% CI 20.79%–22.97%) cases in 2014; 196 (13.15%, 95% CI 12.49%–13.81%) cases in 2015; 260 (17.45%, 95% CI 16.58%–18.32%) cases in 2016; and 553 (37.11%, 95% CI: 35.25%–38.97%) cases in 2017. In 2017, the H7N9 virus outbreak in China was more serious than the previous four years, and it lasted longer than the preceding four years. The outbreak of human infection with the H7N9 virus has obvious seasonality, with winter and spring being the peak periods. The cumulative number of cases in December each year and from January to March of the following year was 1106, accounting for 74.23% of the total number of epidemic cases in the country. The eastern and southeastern coastal areas of China were most affected by the H7N9 virus. There were 255 cases in Jiangsu Province, 258 cases in Guangdong Province, and 305 cases in Zhejiang Province.
      Figure 2
      Figure 2The number of H7N9 virus human cases in various provinces and autonomous regions in China from 2013 to 2017.

      Kernel density estimation

      According to the kernel density estimation results from 2013 to 2017 (Figure 3), human infection with the H7N9 virus had obvious spatial accumulation. From 2013 to 2017, Suzhou, Wuxi, Shanghai and its surrounding areas were high-risk areas, and large-scale human infections with H7N9 avian influenza occurred. In 2014 and 2015, Shantou, Huizhou and its surrounding areas in Guangdong Province were high-risk areas. In 2017, areas affected by the H7N9 virus in mainland China expanded greatly, and high-risk areas increased, mainly including southern Jiangsu Province, Anhui Province, northern Zhejiang Province, and northern Jiangxi Province.
      Figure 3
      Figure 3Estimated kernel density estimation risk maps of human infection with the H7N9 virus in China from 2013 to 2017.

      Standard deviational ellipse

      According to the visualization results of the standard deviation ellipse analysis (Figure 4) and the parameter table of the standard deviation ellipse (Table 1), human infection with the H7N9 virus in mainland China from 2013 to 2017 was mainly concentrated in the Yangtze River Delta and Pearl River Delta. Over time, the flu epidemic had a distinct tendency to spread to inland areas of China. In 2013, the epidemic spread from Jiangsu Province to Guangdong Province. In the first half of 2014, the distribution of influenza epidemics was the same as the previous year, mainly in Jiangsu Province, Guangdong Province, and the southeast coastal areas in between. However, in the second half of 2014, the influenza epidemic spread from the southeastern coastal areas of China to the inland areas of China. In 2015, the flu epidemic mainly occurred in the Yangtze River Delta and Pearl River Delta regions of China. In 2016, the epidemic trended from the southeast coastal areas to northern China. In 2017, influenza epidemics occurred in most parts of mainland China, mainly in the southeast coastal and central regions of China.
      Figure 4
      Figure 4Standard deviation ellipse analysis charts of human infection with the H7N9 virus in China from 2013 to 2017.
      Table 1The parameter table of the standard deviation ellipse analysis, followed by the Center X, Center Y coordinates of the center mean of the ellipse, XStdDist (major axis), YStdDist (short axis) and the direction of the ellipse rotation.
      MonthCenter XCenter YXStdDistYStdDistRotation
      2013/031378317.19634019161.8963208133.527773521.6426120.3270
      2013/041333021.45343904349.3198288507.0557335923.2121179.0448
      2013/101405081.52713765952.507826589.7299567399.310630.6694
      2013/121075784.08723277047.7523146238.9441831661.276832.3199
      2014/011219724.04463507420.5277247652.7684719612.389033.01015
      2014/021046645.12083448761.3494266354.3498809799.041323.89518
      2014/031143465.42133567137.4001196663.5612634305.522215.32756
      2014/041252686.58433857551.3079226619.0533977840.432322.91790
      2014/051174164.58403625610.1029110574.5481815955.612716.29187
      2014/111100612.82983477396.60852007648.769182436.7854128.1733
      2014/12945731.221133773240.38341663284.892456496.0260126.0826
      2015/011196071.80893360152.6421540451.8931683872.0383158.4022
      2015/021150549.61613466806.7436276561.1127635312.557218.2407
      2015/031395639.01193633136.7243150823.3433576861.981721.5549
      2015/041198602.12653992233.4505268999.8272223008.6043114.4283
      2015/051314676.37893940769.9370202591.2234419171.5422155.0412
      2015/091431788.78043873562.225414207.92347125795.67711.22428
      2015/111343158.45953667250.057022954.40760551870.527819.23826
      2015/121338042.76813639623.8283115389.9658659759.768526.4434
      2016/011259381.32843681873.7188240905.3073598434.338341.36630
      2016/021208916.38323567977.2609266634.454673512.354818.34036
      2016/031274914.49223736220.4102333630.2359483558.92957.68872
      2016/041278452.58133938739.7491132842.7306513700.137025.1128
      2016/051046279.76384046340.6724247451.0134526914.2728170.7298
      2016/061244866.19914303213.4211262091.11511036573.527165.8901
      2016/111412770.76433862312.186186541.93885484646.209414.26876
      2016/121285448.11883807475.8261268732.8412583700.033724.60533
      2017/011129310.07233715905.3415417988.8612561584.200936.80384
      2017/02959770.931863683194.7732497715.3064705958.595034.81136
      2017/03515998.885833590073.31491300818.123517959.457897.44345
      2017/04417707.061144250522.2288996643.0224789229.089167.74492
      2017/05688716.514294277927.7089834340.9168639851.979850.89942
      2017/06474562.946143439544.7357949603.37181067291.051149.8244
      2017/071255930.82123730680.925078409.54115860787.6189171.3683
      2017/08115352.887374182068.11341690148.411451802.4114127.3045

      Spatial and temporal scanning cluster analysis

      Spatial and temporal scan cluster analysis results (Figure 5) showed that there are six clusters, all of which are statistically significant (P < 0.01). These clusters indicate that vast influenza cases occurred during the gathering time and in the gathering area. Cluster 1 indicated that the gathering time was from April 2017 to September 2017, and the gathering area involved Beijing, Tianjin, Liaoning Province, Inner Mongolia, Hebei Province, Shanxi Province, and Shandong Province. Cluster 2 suggested that the gathering time was from January 2017 to March 2017, and the gathering area involved Hubei Province, Henan Province, Chongqing City, Hunan Province, Jiangxi Province, and Anhui Province. Cluster 3 gathered from January 2017 to November 2017, and the gathering area mainly involved Shanghai and its surrounding areas. Cluster 4 suggested that the gathering time was from April 2017 to June 2017, and the gathering area involved Sichuan Province, Yunnan Province, and Guizhou Province. Cluster 5 indicated that the gathering time was from July 2013 to March 2015, and the gathering area involved Guangdong Province, Fujian Province, and Jiangxi Province. Cluster 6 indicated that the gathering time was from July 2014 to December 2014, and the gathering area was mainly Urumqi, Xinjiang. It showed that human infection with the H7N9 virus occurred in most parts of China in 2017, and the clusters were mainly located in the Central China Plain, Southeast Coast, and North China Plain.
      Figure 5
      Figure 5Spatial and temporal analysis results of human infection with the H7N9 virus in China from 2013 to 2017.

      Correlation analysis

      The results of the single-factor correlation analysis (Table 2) showed that human infection with the H7N9 virus had a positive correlation with the lowest air pressure, maximum air pressure, and average air pressure (P < 0.05). The strongest positive correlation was with maximum air pressure. The influenza epidemic of human infection with the H7N9 virus was negatively correlated with minimum temperature, maximum temperature, daily precipitation, average temperature, mean water pressure, average minimum temperature, average maximum temperature, sunshine hours, and maximum daily precipitation (P < 0.05). The negative correlation with the average minimum temperature was the strongest.
      Table 2Correlation between the number of human infections with the H7N9 virus and meteorological factors.
      Pearson CorrelationP95% CI
      LowerUpper
      Lowest air pressure0.221<0.0010.0560.105
      Minimum temperature−0.293<0.0010.054−0.398
      Maximum air pressure0.263<0.0010.0540.157
      Maximum temperature−0.280<0.0010.054−0.388
      Daily precipitation−0.1490.0110.058−0.259
      Average air pressure0.259<0.0010.0540.150
      Average temperature−0.323<0.0010.053−0.430
      Mean water pressure−0.282<0.0010.055−0.386
      Average minimum temperature−0.312<0.0010.054−0.418
      Average maximum temperature−0.314<0.0010.054−0.420
      Sunshine hours−0.1380.0190.057−0.250
      Maximum daily precipitation−0.1760.0030.058−0.282

      Establishment of a risk assessment agent-based model

      Based on the Repast Simphony 2.8 simulation platform, this study selected 12 relevant meteorological factors through correlation analysis to construct an MRM. Selected meteorological factors were lowest air pressure, minimum temperature, maximum air pressure, maximum temperature, daily precipitation, average air pressure, average temperature, mean water pressure, average minimum temperature, average maximum temperature, sunshine hours, and maximum daily precipitation. In this study, six impact factors of population, the gross output value of farming, poultry output, number of slaughter poultry, distance to water, and NDVI were used to construct an ERM. In this study, the MLP neural network model was used to determine the risk decision rules, and the RDM was constructed. In addition, because the influenza epidemic had obvious seasonality and explosive growth in winter and spring, this study added time (month) as an impact factor in the RDM to improve the fitting effect of the model.
      The model divided the risk areas according to the number of epidemic cases per month, which consisted of four categories (MC: monthly cumulative number of cases): high-risk areas (MC > 10), medium-risk areas (5 < MC ≤10), low-risk areas (1 < MC ≤5), and lower-risk areas (MC = 1). In this study, 70% of the data were used as the training model and 30% as the test data. According to the ROC curve (Figure 6), the agent-based risk assessment model had high-effect fitting. The fitting accuracy of the four types of areas of high-risk, medium-risk, low-risk, and lower-risk was 0.825, 0.853, 0.672, and 0.795, respectively. The standardization results according to the importance of each factor in the model (Figure 7) showed that the mean water pressure factor was the most important to the model. The running process diagram of the risk assessment agent-based model was exported to the repast platform and visualized in ArcGIS10.8 software (Figure 8). During the operation of the model, the number of red dots was used to indicate the risk of the flu epidemic. In this study, ArcGIS 10.8 software was used to compare the actual situation of influenza epidemics at different times with the estimated results of the model (Figure 9). Obviously, this model had a high fitting effect on the risk areas of human infection with the H7N9 virus in mainland China.
      Figure 6
      Figure 6A ROC curve diagram of the fitting accuracy of the model. The fitting accuracy was high-risk: 0.825, medium-risk: 0.853, low-risk: 0.672, and lower-risk: 0.795.
      Figure 7
      Figure 7The standardized importance of each parameter in the risk assessment agent-based model.
      Figure 8
      Figure 8A process diagram of the execution of the risk assessment agent-based model. The number of red dots indicates the degree of risk of human infection with the H7N9 virus in the area.
      Figure 9
      Figure 9Comparison diagram of simulated risk and the real risk of the risk assessment agent-based model. A) December 2015; B) January 2016; C) February 2016; D) December 2016; E) February 2017. MC represents the cumulative number of cases in a month. High risk: MC > 10; medium risk: 5 < MC ≤10; low risk: 1 < MC ≤5; lower risk: MC = 1.

      Discussion

      From February 2013 to November 2017 there were five outbreaks of H7N9 virus in people in China. This is an infectious disease caused by the avian influenza virus, which has attracted widespread attention (
      • Horby P.
      H7N9 is a virus worth worrying about.
      ,
      • Uyeki T.M.
      • Cox N.J.
      Global concerns regarding novel influenza A (H7N9) virus infections.
      ). The overall trend analysis and kernel density estimation results of the epidemic case data in this study show that there were five outbreaks of flu in China. The epidemic first broke out in Shanghai in February 2013. In 2014 and 2015, the Yangtze River Delta and Pearl River Delta were high-risk areas. In 2016, the epidemic was under control and the epidemic was mainly concentrated in the Yangtze River Delta. In 2017, the epidemic spread to inland areas of China, which is consistent with previous research conclusions (
      • Bui C.M.
      • Gardner L.
      • MacIntyre R.
      • Sarkar S.
      Influenza A H5N1 and H7N9 in China: a spatial risk analysis.
      ,
      • Chen Y.
      • Wen Y.
      Spatiotemporal distributions and dynamics of human infections with the A H7N9 avian influenza virus.
      ,
      • Cowling B.J.
      • Jin L.
      • Lau E.H.Y.
      • Liao Q.
      • Wu P.
      • Jiang H.
      • et al.
      Comparative epidemiology of human infections with avian influenza A H7N9 and H5N1 viruses in China: a population-based study of laboratory-confirmed cases.
      ,
      • Shan X.
      • Wang Y.
      • Song R.
      • Wei W.
      • Liao H.
      • Huang H.
      • et al.
      Spatial and temporal clusters of avian influenza a (H7N9) virus in humans across five epidemics in mainland China: an epidemiological study of laboratory-confirmed cases.
      ). The flu epidemic spread to central China, as well as parts of the western and northern regions such as Sichuan Province and Beijing.
      The number of influenza epidemic cases peaked in the winter and spring each year, while the number of cases significantly decreased in the summer and autumn. This shows that the flu epidemic is controlled, to a certain extent, in summer and autumn. Risk trend analysis results demonstrated that the most severe epidemic areas are the Yangtze River Delta and Pearl River Delta, which are consistent with other studies (
      • Wang D.
      • Yang L.
      • Zhu W.
      • Zhang Y.
      • Zou S.
      • Bo H.
      • et al.
      Two outbreak sources of influenza A (H7N9) viruses have been established in China.
      ). The three provinces with the largest number of influenza cases are Jiangsu, Guangdong, and Zhejiang.
      A standard ellipse analysis was performed on the monthly influenza epidemic data, and the results showed that the 2013 epidemic spread among Shanghai, Zhejiang, and Guangdong provinces. In the first half of 2014, the epidemic mainly spread in the southeast coastal areas, and in the second half of the year there was a clear trend of spreading inland. Influenza epidemics in 2015 and 2016 were mainly spread in the southeast coastal area, but in the second half of 2016 there was a tendency to spread to northern China. In 2017, the flu epidemic gradually spread to the central and northwestern regions of China. The results of the spatial and temporal analysis also showed that the 2017 epidemic had significant spatiotemporal aggregation in China's North China Plain, Central China Plain, and Southwest China. This shows that the flu epidemic spread widely in 2017 and had a great impact on most parts of China.
      The results of the overall trend analysis, kernel density estimation analysis, and standard ellipse analysis in this study all indicate that the epidemic is mainly concentrated in the southeastern region of China and spreads within the region. This may be due to the developed economy, large population, and high mobility of China’s southeast coastal areas. The flu epidemic has existed in China for a long time and has a wide range of distribution. The number of cases is higher in winter and spring than in other seasons. This may be due to the large population flow during the Spring Festival in China, and the mass migration of people from the southeast coast moving to the inland areas of China, which intensified the spread of the epidemic.
      The analysis of the correlation between influenza epidemics and meteorological factors in the study showed that lowest air pressure, minimum temperature, maximum air pressure, maximum temperature, daily precipitation, average air pressure, average temperature, mean water pressure, average minimum temperature, average maximum temperature, sunshine hours, and maximum daily precipitation may be important factors affecting the spread of the epidemic. This is consistent with other research conclusions (
      • Fang L.-Q.
      • Li X.-L.
      • Liu K.
      • Li Y.-J.
      • Yao H.-W.
      • Liang S.
      • et al.
      Mapping spread and risk of avian influenza A (H7N9) in China.
      ,
      • Hu W.
      • Zhang W.
      • Huang X.
      • Clements A.
      • Mengersen K.
      • Tong S.
      Weather variability and influenza A (H7N9) transmission in Shanghai, China: a Bayesian spatial analysis.
      ,
      • Li X.-L.
      • Yang Y.
      • Sun Y.
      • Chen W.-J.
      • Sun R.-X.
      • Liu K.
      • et al.
      Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China.
      ,
      • Zhang P.
      • Wang J.
      • Atkinson P.M.
      Identifying the spatio-temporal risk variability of avian influenza A H7N9 in China.
      ) that temperature and precipitation are important risk factors for human infection with the H7N9 virus. This shows that the cold climate in winter and spring, and the large temperature difference may aggravate the spread of the H7N9 virus. This is also in line with the fact that the outbreak of the epidemic in winter and spring in China has occurred on a large scale.
      Compared with other related research, the risk assessment agent-based model set out in the present study established, combined with relevant research conclusions, the impact of temperature and precipitation (
      • Fang L.-Q.
      • Li X.-L.
      • Liu K.
      • Li Y.-J.
      • Yao H.-W.
      • Liang S.
      • et al.
      Mapping spread and risk of avian influenza A (H7N9) in China.
      ,
      • Hu W.
      • Zhang W.
      • Huang X.
      • Clements A.
      • Mengersen K.
      • Tong S.
      Weather variability and influenza A (H7N9) transmission in Shanghai, China: a Bayesian spatial analysis.
      ,
      • Li X.-L.
      • Yang Y.
      • Sun Y.
      • Chen W.-J.
      • Sun R.-X.
      • Liu K.
      • et al.
      Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China.
      ) and live poultry (
      • Virlogeux V.
      • Feng L.
      • Tsang T.K.
      • Jiang H.
      • Fang V.J.
      • Qin Y.
      • et al.
      Evaluation of animal-to-human and human-to-human transmission of influenza A (H7N9) virus in China, 2013–15.
      ,
      • Wang D.
      • Yang L.
      • Zhu W.
      • Zhang Y.
      • Zou S.
      • Bo H.
      • et al.
      Two outbreak sources of influenza A (H7N9) viruses have been established in China.
      ,
      • Wu P.
      • Peng Z.
      • Fang V.J.
      • Feng L.
      • Tsang T.K.
      • Jiang H.
      • et al.
      Human infection with influenza A (H7N9) virus during 3 major epidemic waves, China, 2013–2015.
      ) on the spread of the epidemic. It also added meteorological factors and environmental factors into the model. Due to the seasonality of the epidemic case data, time (months) was also added to the model as an impact factor. Due to the complex relationship between the spread of the influenza epidemic and other factors, this study adopted the MLP neural network with better nonlinear fitting to establish the interaction rules of the model to improve the fitting effect of the model. The model effectively reflected the risk areas of human infection with the H7N9 virus under different weather conditions, time conditions (months), and environmental factors, and provided help for the defense and monitoring of influenza epidemics.

      Conclusion

      This study found that human infection with the avian influenza A (H7N9) virus in China has obvious seasonality, and the high-risk areas for influenza epidemics were mainly in the Yangtze River Delta and Pearl River Delta. With time, the influenza epidemic gradually spread to inland areas of China. In 2017, there were multiple temporal and spatial clusters of influenza epidemics in the Chinese mainland. The risk assessment agent-based model proposed in this study achieved a high fitting effect on the epidemic, and considered factors such as meteorological conditions, time conditions (months), and environmental factors, which can provide effective help for controlling the epidemic.

      Contributions

      Dongqing Huang: conceptualization, methodology, software, data curation, writing - original draft. Wen Dong: supervision, funding acquisition. Qian Wang: investigation, data curation, writing - original draft.

      Funding

      The study was supported by the National Natural Science Foundation of China (Grant No. 41661087 ).

      Conflict of interest

      None of the authors have any conflicts of interest to declare.

      Ethical approval

      Approval was not required.

      Acknowledgments

      The study was supported by the National Natural Science Foundation of China (Grant No. 41661087). The authors express their sincere gratitude for the funding of this research.

      References

        • Bielecka E.
        • Pokonieczny K.
        • Borkowska S.
        GIScience theory based assessment of spatial disparity of geodetic control points location.
        ISPRS Int J Geo-Information. 2020; 9: 148
        • Bui C.M.
        • Gardner L.
        • MacIntyre R.
        • Sarkar S.
        Influenza A H5N1 and H7N9 in China: a spatial risk analysis.
        PLoS One. 2017; 12e0174980
        • Chen Y.
        • Wen Y.
        Spatiotemporal distributions and dynamics of human infections with the A H7N9 avian influenza virus.
        Comput Math Methods Med. 2019; 20199248246https://doi.org/10.1155/2019/9248246
        • Cowling B.J.
        • Jin L.
        • Lau E.H.Y.
        • Liao Q.
        • Wu P.
        • Jiang H.
        • et al.
        Comparative epidemiology of human infections with avian influenza A H7N9 and H5N1 viruses in China: a population-based study of laboratory-confirmed cases.
        Lancet. 2013; 382: 129-137
        • Fang L.-Q.
        • Li X.-L.
        • Liu K.
        • Li Y.-J.
        • Yao H.-W.
        • Liang S.
        • et al.
        Mapping spread and risk of avian influenza A (H7N9) in China.
        Sci Rep. 2013; 3: 1-8
        • Ge E.
        • Zhang R.
        • Li D.
        • Wei X.
        • Wang X.
        • Lai P.-C.
        Estimating risks of inapparent avian exposure for human infection: avian influenza virus A (H7N9) in Zhejiang Province, China.
        Sci Rep. 2017; 7: 1-8
        • Gilbert M.
        • Golding N.
        • Zhou H.
        • Wint G.R.W.
        • Robinson T.P.
        • Tatem A.J.
        • et al.
        Predicting the risk of avian influenza A H7N9 infection in live-poultry markets across Asia.
        Nat Commun. 2014; 5: 1-7
        • Gomes EC de S
        • Mesquitta MC da S
        • Wanderley L.B.
        • de Melo FL
        • Guimarães RJ de PS
        • Barbosa C.S.
        • et al.
        Spatial risk analysis on occurrences and dispersal of Biomphalaria straminea in and endemic area for schistosomiasis.
        J Vector Borne Dis. 2018; 55: 208-214
        • Horby P.
        H7N9 is a virus worth worrying about.
        Nat News. 2013; 496: 399
        • Hu W.
        • Zhang W.
        • Huang X.
        • Clements A.
        • Mengersen K.
        • Tong S.
        Weather variability and influenza A (H7N9) transmission in Shanghai, China: a Bayesian spatial analysis.
        Environ Res. 2015; 136: 405-412
        • Hu Y.
        • Wang F.
        • Guin C.
        • Zhu H.
        A spatio-temporal kernel density estimation framework for predictive crime hotspot mapping and evaluation.
        Appl Geogr. 2018; 99: 89-97
        • Iuliano A.D.
        • Jang Y.
        • Jones J.
        • Davis C.T.
        • Wentworth D.E.
        • Uyeki T.M.
        • et al.
        Increase in human infections with avian influenza A(H7N9) virus during the fifth epidemic - China, October 2016-February 2017.
        MMWR Morb Mortal Wkly Rep. 2017; 66: 254-255https://doi.org/10.15585/mmwr.mm6609e2
        • Li X.-L.
        • Yang Y.
        • Sun Y.
        • Chen W.-J.
        • Sun R.-X.
        • Liu K.
        • et al.
        Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China.
        Sci Rep. 2015; 5: 1-8
        • Li Z.
        • Fu J.
        • Lin G.
        • Jiang D.
        Spatiotemporal variation and hotspot detection of the avian influenza A(H7N9) virus in China, 2013–2017.
        Int J Environ Res Public Health. 2019; 16https://doi.org/10.3390/ijerph16040648
        • Liu Z.
        • Fang C.-T.
        A modeling study of human infections with avian influenza A H7N9 virus in mainland China.
        Int J Infect Dis. 2015; 41: 73-78
        • North M.J.
        • Collier N.T.
        • Ozik J.
        • Tatara E.R.
        • Macal C.M.
        • Bragen M.
        • et al.
        Complex adaptive systems modeling with Repast Simphony.
        Complex Adapt Syst Model. 2013; 1: 1-26
        • Shan X.
        • Lai S.
        • Liao H.
        • Li Z.
        • Lan Y.
        • Yang W.
        The epidemic potential of avian influenza A (H7N9) virus in humans in mainland China: a two-stage risk analysis.
        PLoS One. 2019; 14e0215857
        • Shan X.
        • Wang Y.
        • Song R.
        • Wei W.
        • Liao H.
        • Huang H.
        • et al.
        Spatial and temporal clusters of avian influenza a (H7N9) virus in humans across five epidemics in mainland China: an epidemiological study of laboratory-confirmed cases.
        BMC Infect Dis. 2020; 20: 1-7
        • Tanner W.D.
        • Toth D.J.A.
        • Gundlapalli A.V.
        The pandemic potential of avian influenza A (H7N9) virus: a review.
        Epidemiol Infect. 2015; 143: 3359-3374
        • Uyeki T.M.
        • Cox N.J.
        Global concerns regarding novel influenza A (H7N9) virus infections.
        N Engl J Med. 2013; 368: 1862-1864
        • Virlogeux V.
        • Feng L.
        • Tsang T.K.
        • Jiang H.
        • Fang V.J.
        • Qin Y.
        • et al.
        Evaluation of animal-to-human and human-to-human transmission of influenza A (H7N9) virus in China, 2013–15.
        Sci Rep. 2018; 8: 1-7
        • Wang D.
        • Yang L.
        • Zhu W.
        • Zhang Y.
        • Zou S.
        • Bo H.
        • et al.
        Two outbreak sources of influenza A (H7N9) viruses have been established in China.
        J Virol. 2016; 90: 5561-5573
        • Wu P.
        • Peng Z.
        • Fang V.J.
        • Feng L.
        • Tsang T.K.
        • Jiang H.
        • et al.
        Human infection with influenza A (H7N9) virus during 3 major epidemic waves, China, 2013–2015.
        Emerg Infect Dis. 2016; 22: 964
        • Xiang N.
        • Iuliano A.D.
        • Zhang Y.
        • Ren R.
        • Geng X.
        • Ye B.
        • et al.
        Comparison of the first three waves of avian influenza A (H7N9) virus circulation in the mainland of the People’s Republic of China.
        BMC Infect Dis. 2016; 16: 1-12
        • Xiang N.
        • Li X.
        • Ren R.
        • Wang D.
        • Zhou S.
        • Greene C.M.
        • et al.
        Assessing change in avian influenza A (H7N9) virus infections during the fourth epidemic—China, September 2015–August 2016.
        Morb Mortal Wkly Rep. 2016; 65: 1390-1394
        • Zhang P.
        • Wang J.
        • Atkinson P.M.
        Identifying the spatio-temporal risk variability of avian influenza A H7N9 in China.
        Ecol Modell. 2019; 414108807
        • Zhang X.
        • Zou L.
        • Chen J.
        • Fang Y.
        • Huang J.
        • Zhang J.
        • et al.
        Avian influenza A H7N9 virus has been established in China.
        J Biol Syst. 2017; 25: 605-623