- Research article
- Open Access
- Open Peer Review
Spatial analysis of hemorrhagic fever with renal syndrome in China
© Fang et al; licensee BioMed Central Ltd. 2006
- Received: 11 October 2005
- Accepted: 26 April 2006
- Published: 26 April 2006
Hemorrhagic fever with renal syndrome (HFRS) is endemic in many provinces with high incidence in mainland China, although integrated intervention measures including rodent control, environment management and vaccination have been implemented for over ten years. In this study, we conducted a geographic information system (GIS)-based spatial analysis on distribution of HFRS cases for the whole country with an objective to inform priority areas for public health planning and resource allocation.
Annualized average incidence at a county level was calculated using HFRS cases reported during 1994–1998 in mainland China. GIS-based spatial analyses were conducted to detect spatial autocorrelation and clusters of HFRS incidence at the county level throughout the country.
Spatial distribution of HFRS cases in mainland China from 1994 to 1998 was mapped at county level in the aspects of crude incidence, excess hazard and spatial smoothed incidence. The spatial distribution of HFRS cases was nonrandom and clustered with a Moran's I = 0.5044 (p = 0.001). Spatial cluster analyses suggested that 26 and 39 areas were at increased risks of HFRS (p < 0.01) with maximum spatial cluster sizes of ≤ 20% and ≤ 10% of the total population, respectively.
The application of GIS, together with spatial statistical techniques, provide a means to quantify explicit HFRS risks and to further identify environmental factors responsible for the increasing disease risks. We demonstrate a new perspective of integrating such spatial analysis tools into the epidemiologic study and risk assessment of HFRS.
- County Level
- Hemorrhagic Fever With Renal Syndrome
- Spatial Autocorrelation Analysis
- Exploratory Spatial Data Analysis
- Annualize Average Incidence
Hemorrhagic fever with renal syndrome (HFRS) is a zoonosis caused by different species of hantavirus (HV). China is the most severe endemic country, 90% of the total HFRS cases in the world were reported . Although integrated intervention measures involving rodent control, environment management, and vaccination are being implemented, HFRS remains a public health problem with 20,000–50,000 human cases annually in mainland China. The incidence of HFRS shows high variabilities at both provincial and county levels. Economic development, urbanization, human mobility, and environment and climate changes were thought to be related to incidence and spatial distribution of HFRS . The HFRS incidence has been increasing in some metropolises and provincial capital cities in recent years . A better understanding of the spatial distribution patterns of HFRS would help to identify areas and population at high risk.
The spatial analyses, such as spatial smoothing and cluster analysis are commonly used to characterize spatial patterns of diseases [2–9, 20]. Spatial smoothing is used to reduce random variation associated with small populations and enables observations of gradients or holes of disease incidence that may not apparent from direct observation of raw data [2, 10, 11]. Spatial autocorrelation analysis was performed to detect significantly difference from a random spatial distribution of HFRS cases [15, 18]. Spatial cluster analysis is applied to identify whether cases of disease are geographically clustered [12–14]. In this study, we conducted GIS-based spatial analyses involving spatial smoothing, exploratory spatial data analysis (ESDA) and spatial scan statistic to characterize geographic distribution pattern of HFRS cases. Spatial scan statistic was used to identify areas and population at high risk at the county level, which corrects for multiple comparisons, adjusts for the heterogeneous population densities among the different areas, detects the foci without prior specification of suspected location or size thereby overcoming pre-selection bias, and allows for adjustment of confounders [12, 16, 19].
Data collection and management
Records on HFRS cases between 1994 and 1998 were obtained from the National Notifiable Disease Surveillance System. For conducting a GIS-based analysis on the spatial distribution of HFRS, the county-level polygon map at 1:1,000,000 scale was obtained, on which the county-level point layer containing information regarding latitudes and longitudes of central points of each county was created. Demographic information based on 1995 census was integrated in terms of the administrative code . All HFRS cases were geocoded and matched to the county-level layers of polygon and point by administrative code using the software ArcGIS8.3.
GIS mapping and smoothing
To alleviate variations of incidence in small populations and areas, annualized average incidences of HFRS per 100,000 at each administrative region over the 5 year-period were calculated, and spatial rate smoothing was implemented.
Based on annualized average incidence, all counties were grouped into four categories: non-endemic area, low endemic area with annualized average incidence between 0 and 5 per 100,000, medium endemic area with the incidence between 5 and 30 per 100,000, and high endemic area with the incidence over 30 per 100,000. The four types of counties were color-coded on maps.
To assess the risk of HFRS in each county, an excess hazard map was produced. The excess hazard represents the ratio of the observed incidence at each county over the average incidence of all endemic areas, the later was calculated by the number of cases over the total number of people at risk instead of the annualized incidence of a county .
The technique of spatial rate smoothing was employed to annualized average incidence of HFRS. The smoothed incidence was computed from the total number of cases in a spatial "window" divided by the total number of people at risk within the "window", which was specified using a spatial weights file including both county and its neighbor counties' locations. Each smoothed incidence was calculated once the "window" core overlapped with a county center. So the first step in the analysis was to construct a spatial weights file that contained information on "neighborhood" structure of each county. The k-nearest neighbor criterion ensured each observed object had exactly the same number (k) of neighbors. In the analysis six neighbors were chosen for each county by k-nearest neighbor criterion. The second step was to load the weight file and carry out smoothing analysis .
To establish a continuous distribution map of HFRS, a spatial interpolation was conducted using the established county-level point layer. Inverse distance weighting (IDW) method was used due to lack of normality of distribution of annualized average incidence and difficulty of transformation (into to normal distribution).
Spatial autocorrelation analysis
Global spatial autocorrelation analysis was performed in GeoDa0.9.5-i software. Moran's I spatial autocorrelation statistic was calculated and visualized in the form of Moran Scatter Plot. First, a contiguity-based spatial weight was constructed for each county by creating a rook contiguity weights file. Spatial autocorrelation statistics for HFRS incidence were calculated based on the assumption of constant variance. This assumption was usually violated when incidence at county level varied greatly in different populations. The Assuncao-Reis empirical bayes standardization (i.e. a function in GeoDa) was performed to adjust for the violation of the assumption. Secondly, Moran's scatter plot was produced with a spatial lag of incidence on the vertical axis and a standardized incidence on the horizontal axis. Any observation beyond two standard deviations was categorized as outlier. Thirdly, a significant test was performed through the permutation test, and a reference distribution was generated under an assumption that the incidence was randomly distributed. The number of permutation test was set to 999 and the significance level was set as 0.001.
Spatial cluster analysis
Spatial cluster analysis was performed to detect spatial clusters of HFRS cases. "Spatial scan statistics" was used to test the null hypothesis that the relative risk (RR) of HFRS was the same between any county groups, or collection of county groups, and the remaining county groups. Areas with differing sizes were scanned without knowledge on cluster size and location to avoid selection bias. SaTScan software, designed specifically to implement this test, imposed a circular window on the map . This window moved over the study region and centered on the centroid of each county. The area within the circular window varied in size from zero to some upper limit (a maximum radius of the circular window set in virtue of the proportion of the whole population) specified by the user, never including > 50% of the total population. Possible clusters were tested within the variable window around the centroid of each county group. Whenever the window finds a new case, the software calculates a likelihood function to test for elevated risk within the window in comparison with those outside the window. The likelihood function for any given window was proportional to: (d/n)d([D - d]/[D - n])(D - d) I(), where D is the total number of cases, d is the number of cases within the window, and n is the expected number of cases. If SatScan was scanning for higher incidences, the indicator function I() was 1 when cases in the window are more than expected, otherwise it would be 0. In this study, retrospective spatial cluster analysis for higher incidences was used, in which the maximum window radius was set to be smaller than 20% of the total population. Smaller maximum radius (≤ 10% of the total population) was also tried to look for possible subclusters. For each window of varying position and size, the software tested the risk of HFRS within and outside the window, with the null hypothesis of equal risk.
Spatial distribution of HFRS in China
Spatial autocorrelation of HFRS in China
Spatial autocorrelation analyses for annualized incidence of HFRS in mainland China from 1994 to 1998
Moran 's I
The distribution of HFRS clusters
In the study, exploratory spatial data analysis and spatial cluster analysis of HFRS were conducted at county level of mainland China. We mapped HFRS from different aspects such as crude incidence, excess risk, spatial smoothed incidence, and incidence with IDW, evaluated the spatial pattern and highlighted geographic areas with significant high incidence of HFRS in mainland China. Furthermore, this study demonstrated that additional tools necessary for disease surveillance could be provided for public health officials using existing health data, GIS and spatial scan statistics.
The study showed that the spatial distribution of HFRS in mainland China was nonrandom and clustered with a Moran's I of 0.5044 (p = 0.001) from 1994 through 1998. Spatial cluster analysis identified 16.51% total population and 26 areas increased HFRS risk when a maximum spatial cluster size of ≤ 20% total population was used. Additional cluster analysis based on a maximum spatial cluster size of ≤ 10% total population identified 39 subclusters occupied by 18.42% of the total population, which had statistically significant (p < 0.01) increased HFRS risk. The results suggest that there were "hot-spots" of HFRS in a number of areas in China, which were also the priority areas of public health planning and resource allocation for preventing HFRS. For instance, there were large areas (> 10,000 km2) of increased HFRS risk existed in Shandong, Hebei, Heilongjiang, Hunan, Zhejiang, Jiangxi, and Guangxi provinces, and some small areas (≤ 10,000 km2) with increased HFRS risk in some other provinces of central, eastern and north-eastern China.
The spatial distribution of HFRS was correlated with density, species and infection rate of rodents as the major animal reservoirs, which were influenced possibly by natural and social-economic environmental conditions such as the elevation, land use, soil type, vegetation, precipitation, atmospheric temperature, et al [21, 22]. To identify and measure quantitatively the most important determinants of HFRS distribution, and to assess the burden of illness due to HFRS, more detailed epidemiological investigations need to be carried out. Clusters with significantly high incidence of HFRS identified will be helpful of investigating the underlying causes of increased risk in the identified areas, landscape attributes and identification of the environmental variables characteristic of high-risk areas with different acreage. Environmental and landscape characteristics, socio-economic factors associated with increased risk for HFRS infections need to be studied.
This study has shown the presence of 'hot-spots' of HFRS in mainland China. The study has also demonstrated that using existing health data, GIS and GIS-based spatial statistical techniques could provide an opportunity to clarify and quantify the health burden from HFRS within highly endemics areas, and also lay a foundation to pursue further investigation into the environmental factors responsible for increased disease risk. To implement specific and geographically appropriate risk-reduction programs, the use of such spatial analysis tools should become an integral component in the epidemiologic description and risk assessment of HFRS.
The authors extend the appreciation to Huaxin Chen, Yalan Liu, Hua Yang for providing the data. The study was supported by Natural Science Foundation of China (grant number: 30590370, 30590374), Commission of the European Communities (grant number: SP22-CT-2004-003824) and Beijing Natural Science Foundation (grant number: 7061005).
- Bai X, Huang C: Study farther on hemorrhagic fever with renal syndrome. Chin J Infect Dis. 2002, 20: 197-198.Google Scholar
- Curtis A: Using a spatial filter and a geographic information system to improve rabies surveillance data. Emerg Infect Dis. 1999, 5: 603-606.View ArticlePubMedPubMed CentralGoogle Scholar
- Nkhoma ET, Chiehwen EH, Victoria IH, Harris AM: Detecting spatiotemporal clusters of accidental poisoning mortality among Texas counties, U.S., 1980–2001. Int J Health Geog. 2004, 3: 25-37. 10.1186/1476-072X-3-25.View ArticleGoogle Scholar
- Frank C, Fix A, Pena C: Strickland G. Mapping Lyme disease for diagnostic and preventive decisions, Maryland. Emerg Infect Dis. 2002, 8: 427-429.View ArticlePubMedPubMed CentralGoogle Scholar
- Odoi A, Martin SW, Michel P, Middleton D, Holt J, Wilson J: Investigation of clusters of giardiasis using GIS and a spatial scan statistic. Int J of Health Geog. 2004, 3: 11-21. 10.1186/1476-072X-3-11.View ArticleGoogle Scholar
- Glass GE, Schwartz BS, Morgan JM, Johnson DT, Noy PM, Israel E: Environmental risk factors for Lyme disease identified with geographic information systems. Am J Public Health. 1995, 85: 944-948.View ArticlePubMedPubMed CentralGoogle Scholar
- Morrison AC, Getis A, Santiago M, Rigau-Perez JG, Reiter P: Exploratory space-time analysis of reported dengue cases during an outbreak in Florida, Puerto Rico, 1991–1992. Am J Trop Med Hyg. 1998, 58: 287-298.PubMedGoogle Scholar
- Mott KE, Nuttall I, Desjeux P, Cattand P: New geographical approaches to control of some parasitic zoonoses. Bull World Health Organ. 1995, 73: 247-257.PubMedPubMed CentralGoogle Scholar
- Zeman P: Objective assessment of risk maps of tick-borne encephalitis and Lyme borreliosis based on spatial patterns of located cases. Int J Epidemiol. 1997, 26: 1121-1129. 10.1093/ije/26.5.1121.View ArticlePubMedGoogle Scholar
- Rushton R, Lolonis P: Exploratory spatial analysis of birth defect rates in an urban population. Stat Med. 1996, 15: 717-726. 10.1002/(SICI)1097-0258(19960415)15:7/9<717::AID-SIM243>3.0.CO;2-0.View ArticlePubMedGoogle Scholar
- Talbot TO, Kulldorff M, Forland SP, Haley VB: Evaluation of spatial filters to create smoothed maps of health data. Stat Med. 2000, 19: 2399-2408. 10.1002/1097-0258(20000915/30)19:17/18<2399::AID-SIM577>3.0.CO;2-R.View ArticlePubMedGoogle Scholar
- Kulldorff M: A spatial scan statistic. Communications in Statistics: Theory and Methods. 1997, 26: 1481-1496.View ArticleGoogle Scholar
- Kulldorff M, Nagarwalla N: Spatial disease clusters: detection and inference. Stat Med. 1995, 14: 799-810.View ArticlePubMedGoogle Scholar
- Kulldorff M, Feuer EJ, Miller BA, Freedman LS: Breast cancer clusters in the northeast United States: a geographic analysis. Am J Epidemiol. 1997, 146: 161-170.View ArticlePubMedGoogle Scholar
- Anselin L: Local indicators of spatial association (LISA [J]). Geog Analy. 1995, 27: 93-115.View ArticleGoogle Scholar
- Kulldorff M, Information Management Services, Inc: SaTScan™ v6.0: Software for the spatial and space-time scan statistics. 2005, [http://www.satscan.org/]Google Scholar
- the institute of geographical sciences and natural resources research, the Chinese academy of sciences: China natural resources database. 2005, [http://www.data.ac.cn]Google Scholar
- Anselin L, Syabri I, Kho Y: GeoDa: An Introduction to Spatial Data Analysis. 2005, [http://www.csiss.org]Google Scholar
- Chaput EK, Meek JI, Heimer R: Spatial analysis of human granulocytic ehrlichiosis near Lyme, Connecticut. Emerg Inf Dis. 2002, 8: 943-948.View ArticleGoogle Scholar
- Bailey TC: Spatial statistical methods in health. Cad saude public. 2001, 17: 1083-1098.Google Scholar
- Cao W, Fang S, Li C, Sun J, Fang L, Zhang X: Environmental risks of hemorrhagic fever with renal syndrome in China: the use of geographic information systems in a landscape epidemiological approach. The 3rd Asian-Pacific Congress of Epidemiology. 2001, Kitakyushu, JapanGoogle Scholar
- Chen H, Qiu F: Studies on the environment structure of natural nidi and epidemic areas of hemorrhagic fever with renal syndrome in China. Chin Med J. 1994, 107: 107-112.PubMedGoogle Scholar
- Zhang Y, Xiao D, Wang Y, Wang H, Sun L, Tao X, Qu Y: The epidemic characteristics and preventive measures of hemorrhagic fever with renal syndrome. Chin J Epidemiol. 2004, 25: 466-469.Google Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2334/6/77/prepub
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.