Introduction
During the past decade, outbreaks of infectious disease have been a major cause for concern (Kahn 2009), particularly in the developing countries where control is restricted by inadequate infrastructural and financial resources. Infectious diseases occurring worldwide, such as HIV/AIDS, tuberculosis, malaria and diarrhoea diseases have been shown to present significant public health concerns in terms of morbidity and mortality (Tanser and Le Sueur 2002). Hence, the distribution of their pathogenic agents is very important for disease surveillance, control and prevention. By creating a link between transmission of infectious diseases, and spatial and spatio-temporal proximity, researchers are now able to use geographical information system (GIS) in infectious disease epidemiology and to inform decision makers.
Spatial analysis and GIS
When carrying out epidemiological investigation, the type of data to be collected usually depends on the purpose of the investigation. Where an investigation is been carried out with the intention of determining the pattern of distribution of an infectious disease, GIS has proven to be a very useful system (Pfeiffer et al. 2008)). Like many other computer-based system, it can be used for data collection, storage, management and analysis. It is also used for modelling and decision making. If spatial data such as coordinate locations, address or postcode are collected; GIS allows epidemiologist to perform the additional task of studying the relationship between different types of epidemiological and/or spatial data that will otherwise not be analysed together and to produce maps for investigating the spatial pattern of disease distribution. According to Pfeiffer et al. (2008), the objectives of spatial epidemiological analysis are the description of spatial pattern, identification of disease clusters and explanation or prediction of disease risk. To achieve these objectives (in relation to an infectious disease) is to demonstrate the application of GIS and spatial analysis in the surveillance, control and prevention of disease.
Identifying Patterns in Disease Distribution
In discussing the application of GIS and spatial analysis in infectious diseases epidemiology, the diagrammatic representation of a spatial analysis framework (See Figure 1) adapted from Bailey and Gatrell (1995) is a useful stating point. One of the most commonly used spatial analysis method in epidemiological investigation of infectious diseases is visualisation (Pfeiffer et al. 2008). In order to identify spatial trend in an epidemiological dataset when carrying out an epidemiological investigation an infectious disease, spatial data contained in the dataset should be plotted. This usually leads to the creation of maps that can be used to describe any existing pattern in the disease distribution (descriptive analysis), allowing for significant appreciation of disease pattern, identification of risk factor and also to effectively communicate the findings of epidemiological analysis for policy discussion. Because of these contributions that descriptive spatial analysis can make to infectious disease control / prevention programmes, routinely collected surveillance/epidemiological data mostly include spatial data. The study of cholera transmission in London in the mid-1850s by John Snow is a classical example of the application of visual representation of spatial data in infectious disease epidemiology (Nelson et al. 2007). He plotted a map of cholera death (point data) in relation to the Broad Street pump (Figure 1) and used it to summarise his data to his audience and also as a tool to try to convince others of his conclusion. Upon visualization of disease map, any unusual aggregation of cases of disease spatially and/or temporally (not related to the distribution of the population at risk), will require further investigation (Pfeiffer et al., 2008).
Figure 1: John Snow's 1954 cholera-outbreak map of London showing the clusters of cholera cases. (Source: Wikipedia)
The Snow's example represents a situation where point data was used to give a simple picture of a disease event; however when the cases of disease in a single area is large, it becomes difficult to plot a clear map using point data. In such situation, data from each defined area, county or country are aggregated into a single representative value (Pfeiffer et al., 2008). Maps produced from aggregated data are commonly used to visualise spatial distribution of disease and disease prevalence, but they have certain limitations that cannot be discussed here.
Thumbi et al. (2010) related that in the past only few studies focused on the spatial dimension of trypanosomiasis prevalence because of the high cost associated with creating tse-tse distribution maps through ground-based vector surveys. However, these processes have been made easier and cheaper by GIS software, which can be used to produce more accurate maps. In addition, the use of GIS also allows for effective analysis of the different factors that could influence the pattern of distribution of trypanosomiasis (Thumbi et al. 2010). Plotting spatial data on a map may reveal a pattern in the disease distribution requiring further investigation to explain the observation.
Explaining Disease Distribution
Upon visualization, patterns in disease distribution that are believed or confirmed to be different from what it should have been are subjected to statistical analysis. Unusually patterns in disease distribution are generally classified as regular, random or clustered, with disease clustering being the most commonly observed pattern (Pfeiffer et al. 2008). According to the popular definition of disease clustering proposed by Wakefield et al. (2000), "a disease is clustered if there is residual spatial variation in risk after known influences have been accounted for". Investigating disease clustering is an important epidemiological process because it helps to determine if the clustering is statistically significant and worthy of further investigation, or if the clustering may have occurred by chance or as a reflection of population distribution (Pfeiffer et al. 2008).
In a study carried out to assess the effect of putative risk factors on the spatial distribution of BSE and to improve the effectiveness of BSE control programs in Great Britain, Stevenson et al. (2005) investigated the area-level risks for Bovine spongiform encephalopathy (BSE) in British cattle before and after the July 1988 meat and bone meal feed ban. Aggregated data were used to produce choropleth maps for visualization and descriptive analysis of BSE distribution, pre- and post-ban. The area-level risk of BSE for two birth cohorts: those animals born before the introduction of statutory control measures on 18 July 1988 (termed the pre-control cohort), and those animals born between 18 July 1988 and 30 June 1997 (the post-control cohort) were evaluated and then standardised mortality ratios and the standard errors of the standardised mortality ratios for BSE for the pre- and post-control cohorts were plotted for each cohort as choropleth maps. Figure 2a showed a uniformly distributed but unexplained BSE risk across Great Britain before the ban, whereas after the ban (Figure 2c), there was spatially aggregated areas of unexplained risk in the northern and eastern regions of England.
Exploring the spatial dataset, Spearman's rank correlation coefficient was used to quantify the relationships between BSE standardised mortality ratios and the explanatory variables, whereas, the presence of unexplained spatial aggregation in the data was assessed by quantifying the similarity of the residual values for areas defined as adjacent using Moran's I statistic with the aim of looking for autocorrelation. The outcome of the statistical analysis indicated spatial autocorrelation.
Figure 2: Choropleth maps of area-level standardised mortality ratios (SMRs) and standard errors of area-level SMRs for BSE in the population of cattle present in Great Britain from 30 June 1986 to 30 June 1997: (a) and (b) the pre-control cohort, (c) and (d) the post-control cohort. Pre-control cohort: cattle born up to and including 18 July 1988; post-control cohort: cattle born between 18 July 1988 and 30 June 1997.
To quantify the influence of hypothesised risk factors on area-level relative risk of BSE, Bayesian Poisson models were constructed and this led to the belief that for cattle born after the ban on feeding ruminant-derived protein to ruminants, the easterly shift in BSE risk (Figure 2c) implies that control measures were less-successfully applied in the Eastern counties of England compared with other regions of Great Britain. The findings from the investigation support the belief that low level cross-contamination of cattle feed by pig feed play a role in influencing the incidence of BSE and this could help to target and/or strengthen control efforts for BSE.
There are a number of other statistical methods also available for investigating possible disease clustering to determine whether the clustering is statistically significant and in need of further investigation or whether the observed pattern is a matter of chance (Pfeiffer et al. 2008).
Predicting Disease Distribution
A number of factor including environmental, climatic or biological factors have been found to have the ability to influence the distribution pattern of some infectious diseases. As a general note, the type of factor influencing a particular infectious disease usually depends on the disease in question. One very important advantage of knowing the factors influencing a disease distribution is the ability to predict the pattern of distribution of the disease by looking at the relationship between the occurrence of the disease and the known influencing factor(s)/explanatory variable or quantify the effect of a set of explanatory variables on the spatial distribution of the disease. GIS is a very useful system in this respect because of its ability to store, manage and analysis both spatial and non-spatial data in the same geographical framework. In combination with remotely sensed (RS) environmental data, the explanatory power of GIS is further enhanced (Hay et al. 2000).
In order to predict the possible pattern of spread of an infectious disease, an understanding of the epidemiology of such a disease is essential. For example it was narrated by Rogers et al. (2003) that West Nile Fever (an infectious disease with only one principal vector species) first appeared in New York, USA in 1999 and within 3years spread to 44 US states and Canada, causing clinical infection in 4,000 people with 284 deaths. But because West Nile Fever is not typical to the American continent it would have been expected that it may just have a few competent vectors, but in August 2003, 43 mosquito species in the United States were found to have West Nile virus, its RNA or antigen. They commented that, the spread of the disease would have been less surprising, if only the large number of potential vectors were known in advance. Knowing a disease vector is one step in predicting disease spread, the next step being to gather all available information regarding the distribution, incidence and prevalence of the disease which can then be stored with GIS (Rogers et al. 2003). It is very important for data (to be used in predicting disease spread) stored in GIS to contain both the disease data and all available data on the explanatory variable (environmental, climatic or biological).
An important component of GIS is Satellite imagery because of the ability of satellite sensors to produce data containing information about the condition of rainfall, temperature, humidity and vegetation on the Earth's surface which are important for the transmission of infectious disease pathogens by a disease vector or intermediate host (Rogers et al. 2003). When all the required data for an infectious disease are available, statistical methods (e.g. regression modelling) is used to quantify the effect of a set of explanatory variables on the spatial distribution of the disease and then the statistical relationships established between disease and environmental datasets are applied at the full spatial resolution of the disease dataset to produce a risk map (Pfeiffer et al. 2008).
Conclusion
In many ways, GIS and spatial analysis have proven to be very effective in the surveillance, control and/or prevention of infectious disease, some of which are presented in this write-up. With the increasing threat pose by infectious diseases, especially zoonotic infectious diseases in recent times, it is hoped that disease surveillance, control and prevention will become even more feasible with the use of GIS and spatial analytical techniques.
