Spatial Analysis of Cholera Prevalence in Kumasi Area in Ghana

Introduction

Current Global Health reports show very discouraging figures of people getting infected with communicable diseases in most populous places (Luquero, 2011). Governments and the international community are finding it very hard to strike a balance between economic growth and fight of communicable diseases such as Cholera and others. Cholera is one of the diseases which are p[roving very hard for governments to ensure a healthy population. It has claimed many lives in major parts of the world and specifically in Africa. It is highly infectious diseases which need highly cautious preventive measures to address it accordingly.

The disease is caused by a bacterium (v.Cholerae) which is usually passed through faecal matter in the environment. The diseases usually affect the intestines causing severe vomiting and diarrhoea. It commonly affects the society which lacks the clean water mostly in the developing countries. It is able to survive in any environment, a fact which makes it hard to eradicate it. Since some places are highly likely to contract these diseases, it is very proper to ensure that we use the GIS information and analysis of the Cholera spatial information to determine the areas which are most affected and the predisposing factors for this epidemic. Majorly, the research implements the spatial epidemiology to find the relationship between the diseases and the common risk factors. The spatial information and the GIS data will be required to provide the health officials with the right view of the causative factors for the disease.

The research will use the environmental variables such as the refuse dumps and the water reservoirs which are obtained from the satellite images.T he concentration of these variables and the cholera incidence on the ground will be mapped in GIS and the data analyzed using the spatial statistical methods. The main objective is to find the distribution of environmental factors which increases the risks of Cholera spread in Kumasi, in Ghana by using the GIS information and spatial statistical methods of analysis.

Materials and Methods

The method used to determine the spread of Cholera and the risk factors is referred to as Spatial Epidemiology. This method uses the geographical, environmental and behavioural factors to determine the relationships between diseases and their risk factors. The spatial information is very necessary for battling the epidemiology since it is believed that those who are closer to the spatial risk factors are at higher chances of contracting the disease.

The Study area is the Kumasi, Ghana where the mapping of the spatial factors and the spread of diseases is made. The most places have a large number of refuse dumps and water reservoirs which form the most important area for studying their relationship with the spread of Cholera. The distance is mapped between these variables and the community to depict the spread of the disease. Kumasi is one of the largest metropolitan areas in Ghana. It is part of the middle belt of Ghana with the latitude of 6.040N and longitude 1.280W on an area of approximately 220km.The report presented by Duker and Osei (2008).It has approximately 1.2 million people which are a larger percentage of Ashanti’s population. In the study, 68 sections were used to derive the data. The area receives two major seasons which is basically described as the rainy and dry season. April and July receive a large amount of rainfall while September to November is a short rainfall season.82 % of people in the area have access to piped water but there is still large population using the running water for cooking and drinking. According to Osei et al. (2010),

Fig 1: Regional Map of Ghana (Left) and the Kumasi region (Right)

The data was obtained from Osei (2010). This data was obtained from Kumasi Metropolitan Disease Control Unit (DCU) in 2005 where the worrying outbreak of Cholera was reported in the region. It consists of the spatial representation of the number of people infected in the community within 72 days that the outbreak lasted. The community is represented in a shapefile in X and Y coordinates planes. The attributes include the number of people affected by cholera, the population and raw rates. Raw rates are obtained by calculating the population estimates presented in Ghana Statistical Service (GSS). The Cholera rates of are obtained by taking the cases reported divided by the population and multiplied by the 10,000 factor to make data more intuitive.

Refuse Dump data

Kumasi has the large number of old refuse dumps which majorly contain organic materials, papers and some few metals. It also has inert materials which are majorly formed through the wood ash and charcoal. Waste management is done through the use of contracted organizations. These organizations are picked by Waste Management Department in Kumasi (WMD). The solid management is done through communal collection responsibility and house to house collection. Collection organizations use containers which are placed in various households then dumped in the landfills on the periphery of the metropolis. Despite the fact that wastes containers are placed at various houses, 50% of the households dump their wastes away from the dumpsites and a number of them dump them in the nearby rivers.

Refuse dump data is derived from the Osei and Duke (2010) research done on 2005 using the GPS. 124 refuse dumps are obtained and mapped as per the  GPS. The data is presented in shapefiles indicating X and Y coordinates in meters.

The rivers data is obtained from the same Osei (2010) data which were presented in digital maps in 2005. Rapid eye data sensors were also used to obtain the concentration and locations of risk reservoirs which can be identified with rapid breakouts of Cholera in Kumasi. Image capture and Rapid Eye sensors were commonly used to make decisions basing on the earth observation imagery. Rapid Eye has 5 satellites constellations which produce the 5-meter resolution. It travels on the orbital lanes and can provide high-quality images when focused in a specific area.

The study area was focused in the Rapid Eye scanning using the RapidEye ortho-level 3A specification. The images obtained were then mosaicked using ERDAS Imagine 0f 2011 and the area divided into subsets from the larger image. The small subset is good for us into a very specific area and finds the reservoirs and the images for the residential settlements.

The information about the refuse dumps, settlements in the community, rivers and rivers were obtained from the Osei (2010) research. Osei indicates that the shapefile which were entered into the ArcMap was obtained in Ghana Transverse Mercator System (GTM). They were later reassigned into the Ghana Grid System and later translated to the Universal Transverse Mercator UTM_WGS 1984 and overlaid with the RapidEYe by the help of image processing techniques. The software which made this research successful include the ERDAS, ArcGIS, R-software and open geodata.

Fig 2: RapidEye image

Analysis  methods and the Flow Chart

The image of the is was organized to depict a land cover map The cholera data were integrated with the refuse dump data, water bodies and rivers data and viewed through the GIS perspective (Bailey, & Gatrell, 2015). The modelling techniques for the data were chosen as the spatial analysis and regression modelling which were then carried out on the data.

Figure 3: Process followed in Research

Image Analysis

Pre-processing analysis

The images obtained through RapiEye sensing were placed in the ERDAS imagine and orthorectified to produce a large point of view. All parts and sections of the image were put under put under the right coordinates on the real ground. This was done with the help of a projection system kept by the Ghana WGS.

Image Classification

The classification of the image was done when the pixels of the image are matched with the spectral appearance of the real ground (Fotheringham& Rogerson, 2014). Pixels are sorted and placed in different classes or data categories determined by the data files values. Classes have specific set of criteria which should be satisfied, therefore, pixels in a particular class must satisfy such criteria. Image data is then converted to thematic data using the maximum likelihood algorithm. The classification process should be monitored closely and the data was validated to obtain assess the accuracy. The accuracy was validated by use of 84 points picked and compared to the actual data on the map. It was then analysed by putting the data on the error matrix to correct the mistakes done in classification

Integration and Visualization

Since the communities in Kumasi do not have classified boundaries between them, there was a need to draw some arbitrary boundaries which could be used to analyze data. It is done visually using the natural boundaries such as streams and the clusters which are found at some various locations. The various reservoirs were turned to shapefiles after extraction from the RapidEye Images. The streams and refuse dumps are analyzed differently on different scales.

Spatial Factor Maps

Since the hypothesis is that Cholera is caused by poor sanitation, it is believed that water bodies such as the resources, streams/rivers and the refuse dumps are some of the predisposing factors. The following predictions are then made on such hypothesis:

• People living near refuse dumps are likely to contact the disease.
• Areas with high density refuse dumps are likely to get the disease.
• Those living close to reservoirs are had high chances of getting the disease.

The spatial relationship between the disease and (i) distance to refuse dumps (ii) distance to cholera reservoirs (iii) refuse dump density. Four spatial maps were drawn by use of ArcMap. The spatial maps are overlaid with the centroids which represent the disease prevalence of the case of cholera in this areas. The four variables include:

• Proximity to refuse dumps-distance between each pixel to the refuse dump
• Refuse dump density- the number of refuse dumps per area
• Proximity to digitized reservoirs-Distance between each point or the pixel to the water bodies in the nearest locations. It includes rivers and the streams shown in the spatial maps.
• Proximity to classified reservoirs- The proximity of the points of incidence to the images of the reservoir obtained through the RapidEye image.

Proximity to classified reservoirs uses information obtained from rapid eye image while the proximity to the reservoir variables is information obtained from the digitized pixels from the topographic map from Osei et al. (2010). Reservoirs and the rivers are from the same geographical location but measured using different scales. Polygon boundaries of the community are used to present the cholera information which is then entered into the database. The community attributes include the centroids, population, raw rates and the explanatory variables presented above.

Proximity to Refuse Dumps

The distance surface was obtained using the spatial analysts and Distance toolbox on the ArcMap.the refuse layer is used as the necessary input point to the source. It is the calculation to the closest source.1 km radius was used to enable accurate measurement from one centroid to the refuse dump.

Fif 4: Distance near a refuse dump

The density of Refuse Dumps

The kernel surface shows the relationship between the centroids and the density of the refuse dumps. The magnitude is calculated in each area using the kernel density paradigm. The values vary from the point location and reduce as you move away from the point. It reaches zero at the 1km mark of the demarcated area.

The density values between points tool and Extract values were extracted using the spatial analyst extension. It is then recorded in the community feature class.

Fig 5: Proximity to digitized reservoirs

Inputs are the digitized streams from Osei et al. (2010). The output is presented in the image below depicting the pixel with the potential cholera reservoir.

Fig 6: Proximity to Digitized Reservoirs

Proximity to classified reservoirs

This variable describes the relationship between each an every pixel to the nearest reservoir images obtained from the RapidEye images. Which offered the nearest distance to the communities around.

Mapping and Geovisualisation

The relationship between cholera prevalence and risk factors is obtained generated before measurement and modelling of spatial relationships. The task was to be followed by the attempt to establish spatial units as characterized by lower or higher counts of Cholera cases. The maps generated include Choropleth maps, proportional symbol maps, Choynowskys probability map and four probability maps.

Spatial Analysis

The spatial analysis is done through two principal methods which include autocorrelation analysis and spatial covariates. The autocorrelation analysis refers to the establishment of spatial autocorrelation to the immediate neighbouring settlements. The steps for autocorrelation analysis involves choosing the neighbourhood, making the spatial weight matrix and implementing the statistical tests by use of a matrix.

Results and Analysis

The reservoirs, built-up areas and urban centres forest dense vegetation and grasslands were mapped into landcover. Emergent vegetation was evident from the mapped map with the water bodies. The landcover locations were merged into five classes. The image is presented below as shown by Osei et al (2010).

The rivers, when laid upon the map, shows a very interesting relationship between the effects of cholera and its environment. The 85 pixels was used as a sample and reference to the real world class and it was maintained at 83.53 % reference percentage. The image could not show the refuse dumps but it is safe to use it. The refuse dump layer on the map is attributed to the Osei dam.

Accuracy levels are indicated in the table below:

The cholera maps are indicated in the image below. The rates per 10,000 people are shown in the map drawn to depict the rates in every 10, 00 people infected by the disease. The community is coloured depending on the prevalence of the dishes as shown on the map.red colour indicates high rates while blue colour has lowest incidences. The legend acts to dep[ict the level of the values with respect to the map.

Fig 8: Incidence cases versus the incidence rate

The thematic maps presented above shows the relationship between the cholera infections and the communities affected. High prevalence of the disease is witnessed at the upper part of Kumasi.

The size of each symbol is proportional to the magnitude of the prevalence of the disease. The map uses counts per rate of 10, 000 people. To characterize the distribution of Cholera incidences, Refuse dumps have been indicated by use of black dots and river segments used the blue colours.

Fig 9: Incidence  counts versus incidence rate

The probability map based on Choynowski brings two ends of measured chances together so that the indications of spatial units, the communities have the right level of the high and low communities are plotted separately as indicated in the figure below Figure 8 shows the raw rates, relative risks and Poisson probability map values.raw rates and other variable handled by this research uses position cumulative distribution fucntion.communities with lower values have the values with the lower tail while the ones have the values at upper tail. The summary presents the high statistics of data obtained through the R function. The function shows the application of crude rates, relative risks, causes and Poisson probabilities.

Fig 9: Chomsky map presentation

Fig 10: Probability map versus Risk Map

The map depicting poison chances show the extreme count while the relative risk map depicts ration of an expected map and the other relevant counts multiplied by 100.

Fig 11: Rates Map versus Count Map

The raw rate map presented the incident counts per the community while the second pat which is expected count presents the incidents count as per the entire global count.

Spatial Analysis

The extent of the Cholera prevalence is witnessed within the environment can be shown through the use of Moran’s Index. Randomization procedure is also run through R function yielding significant  values in the map as  shown below:

Table 1: Morans’ Spatial Correlogram

Moran’s 1=0.138, P=0.045The Moran’s 1 and two-sided chance value is presented by the value of 6. The table below presents the correlogram for CASES.

The scatter plots are also used to depict the relationship between the incidences and risk factors. The steepness of the scatter plot represents Morans L which shows the autocorrelation and clustering in a dataset. The quadrants of the scatterplot are shown by the use of four characteristics starting with the x-axis, followed by y, high level and the low levels. Clusters correspond to the quadrants in spatial outliers.

Fig 12: Cases verses Lags

Autoregressive Modelling

The statistics presented below shows the relationship between the period of studies and the cases in the community. The reported cases range from 1 to 76. The recorded mean during the study is at 13.96 while the standard deviation is 14.19.

Summary of spatial statistics results for the OLS regression fro digitised rivers are presented in table B while the other values are presented in model  A and the regressions for the RaidEye values is in Model B. Model A is graphically depicted in table 1 and model B is depicted in table 2.

Table 2: Variables used in spatial modelling

The proximity of the community shows the significant or direct relationsho=ip with Cholera incidence. The proximity of the community to the digitized reservoirs also shows high chances of getting Cholera. The use of ND_CR in model  B is used to show the improvement of Model in A. Both models OLS and CAR show that Den_RD is not a direct contributor to Cholera incidences (for table 1p-value =0.83 and in table 2, P-Value +0.42). Exclusion of Den_RD is shown in model B and A.

Model B regression

Table 4: Model B regression

Table 5: Model A regression

Table:6: Model B CAR regression

Discussion

The level of accuracy in the land cover runs to 83.53% with Kappa statistics are at 0.80. Accuracies of the water reservoirs (classified) stand at 100 %. The users’ accuracy on the report is the chance that the values in the reference are indeed labelled as that. When the image of the river obtained in OSEi is overlaid on the classified map, it confirms and depicted the presence of the predisposing factors in the image. Classified reservoirs show the additional risk factors which are not available in the digitized map. Classification of water bodies, therefore, is the validation of the availability of the digitized data on the overall image.

In the autocorrelation analysis, there is the spatial and positive correlation of Cholera outbreaks in Kumasi which validates the spatial mapping of clusters in the central and upper parts of Metropolis and low cases or clusters recorded in the periphery. Morans I is sen as a fluctuating lot between negative and positive value. The high correlation factors are seen in the upper parts which cement the earlier statistics found at the metropolis. The neighbourhood structure presented by the structure explains low p-Value for Kumasi Metropolis.

The reason for the high number of clusters at the central part is due to a large number of settlements at the central part as shown by the Google Earth image. The images of the Google earth shows a large number of houses which can be interpreted as the contributing factors to the high concentration of settlements and hence high chances for contacting Cholera.

The autoregressive models improved by CAR Model A shows the direct relationship between cholera incidences and the refuse dumps (p-Value <0.001) and there is no direct significance with the digitized reservoirs (p-Value=0.073 and p>0.05). It is therefore evident from the values shown above that there is an inverse relationship between incidences and the classified reservoirs. The comparison of results depicted in CAR A to CAR B may also impact the results and conclusions obtained.

The disparities are however obtained by the presence of ponds and stagnant waters which were not documented in the digitized water bodies. Such disparities may have affected the results but it is a good reminder of the significance ofRapidEye images in analysing environmental factors. Satellite images provide a good view of the spatial elements affecting the prevalence of specific diseases. Combination of remote sensing, spatial statistics and GIS  offers the best perspective for assessing the effect of land variables in Cholera incidences.

Limitation of the study

• The research used the communities in form of classes which in esses do not exist. There are no well-defined boundaries which should be used in obtaining research samples. The estimation of boundaries may affect the spatial autocorrelation which varies with the neighbourhood structures. The results should, therefore, be cautiously interpreted since different shapes and sizes of the communities vary.
• The data used are for a single one-year incidence but the accurate analysis could be obtained through several incidence analysis.
• Cholera data are obtained from the community but it does not show the exact locations for affected population assumption of equal risk may not work well with the data analysis.

Conclusion and Recommendation

The study uses sensing image referred to as RapidEye to get the reservoirs on the ground. It employed two common statistical models to depict the cholera incidences with (1) closeness to reservoirs to the topographic map and the refuse dumps (2) proximity to classified reservoirs from refuse dumps and the RapidEye image. The study found out that there is a significant relationship between Cholera incidences and classified reservoirs and refuse dumps than the digitized reservoirs. Maps were used to show the relationship of Cholera prevalence and risk factors in Kumasi metropolis.

It is highly recommended that remote sensing and GIS should be adopted as a way of assessing the prevalence of epidemiology in various areas. Further research in epidemiology prevalence should be done to ensure that proper level of geographic detail is incorporated in the reposts. Remote sensing could not be sued to depict the refuse dumps since the bins are too small to be seen. A higher resolution image should be used to show the refuse dumps. Future research should also incorporate household-specific data such as outbreaks over a certain period of time should be used.

References

Bailey, T. C., & Gatrell, A. C. (2015). Interactive spatial data analysis (Vol. 413). Essex: Longman Scientific & Technical.

Fotheringham, S., & Rogerson, P. (Eds.). (2014). Spatial analysis and GIS. CRC Press.

Luquero, F. J., Banga, C. N., Remartínez, D., Palma, P. P., Baron, E., & Grais, R. F. (2011). The cholera epidemic in Guinea-Bissau (2008): the importance of “place”. PloS one, 6(5), e19005.

Osei, F. B. (2010). Spatial statistics of epidemic data: the case of cholera epidemiology in Ghana. The University of Twente, Faculty of Geo-Information Science and Earth Observation.

Osei, F. B., & Duker, A. A. (2008). Spatial dependency of V. cholera prevalence on open space refuses dumps in Kumasi, Ghana: a spatial statistical modelling. International Journal of Health Geographics, 7(1), 62.

Osei, F. B., Duker, A. A., & Stein, A. (2011). Hierarchical Bayesian modelling of the space‐time diffusion patterns of the cholera epidemic in Kumasi, Ghana. Statistica Neerlandica, 65(1), 84-100.