GEOSPATIAL ANALYSIS AND MODELING OF HIV EPIDEMIC

Abstract

Health is vital for all of us and understanding the determinants of a disease, its spread from person to person and community to community has become increasingly global. There are various factors which are the reasons for the emergence of diseases. The characteristics of the locations (including socio-demographic and environmental exposure) offer a valuable source for epidemiological research studies on health and the environment. Since health is a geographical phenomenon and various factors attributing to the health diagnostics and planning are geography dependent, as such, Geographic Information System (GIS) for health studies serves as an important tool. GIS supported by spatial data infrastructure and vibrant routine health data can give planners valuable information to address the issues related to HIV/AIDS and support monitoring, evaluation and planning. GIS can also be used as an effective tool to manage and monitor HIV/AIDS and related routine activities. As health is largely determined by spatial factors (including the socio-cultural and physical environment, which vary greatly in space), it always has an important spatial dimension. The spatial modeling capacities offered by GIS can help to understand the spatial variation in the incidence of disease, and its co-variation with environmental, social and demographic factors. With more HIV infections than any other country in the world, India gives the impression that HIV infection is common and that there is a severe epidemic in the country. There are 2.47 million persons in India living with HIV, equivalent to approximately 0.36 percent ofthe adult population. Thetransmission route is still predominantly sexual (87.4 percent); other routes of transmission by order of proportion includes perinatal (4.7 percent), unsafe blood and blood products (1.7 percent), infected needles and syringes (1.8 percent) and unspecified and other routes of transmission (4.1 percent. The HIV epidemic is largely concentrated in six Indian states; 0.07% in Uttar Pradesh, 0.34% in Tamil Nadu, 0.62% in Maharashtra, 0.69% in Karnataka, 0.97% in Andhra Pradesh and 1.13% in Manipur state. This research is an attempt to understand how GIS can be put to use for spatial epidemiology, in this case the focus is on HIV incidence. The study area in the present research work covers the state of Andhra Pradesh (AP) in India. AP is the second worst HIV/AIDS affected state of India, after the state of Manipur. National AIDS Control Organization (NACO), New Delhi, Ministry of Health and Family Welfare, Government of India, provided with the HIV data collected over a span of seven years from 2001 to 2007. Data beyond 2007 was not available at the time of study as it was not compiled centrally, he test data as collected from NACO is based on different indicators. This study uses indicators on Voluntary Counseling and Testing Centers (VCTCs) and Prevention of Parent to Child Transmission (PPTCT) Centers. The first part of the research comprises of spatial analysis and geovisualization. In this part, work is done on the spatial mapping of HIV incidence at district level and then the sub-district level (mandate) in Andhra Pradesh. Static maps for years 2001 to 2007 were produced at two different scales to understand the patterns of HIV and its movementthrough space and time. Other tools for geovisulazation like graphs, scatter-plots, cartograms and ring maps were also developed. Temporal animation was used as a technique for dynamic visualization. This work is followed by the HIV Cluster analysis: It helps in identification of the regions of elevated risks which can then be associated with other explanatory variables to understand the reasons behind the higher risks at a given place. Two sets of clustering algorithms were used: The Getis-Ord Gi* algorithm in ArcGIS and spatio-temporal clustering in SaTscan. The results were mapped and compared. Often there is the problem of missing data in the epidemiological studies. There are many techniques in literature which have been designed to find the data at the un-sampled locations. In this part of work, the use of a geostatistical technique- kriging for the generation of risks maps has been attempted. Kriging is a tool to generate isolpleth maps, which interpolates data based on the principle of autocorrelation. Clustering and kriging have been attempted at the mandal level. After this generation of an HIV index using regression analysis was attempted at the district level. Forty four different explanatory variables, out of which thirty five were social, economic, health and demographic factors taken from the District Level Health Survey - 2007, and the Census of India and nine, were derived spatial factors like road density, proximity to roads and type of landuse. HIV incidence per district was correlated with each of the forty four variables using scatter plot analysis and the Pearson's correlation coefficient. Twenty six significant and positively correlated variables were retained. These were fed as input to our HIV model using OLS (Ordinary Least Regression) technique. This part could explain the most crucial factors playing a role in the HIV incidence. This was followed by modeling the movement of females for moving for HIV testing. The analysis used techniques of mapping, clustering and SAR modeling to deal with the issue of movement to different HIV testing centers. The later work involves a prototype study involving the application of unique GIS functionality of network based analysis for the planning of new VCTCs in Hyderabad. This work also highlights the utility of network analysis for finding the shortest route to different hospitals and finding the nearest facilities from a given location anywhere in the network. The final part of the research involves development of a web based application for VCTC location. It involves development of a small web application using the Google Maps API and HIV incidence data from different VCTCs. Using this application one can know the presence of a VCTC in a given mandal and the incidence at mandate having a testing facility. The result of spatial analysis and geovisualization include graphs, cartograms, chloropleth maps and ring maps. Though graphs are informative, they do not help us to understand the geographical variation of the variable in interest. Other results include chorolpleth maps, cartograms, ring map and animation respectively. These maps are especially useful in medical geography to present multiple information through one map and make map reading more convenient. Cluster analysis was done to identify the clusters of high alarms across the study area. The drawback of doing clustering in ArcGIS is that it does not provide spatio-temporal clustering. Though SaTscan provides both spatial and the spatio-temporal clustering but it does not have a graphical support. Hence though the results are more meaningful, they are in the tabular format which needs to be imported to some GIS platform to meaningfully map them and hence needs extra time. The results from both were compared and both show high rate clustering in the south west zone especially in the districts of Guntur, Krishna and West Godavari. In the case of kriging, different kriging models were used Spherical model used for the second order trend removed ordinary kriging gave the lowest RMSE of 10.77 with the standardized mean (0.04123) for the 2007 data. Modeling of HIV incidence was done using the Ordinary Least Squares (OLS) regression technique in ArcGIS environment. For the model to be correctly specified the residuals should not be auto-correlated. Moran's I test was done to check autocorrelation for the residuals and the result was a random pattern with the Moran's I index = -0.03 and a Z score of 0.09 SD. Movement Modeling was a step further from the OLS modeling. SAR (Spatial Auto-regression) modeling was attempted to model the movement of females going for testing. Type of VCTC, number of VCTCs, distance from major roads, road density and number of neighboring mandate were used as the explanatory variables. The dependent variables were the three classes of females going for testing, these were Refs (Iref): females who are referred by the doctors to go for HIV testing, DWs (low): Females who directly walk in to the test centers and Pregnant (Ip): Females who get tested since they have go to a ANC (Anti Natal Clinic) for check up. Models were build using SAR modeling for all the three types. Relatively low R2 values ranging from (0 to 0.05), (0 to 0.07) and (0.01 to 0.3) for Iref, Idw and Ip as the response variables respectively are observed. The highest R2 value equal to 0.307 was observed for the relationship of the type of facilities with Ip. All other variables do not significantly contribute to the incidence of any of the three categories. None of the variables unambiguously explained the behavior of the type of tested females. Therefore, although it seems that females might be moving one cannot exactly capture the movement and the attributed reasons do not fully explain any of the hypothesized phenomena. The regions where the incidence in pregnant women is higher than the general population were identified as the zones of movement and similarly those with high DWs; however no significance or a consistent cause could be attributed to this. The results of the network based analysis gave ideas about where in Hyderabad there is paucity of testing facilities depending upon the population demands and the proximity from roads. It was also shown that such analysis can be used to find the shortest distance between any two pints and the closest hospital facilities from your location. Finally, the web application for VCTC location gives details at the mandal level where facilities for HIV testing are available. With the available data attempts were made to map HIV and bring forth various techniques of visualization. The analysis provided spatial temporal patterns of disease progression through 2001 -2007. The results show a distinct pattern in years 2004 - 07, HIV incidence is always high in the south-western coastal zone of the state and in the state capital Hyderabad. Clustering is considered to be very important in epidemiology and the results of clustering, like visualization confirmed that the disease mainly has 'hot-spots' in the south western edge of the state. An analysis was done to find out the number of VCTCs in each district to check if the high incidence in these zones is because of higher testing centers. The high incidence zones and high VCTCs necessarily do not co- exist which points that there are other factors also which are responsible for elevated HIV presence. One reason may be the presence of a dense highway network in these zones. Also, reports establish that these areas are very conducive for sex-trade because of a high presence of vehicle routing and the coastal trade. These reasons might be contributing to an elevated risk. Modeling to an extent helped in understanding the underlying factors which are playing a major role in higher incidence. Still, it was felt that the factors which play a role are usually spatially varying at much higher scales like mandate. If, detailed data are available at the mandal level, these variations can be captured more profoundly and a more effective model can result. SAR modeling done at the mandal level for the three groups of females showed that one can only partially, i.e. non-significantly, explain the relations with the explanatory variables when explaining movement. This leads us to assume that at the scale of the study and the available data, much of the movement is random and that some additional parameter set should be collected to exactly identify where people are moving and what factors are governing them in their behavior. Network modeling is important for policy makers. Real time data and a web based network application can be a huge asset for improving accessibility to testing centers. Through this study, an attempt was made to maximize the use of GIS in analyzing and understanding HIV scenario in Andhra Pradesh. As a whole, this work can be taken further and the results can be more refined provided long-term and complete data at a higher scale are available. The study is a first-hand attempt to guide people working in the HIV policy areas to locate the particular zones of interest in AP. Also, it would help in identifying the important risk factors which can be laid emphasis on, in order to combat HIV. GIS has been widely used in developed countries in HIV and other disease mapping and modeling, while it is still in infancy in India. IV