The Establishment Of Integrated Model Of Schistosomiasis Japonica Based On Landscape Pattern Analysis And Bayesian Modeling

With the changes of ecological environment, including global warming, "breaking dikes or open sluice for water storing", and human migration, the ecology of Oncomelania hupensis, the intermediate host of Schistosoma japonicum, will be also changed in terms of habitats, biocenosis, and density, which finally resuls in changes of transmission of schistosomiasis japonica. Therefore, the way of using the scientific prediction models to predict the epidemic status of schistosomiasis japonica become more and more important approach in the disease control and prevention, and it is also one of the increasingly priorities in the process of disease control.In this study, we used geographic information systems (GIS) and remote sensing (RS) technology to develop Bayesian models for prediction of the distribution of Oncomelania snail and schistosomiasis integrated with landscape pattern analysis by employing environmental factors, e.g. water, vegetation, and land surface temperature, landscape factors, e.g. land-use/type, and socio-economic factors as covariables, in order to understand and predict the spatio-temporal patterns of schistosomiasis in different environmental settings under the same scale or the same environmental setting with different scales. The following four investigation aspects were performed with the purpose of providing scientific results contributed to formulation of a more effective strategy for control and prevention of S. japonicum transmission in both mountainous and lake regions of China.Firstly, an area-wide ditch map with the boundary of villages was generated in the study area, namely Eryuan county of Yunnan province, by tracing the ditch network on foot by use of a global position system (GPS) unit, and the data on the distribution and density of Oncomelania snails in the study area were extracted from the annual schistosomiasis records of Eryuan county recoreded from 2000 to 2006. The varaiables, e.g. normalized difference vegetation index (NDVI) , wetness, land surface temperature (LST), and land-use/type, were extracted from remote sensing images, and then landscape metrics were calculated. The spatio-temporal Bayesian models with area data were established at village scale, and then spatial Bayesian model with point data was established using the data of snail survey and SPOT5 satellite image at local scale (or snail habitat). The results indicated there was no significant spatial and temporal correlation of live and infected snail densities at village scale, but there was spatial correlation at local scale. Hence, spatial Bayesian model was used to predict the distribution of snails at local scale. The correlation between the snail density and NDVI, wetness and the slope of ditch was significantly presented at village scales, however this correlation was not significant at local scales. The correlation between snail density and mean shape index (MSI) and Shannon's evenness index (SEI) was significantly presented at local scale. A prediction map was generated by the Bayesian model employing with environmental surrogates and landscape metrics at local scale, and findings of the study suggested that decreasing the heterogeneity of the landscape can reduce snail density and the established model by using higher resolution satellite data at local scale was suitable to be applied in the mountainous region.Secondly, residents aged over 5 years old were screened for S. japonicum infection using indirect haemagglutination test (IHA) and micracidium hatching method. Bayesian multilevel models including spatial correlation were built for serological status and the underlying infection status of S. japonicum, respectively, at the three levels, e.g. individual, family and village. The variability of the distribution pattern of the serological status and underlying infection of S. japonicum occured within village boundary. At individual level, all resident were susceptible to be infected with S. japomicum, and health education should be strengthened on all individuals. At family level, reducing the area of paddy farmland, and building methane gas pit can decrease the seroprevalence, and building sanitary breeding stall for livestock can decrease the underlying infection rate, respectively. At village level, changing the landscape heterogeneity and snail density around villages can decrease the seroprevalence and the prevalence of S. japonicum infection, respectively.Thirdly, the data about the distribution and density of snail from 1995 to 2006 in Hanshou county,Hunan province were collected, and NDVI, wetness and LST were also extracted from remote sensing images, different Bayesian models were established to predict the distribution of snail. Results showed the negative temporal correlations in distribution of live and infected snail were occurred. The rate of decline in spatial correlation of snail distribution between points inside embankment of lake was faster than that outside embankment. The spatial structure of live and infected snails outside embankment was similar, but the difference of the spatial structure of those snails in each year was large. The correlation between snail density and NDVI was negatively distributed inside embankment but positively outside embankment. The correlation between snail density and LST outside embankment was negatively presented, but positively occurred with wetness. The correlation between snail density inside embankment was positively related to SEI, but negatively related to LPI. The correlation between infected snails and MSI, SDI (Shannon's diversity index) outside embankment was positively presented. Predication maps showed the snail density still remained at a high level after implementation of the project of breaking dikes or open sluice for water storing implemented, the spatial distribution of snail inside embankment was much more clustered than that outside embankment, and the distribution of most snails outside embankment was located in the northwest marshland outside embankment in Hanshou county.Fourthly, the Bayesian models were established by employing the data collected from the periodical surveillance on schistosomiasis where survey performed more than 3 times during last 10 years, with taking into account of the uncertainty in sensitivity and specificity of diagnostic test(s). Results showed that no significantly temporal correlation was occurred in human infection rate with S. japonicum, and the difference of spatial structure of human infection between each year was significant. The correlation between the prevalence of S. japonicum infection and NDVI was negatively presented significantly. The prediction map of S. japonicum infection in 2002 showed the whole prevalence of S. japonicum infection was at a low level, and the areas where prevalence more than 1% were mostly located along water courses of the Muping lake and the Yuanshui river. While the average prediction prevalence was 2.22% in 2005, and the higher risk areas distributed along water courses as well. The spatial patterns of prediction and predicted error were similar between results of serological test and that of stool test. The project map of prevalence of S. japonicum infection showed the changes of infection in the south areas was not significant, while the prevalence increased significantly in north areas to the Yuanshui river, and it was indicated the impact of the implemention of project on partial abandon areas for water storing on prevalence of S. japonicum was stronger than that of the project on completed abandon areas for water storing.Based on the results from the Bayisian models prediction on distribution of snail and schistosomiasis both in the mountainous region and in the lake region, it is found that the differences were significantly existed in the risk factors, spatio-temporal patterns, and model building ways, ect., these differences lead to different control measures in these different environmental settings. For examples, in lake regions, the same or similar measures can be implemented in a large scale, while specific measures should be applied to adapt the unique characteristics at a small scale in mountainous region, in order to improve the efficacy of different control efforts.In conclusion, we have developed an integrated model based on both landscape analysis and Bayesian modeling to predict the distribution of snail and schistosomiasis, and this integrated Bayesian model approach with landscape analysis will become a powerful and statistically robust tool for estimating and evaluating the control strategy at an appropriate scale.