| With the development of "The Belt and the Road initiative" of China,and more exchanges and cooperation between China and Southeast Asia occurred,the Lancang-Mekong River region has attracted great attention from all of the worlds.Regional transmission with schistosomiasis and liver flukes become an emerging public health problem that has shown a significant impact on the economic development of society.Schistosomiasis is defined as an eliminating disease by the World Health Organization(WHO),and liver fluke is considered as a disease that can be controlled through effective intervention.Hence,from the perspective of regional overall research on the transmission interruption of schistosomiasis and liver flukes,the control and elimination of the diseases are of the important reference value.In this study,we examine the two parasitic diseases on their current status in their endemic countries and regions:(1)Alonge the Yangtze River of China(Schistosoma japonicum),(2)along the Mekong River Basin of Cambodia and Lao People's Democratic Republic(Schistosoma mekongi);(3)Cambodia,Lao People's Democratic Republic,southern Vietnam,and Thailand(Opisthorchis viverrini);(4)Guangxi of China and northern Vietnam(Clonorchis sinensis).Within the purposes of effective control and elimination of those diseases,we established an epidemiological database of schistosomiasis and liver flukes.Using machine learning approaches to access the effects of factors and risk predictions under the perspective of ecological variation and climate changes.On-going surveillance strategies are explored to be further improved together based on the modelling of the number of surveillance sites selected by the best spatial distance.Part 1:Study on the dynamics and risk predictions of schistosomiasis transmission in the Yangtze River Basin and Mekong River BasinPurpose:To explore the risk prediction approaches for the transmission of schistosomiasis japonica in the Yangtze River Basin and schistosomiasis mekongi in the Mekong River Basin by using machine learning models.The key impact factors affecting two schistosomiasis transmission were explored after evaluating the outputs of risk predictions by different models using the best-fitted model,which will provide a scientific basis for eventually eliminating schistosomiasis in Asia.Method:(1)Database of schistosomiasis transmission in the Yangtze River Basin:The data include the results from the 2018 Oncomelania hupensis survey in surveillance sites with the detection of snail presence and absence data in 2018;meteorological data from 2008-2018 in 191 stations,aligned with ecologic,bio-climate,and socioeconomic data.(2)Database of schistosomiasis transmission in the Mekong River Basin:the data include the results from literature review with an extracted prevalence of schistosomiasis at the point or county level in Laos and Cambodia from 2000 to 2018,which also aligned with ecologic,bio-climate,and socioeconomic data.(3)The research on the transmission features of schistosomiasis japonica:First,predicted the snail survival probabilities through 6 machine learning methods;Second,predicted schistosomiasis transmission index(AMSI)using meteorological data by the biology-based model;Third,reused snail information on survival probabilities combined with many variables such as ecologic variables to explore the risk factors of AMSI and the relationship between the factors and AMSI.Third,predicted the transmission risk areas of S.mekongi,through construction machine learning models employed with key variables,such as the ecologic variables.Results:(1)The databased on snail distribution through a survey in 2018 was established with a total of 2,369 surveillance sites in the Yangtze River Basin(Hunan,Hubei,Jiangxi,Anhui,and Jiangsu provinces),and the snail presence rate was 21.6%(511/2,369).The best fitted RF model(AUC=0.922)among six machine learning models,which indicates that the suitable habitats of snails were distributed along the Yangtze River,mainly concentrated in central and southern Anhui,central Hubei,northern Jiangxi,and other regions;The AMSI values of 191 sites ranged from 48.89 to 261 1.43,with an average of 1089.4,including 69 sites with low risk(0<AMSI?900),116 sites with medium risk(900<AMSI?2000),and 6 sites with high risk(AMSI>2000),which also indicates that most parts of the Yangtze River belong to lower risk areas for schistosomiasis transmission;The RF model showed the best performance(RMSE=160.33,R2=0.863),and the main factors that affect AMSI were Bio15(100%),snail survival probability(98.5%),and HFP(95.5%).There exists a linear relation between Bio15(r=0.29,P<0.01),snail survival probability(r=0.13,P=0.03)linear relation and AMSI.(2)For S.mekongi,the infection rate of schistosomiasis mekongi was mainly concentrated in the southern provinces of Champasack in Laos and in the northern provinces of Strung Treng and Kratie in Cambodia,with a maximum rate of 1.1%,ranged 0 from 40.9%.The RF model had the best performance(RMSE=0.037,R2=0.743).The main factors affecting the infection rate of schistosomiasis were Bio10(100%),elevation(88.1%),and Bio15(76.9%),and the relationship between the aforementioned key factors and the rate was nonlinear in the RF model.The risk of schistosomiasis was concentrated in areas along the Mekong River of the southern Champasack province of Laos by RF model prediction.The model predicted a high rate of schistosomiasis up to 30%,which was an underestimated percentage.Conclusion:Under the low prevalence of schistosomiasis in the Yangtze River Basin,the transmission risk of schistosomiasis can be assessed by using AMSI combined with snail survival probability.The machine learning model can be used not only in estimating snail breeding sites but also in accessing the impacts of Bio15 and snail survival probability on AMSI,which showed that seasonal rainfall and snail probabilities have a great on the transmission intensity of schistosomiasis japonica.While,the risk areas of schistosomiasis mekongi distributed along a limited part of the Mekong River Basin,and the infection rate of schistosomiasis mekongi is relevant lower.The transmission risk of schistosomiasis mekongi can be estimated by machine learning model in combination with many variables,such as bio-climate.Bio 18 and elevation have a great impact on the transmission and prevalence of schistosomiasis in the Mekong River Basin.Part 2:Study on dynamics and risk prediction of liver flukes' transmission in the Great Mekong Sub-regionObjective:To explore the transmission characteristics of C.sinensis infections in the Guangxi of China and northern part of Vietnam,as well as O.viverrini infections in the Mekong River Basin,and analyze the risk factors of both species of liver flukes from the perspective of ecology,and find out the geographical isolation boundary between two liver flukes.Methods:(1)Database of C.sinensis:The liver fluke infections within geographic information were from Guangxi China(Data were available from National parasite survey from 2014 to 2016),and Vietnam(accessing data by literature review from 2000 to 2018 in point and county levels),also combined with the eating habitats of consuming raw or undercooked freshwater fish,then aligned with ecologic,bio-climate,and socioeconomic data.(2)Database of O.viverrini:The data was mainly collected from the liver fluke infections through literature review from 2000 to 2018 in Thailand,Laos,Cambodia,and Vietnam within point and county levels,and data on consume raw or undercooked freshwater fish was also collected.Then those data were combined with ecologic,bio-climate,and socioeconomic data.(3)Machine learning approaches were built to estimate the probability of human infection with C.sinensis and O.viverrini.Then we evaluated the accuracy of different prediction models,to select the best performance model in accessing the risk factor of liver fluke and making predictions.In addition,a machine learning classification model was constructed to find out the key factors and boundaries between the two liver flukes.Results:(1)A total of 85 villages and towns in Guangxi were investigated from 2014 to 2016,33 were detected with C.sinensis.38.8%of counties exist consuming raw or undercooked freshwater fish.From 2000 to 2018,a total of 153 literatures were reported on C.siennsis infections in Vietnam,in which 51 sites of C.sinensis infection were found in the northern part of Vietnam,and 31.7%(20/63)of the administrative areas at the prefecture-level exist consuming raw or undercooked freshwater fish.The LM model showed the best fitting and prediction on C.sinensis infection(AUCtraining=0.959,AUCtesting=0.941).The most important factor affecting the infection of C.sinensis was the eating behavior(100%),the probability of infection of C.sinensis in people who eating raw fish was 13 times higher than that of not eating raw fish.(2)From 2000 to 2018,425 sites of O.viverrini infections were reported in Thailand,and eating raw or undercooked fish occurred in 67.5%(52/77)administrative areas at the prefecture-level.In Laos,144 sites were recorded with O.viverrini infections,and 44%(11/25)of administrative areas had the habit of eating raw fish.In Cambodia,O.viverrini infection sites were 134,and eating raw or undercooked fish was 83.3%(15/18)administrative areas at the prefecture-level.In Vietnam,only 18 places found with O.viverrini infections.Though RF model was the best performance on fitting and predicting the O.viverrini infection(AUCtraining=1,AUCtesting=0.824)in the Mekong River Basin.The most important factor to O.viverrini infection is eating raw fish habits(100%),and the probability of infected people who eating raw fish was 3.3 times higher than those who do not eat raw fish.(3)Ecological and bio-climate variables were of significant differences between C.sinensis infection and O.viverrine infection,for example,the average elevation of C.sinensis endemic area is 35.44 m above sea level,which is lower than that in O.viverrini endemic area(160m above sea level).The classification accuracy of LM,RF,GBM,DT,and XGBoost models was 1 for both C.sinensis and O.viverrini infections.The importance analsyis of variables showed that the factors contributing more than 75%to the classification of C.sinensis and O.viverrini were Bio8,Biol,Bio3,and Bio4 respectively.The model can identify the boundary of two species of liver flukes infections(except NNET model).DT,GBM,and XGBOOST models have predicted the potential risk of O.viverrini infections in the western part of Guangxi,except for LM model.According to prediction results of the LM model,the co-endemic areas of two liver fluke infections occurred in four provinces in the northwestern part and in the upper two provinces of Vietnam,which indicating that those areas are potential co-endemicity areas of C.sinensis or O.viverrini infections.Conclusion:The ecological,bio-climate,and other variables in combination with machine learning methods can be used to predict the risk areas of liver fluke's infections in the Mekong River Basin,and eating fish raw habits is the key factor in the infection of liver fluke.The ecological environment patterns of the two species of liver flukes' areas were significantly different,and the machine learning model could distinguish the distributed areas of C.sinensis and O.viverrini according to the ecological environment and bio-climate factors.The co-endemicity areas could co-exist transmission areas of both C.sinensis and O.viverrini.Whereas the co-endemicity areas existed in western Guangxi province of China,as well as in four northwest provinces and two middle and upper provinces of Vietnam.Part 3 The spatial re-sampling effectiveness in surveys of Oncomelania hupensis habitats and Opisthorchis viverrini infectionsPurpose:In order to improve the surveillance system of schistosomiasis and liver fluke,a spatial re-sampling approach was explored to maintain the same predicted accuracy results by application of the RF model in estimation of the Oncomelania snail infestation in the Yangtze River Basin and O.viverrini infections in Mekong River Basin,when reducing the number of surveillance sites at a maximum level,compared with routine surveillance approach.Method:First,based on the database of Oncomelania snail distribution in the Yangtze River Basin and infections in the Mekong River Basin,using the spatial re-sampling methods to split the study areas into multiple hexagons with equal area,each cell grid represented an ecological region,then a random point was selected within each grid.The spatial distance of adjacent hexagons can be changed to control the size and number of hexagons grids.Second,the machine learning model is used for model fitting and prediction of the re-sampling data and evaluated the performance of the model under different spatial distances of spatial re-sampling.Five scenarios of spatial distances were tested for snail infestation in the Yangtze River basin and O.viverrini infections in the Mekong River basin.Results:The results of the RF model to predict snail infestation in the Yangtze River Basin showed that when the spatial distance was set as Okm,5km,10km,50km,100km,and 150km,the number of hexagonal grids were 0,1258,578,126,62,and 29,respectively.The predicted AUC of the model was 0.886,0.889,0.832,0.723,0.815,and 0.857,respectively.With the Kappa values decreased from 0.647 to-0.682,and the consistency also gradually decreased.When the space distance was 5km,the model prediction result is best one,highly consistent with the original data.Results from the RF model to predict the infection of O.viverrini infections in the Mekong River Basin showed when setting the space distance as 0,5km and 10km,50km,100km,150km,the number of hexagonal grids was 0,1318,1257,603,287,122,respectively.Along with the model predicts AUC were 0.823,0.784,0.824,0.682,0.609,0.736,respectively.And their Kappa value were 0.454,0.429,0.903,0.816,0.963,0.176,respectively.It is demonstrated that when the space distance set to 10km,the best model prediction was presented with its result consistent with the original data.Conclusion:Spatial re-sampling approach can be carried out by setting a certain number of hexagonal grids in the study area with a promising prediction result in the study on resampling the distribution of species or disease by using a minimum number of surveillance points.The number of snail surveillance points in the Yangtze River Basin can be set with a 5km space distance,reducing the original surveillance sites down to 1338.While in the Mekong River Basin,the number of live flukes(C.sinensis and O.viverrini)surveillance sites can be set with a 10km space distance,reducing the original surveillance sites down to 1,257.This strategy could achieve the purpose of predicting the risk areas of liver flukes in the Mekong River Basin without losing the accuracy of the surveillance programme. |
PDF Full Text Request