The Application Of Spatiotemporal Projection Model In Eastern Coastal Area Of China

Background:Through the entering of marine civilization era,coastal cities have gradually become the focused region where capital,human resources,advanced science and technology could be efficiently combined.However,with rapid economic development,especially in the southeastern coastal cities,local ecological environment is facing serious pollution,which leads to high health risk for residents.Besides,aloong with the rapid urbanization,people's diet and living habits are changing and could enjoy more convenient medical service.These all lead to the increasing of incidence and detection rates of chronic diseases(such as cancer,cardiovascular and cerebrovascular diseases).Breast cancer(BC)is one of the the main causes of death for female cancer patients in China.Compared to that of inland cities,BC incidence is quite high in coastal cities.New methods and techniques are demanded to analyze the distribution and etiology of BC cancer.Mainstream mapping techniques,like spatiotemporal kriging(STK),can explore the distribution and internal structure of spatiotemporal process data,construct spatiotemporal variogram functions,predict and generate spatiotemporal distribution maps,which can offer valuable information for risk assessment and health management.Yet,the implementation of these techniques experiences substantive and technical complications,which are due mainly to the different characteristics of space and time and their connection to incidence distribution.Methods:Hangzhou city was selected as the study area.All the newly diagnosed female BC cases[International Classification of Disease 10th edition(ICD-10)code:C50]in Hangzhou from 2008 to 2012 were obtained from the Center for Disease Control and Prevention(CDC).According to the patiens' detailed household register information,all these cases were allocated to 200 townships.BC incidence rate of each township was calculated as the number of cases per 100,000 population based on the population data obtained from the Hangzhou Public Security Bureau(PSB).Chi-square test was employed to compare incidence disparities among various sub-populations and regions.After that,based on the spatiotemporal random field(S/TRF)theory,Spatiotemporal Projection(STP) technique coupled with Bayesian Maximum Entropy(BME)were implemented to predict the Spatiotemporal distribution of BC incidence in Hangzhou city.Meanwhile,for comparison purpose,both the STP and the mainstream STK technique were used to generate BC incidence maps in Hangzhou.Finally,Moran's I test,Hot Spot analysis,Anselin Local Moran's I analysis and Space-time Scan Statistics were employed to detect the spatiotemporal pattern of estimated BC incidence.Results:Although BC incidence in Hangzhou was relatively steady during the period of 2008-2012,signigicantly geographical incidence disparity was exsited.People living in subdistrict area showed the highest BC incidence,was about 1.6 times of that in town area and about 3.2 times of that in country area.The inter-connection between space,time,BC incidence,and disease spread velocity established by the STP technique produces a more realistic representation of the actual BC incidence distribution across space-time.By,transforming the space-time domain into a spatial domain,the STP technique eliminates the theoretical and practical difficulties of traditional space-time mapping techniques,thus making the estimation more accurate and the calculation more efficient.STP performedconsiderably better than STK in terms of(1)the STP provides an improved asessment of the space dependence structure of BC disease incidence;(2)lower mean error(ME),mean absolute error(MAE)and root mean square error(RMSE)values of the BC estimates generated by the STP technique,and more accurate STP-generated incidence maps compared to the STK-generated ones;and(3)an almost perfect fit to the 5-year average BC incidence values.According to the results of spatiotemporal pattern analysis,STP-Estimated BC incidence in Hangzhou is temporally stable and spatially heterogeneous,northeast coastal downtown areas were high incidence clusters with significantly higher BC risk,whereas large parts of southwestern inland areas were low incidence clusters with significantly lower BC risk.Estimated BC incidence revealed an increasing trend from the southwest inland region to the northeast coastal region.Conclusions:Compared to the mainstream mapping techniques like STK,the STP technique,based on the theory of S/TRF and BME,can build the inter-connection among space,time,spatiotemporal attributesand spread velocity.Based on the three-fold idea of"Transform-Solve-Backtransform",at the transformation stage,STP technique would temporarily "compress" the time information,solve the BC mapping problem in a simpler domain of the "compressed" time data,then release the "compressed" time data information at the backtransformation(final)stage.STP technique generats BC incidence more accurate,more informative and involve considerably lower computational costthan STK,which could provide foundations for the detection of spatiotemporal distribution of cancer,relating to corresponding environmental risk factors,exploration of cancer etiology and cancer prevention and public health resources allocation.