Spatial And Temporal Distribution And Prediction Of Pulmonary Tuberculosis Based On Spatial Autocorrelation And SARIMA-BPNN Combined Model

Purpose:1.Explore the spatial aggregation characteristics of tuberculosis incidence in China from 2012-2021 through spatial autocorrelation analysis,and find hotspot areas and abnormal areas of tuberculosis incidence.2.Establish the best prediction model using the time series data of monthly incidence rate of TB in China.3.To reveal the spatial and temporal characteristics of TB epidemic in China,and to provide scientific basis for formulating targeted prevention and control strategies,so as to improve the capacity and level of public health surveillance,promote a more rational and effective allocation of health resources,and achieve the purpose of accurate prevention and control of TB.Method:1.The annual incidence rate of tuberculosis in 31 regions(except Hong Kong,Macao and Taiwan)from 2012 to 2021 was collected through the "Public Health Science Data Center" and "China Health Statistical Yearbook"."The monthly incidence rate of tuberculosis in China from January 2012 to December 2022 was collected through the data published by the Bureau of Disease Prevention and Control of the National Health Care Commission and Disease Surveillance.2.ArcGIS 10.2 software was used to draw a grading map of TB prevalence using the 5-level natural interruption point grading method,and Geoda software was used to conduct spatial autocorrelation analysis.3.The monthly incidence rate of tuberculosis in China from 2012-2021 was compiled using Excel 2021 software,and the data from 2012-2021 was used as modeling data,and the monthly incidence rate of tuberculosis in China from January to December 2022 was used as prediction data,and the SARIMA model was built using Eviews 10.0 software,Matlab R2016b software,respectively.platform to construct the BPNN model,and the combined SARIMA-BPNN model.The prediction effects of the three models were judged by analyzing and comparing each goodness-of-fit index(AIC,BC)and error index(RMSE,MAE,MAPE)of the three models.Results:1.during the 10-year period of 2012-2021,the level of tuberculosis reporting rate in China gradually decreases from 70.62/100,000 in 2012 to 45.37/100,000 in 2021.2.During the 10-year period of 2012-2021,Xinjiang and Tibet areas were perennially high incidence areas,and Tianjin,Beijing,Shandong Province,and Shanghai were long-term low incidence areas;the largest number of low incidence areas in 2018 and 2019 included Gansu Province,Ningxia Hui Autonomous Region,Shanxi Province,Tianjin City,Beijing,Shandong Province,and Shanghai.3.The results of global spatial autocorrelation analysis showed that Moran’s I was positive in all regions from 2012 to 2021,with z-values of 3.0269,3.3371,2.9648,3.1086,3.4959,3.6404,3.9656,4.2588,4.2642,and 4.0428,which were all greater than 1.96,and their p-values were are less than 0.05,indicating that there is a global spatial autocorrelation for each year in these 10 years,i.e.,there is a regional trend of aggregation in space.4.Getis-Ord Gi*hotspot results showed that the main hotspot areas of TB incidence were:Xinjiang Hui Autonomous Region,Tibet Autonomous Region,Qinghai Province,Yunnan Province,Sichuan Province,Hunan Province,and Gansu Province,and the main coldspot area was:Hebei Province.5.Anselin Local Moran’s I clustering and outlier analysis results show that there are 5 high-high(HH)clustering areas,including:Xinjiang Uyghur Autonomous Region,Tibet Autonomous Region,Qinghai Province,Sichuan Province,Yunnan Province,and TB hotspot areas basically coincide;there are 2 outlier low-high(L-H)clustering areas,including:Yunnan Province,Gansu Province;there is 1 low-Low(LL)aggregation area:Hebei Province,which is in complete agreement with the cold spot area.6.Using monthly TB incidence data from January 2012 to December 2021 in China,SARIMA(0,1,1)(0,1,1)12 model,BPNN(12-7-1)neural network model,and SARIMA(0,1,1)(0,1,1)12-BPNN(1-5-1)combined model were successfully established,and the fit of the models during the fitting period superiority indicators were RMSE=0.55,MAE=0.35,MAPE=5.66%;RMSE=0.53,MAE,0.41,MAPE,8.3%;RMSE=0.35,MAE,0.31,MAPE,4.2%,respectively.The error indicators predicted by the model for the 12 months of 2022 are RMSE=0.58,MAE=0.43,MAPE=12.51%;RMSE=0.63,MAE=0.46,MAPE=13.67%;RMSE=0.30,MAE=0.20,MAPE=6.01%.It can be seen that the model with the best fitting and prediction is the combined SARIMA-BPNN model,followed by the SARIMA model and finally the BPNN neural network model.Conclusion:1.The incidence of tuberculosis in China’s provinces has been decreasing year by year for the past 10 years,which confirms the effectiveness of tuberculosis prevention and control measures in China,but still falls short of the WHO’s goal of "ending tuberculosis strategy".2.There is a global spatial aggregation of TB incidence in China,with Xinjiang Uyghur Autonomous Region,Qinghai Province,Tibet Autonomous Region,and Sichuan Province being hotspots year-round and also areas of high-HH aggregation,and the prevention and control management of TB in these areas should be strengthened.3.The seasonal characteristics of tuberculosis incidence in China are significant,with the peak incidence in spring and the trough in winter.4.The SARIMA model,BPNN neural network model,and SARIMA-BPNN combination model,which were established based on the time series data of monthly incidence rate of TB in China,all had better performance in predicting the incidence of TB in China,and the SARIMA-BPNN combination model had the best prediction effect,which can be used as the preferred prediction model to provide reference for the formulation of TB prevention and control measures.