| Objective The spatial and temporal clustering of brucellosis in Gansu Province was explored in order to identify the high risk clustering period and area,and further explored the cumulative lag effect of various meteorological factors on the brucellosis incidence in Yongchang county and Jingyuan county,and establish the efficient prediction model,providing references for further active and effective work of brucellosis prevention and control.Methods The study collected monthly reported incidence data of brucellosis in Gansu Province and various districts and counties from 2015 to 2019,as well as monthly average meteorological data of Yongchang County and Jingyuan County during the same period.Exploring the spatiotemporal clustering of brucellosis in Gansu Province through spatial autocorrelation analysis and spatiotemporal scanning analysis.In the high-risk cluster area,Yongchang County in the northwest region and Jingyuan County in the central region were selected as the research areas.The meteorological variables related to brucellosis in the two regions were screened through Spearman correlation analysis,and the cumulative lag effect of meteorological factors on brucellosis in Yongchang County and Jingyuan County was explored by establishing the distribution lag nonlinear model.Random forest prediction model based on historical case number and meteorological factors was established in Yongchang County,and SARIMA model was established to evaluate the prediction effect of the model on brucellosis according to R~2 and RMSE.Results 1.The incidence of brucellosis in Gansu Province from 2015 to 2019 was conducted the examination of linear-by-linear association,and the results showed that the overall incidence of brucellosis was decreasing year by year(P<0.05).The incidence curve of brucellosis has an obvious seasonal trend,and the major peak is from April to July,while the minor peak is from September to October.2.The global spatial autocorrelation results showed that the Moran’s I value of brucellosis incidence in Gansu Province from 2015 to 2019 were 0.442,0.422,0.368,0.418 and 0.483,and Z value were 5.47,6.11,5.78,7.00 and 7.25 respectively(P<0.01).Local spatial autocorrelation results showed that high-high concentration areas of brucellosis from 2015 to 2019 in Gansu Province were mainly located in the northeast and central and western regions of Gansu,and the low-low concentration areas were mainly located in southeast of Gansu.The results of spatial-temporal scanning analysis showed that four different levels of aggregation areas were detected from 2015 to 2019 in Gansu Province.The first level aggregation area included 16counties,and the aggregation period was 2018 to 2019;The secondary aggregation area included 20 counties,and the aggregation period was 2015 to 2016;The third level aggregation area included 7 counties,and the aggregation period was 2015;The fourth level aggregation area included one county,and the aggregation period was 2019(P<0.05).3.The exposure-response relationship between the monthly average temperature and brucellosis in Yongchang was J-shaped.When the temperature was 20.5°C with a lag of 0 days,the RR value was the largest,which was 1.60(95%CI,1.130-2.265).At a temperature of-11°C and a lag of 0 days,the RR value was the smallest,0.53(95%CI,0.336-0.849).4.The exposure-response relationship between the monthly average wind speed and brucellosis in Yongchang was W-shaped.When the wind speed was 2m/s with a lag of 0 days,the RR value was the largest,which was 0.19(95%CI,0.049-0.788).5.The exposure-response relationship between the monthly air pressure and brucellosis in Yongchang and Jingyuan was inverted V-shaped.The RR value was the largest in Yongchang when the air pressure was 790.5h Pa and lag 1 month,which was1.12(95%CI,1.008-1.246),and the RR value was the smallest when the air pressure was 806.5h Pa and lag 2 months,which was 0.11(95%CI,0.025-0.497).The RR value was the largest in Jingyuan when the air pressure was 857h Pa and lag 2 months,which was 1.58(95%CI,1.153-2.171),and the RR value was the smallest when the air pressure was 861.5h Pa and lag 2 months,which was 0.84(95%CI,0.737-0.977).6.The exposure-response relationship between the monthly average sunshine duration and brucellosis in Jingyuan was S-shaped.The maximum RR value was 1.80(95%CI,1.123-2.875)when the sunshine duration was 305h and lag 2 months.The min RR value was 0.69(95%CI,0.487-0.991)when the sunshine duration was 295h and lag 0 days.7.The exposure-response relationship between the monthly average relative humidity and brucellosis in Jingyuan showed the inverse J-shaped.The RR value was the largest in Jingyuan when the relative humidity was 36%and lag 1.5 months,which was 1.41(95%CI,1.008-1.962),and the RR value was the smallest when the relative humidity was 68%and lag 0 days,which was 0.68(95%CI,0.474-0.997).8.R~2 and RMSE of training set in random forest model were 0.903 and 1.609;The R~2 and RMSE of the test set were 0.802 and 3.015,respectively,while R~2 and RMSE in SARIMA model were 0.530 and 7.008.Conclusions 1.The incidence of brucellosis in Gansu Province from 2015 to 2019has an obvious seasonal trend,mainly concentrated in summer and autumn.2.The incidence of brucellosis from 2015 to 2019 in Gansu Province had obvious spatial clustering,and the high incidence cluster area shifted from the northeast to the central and western countries of Gansu Province.Among them,Yongchang,Minqin,Liangzhou and Pingchuan were the main gathering areas,which were the key areas of brucellosis prevention and control in Gansu Province.3.The nonlinear hysteresis relationship between the monthly average temperature and the incidence of brucellosis in the Yongchang area is J-shaped,and it is W-shaped with wind speed.The non-linear lag relationship between relative humidity and brucellosis incidence in Jingyuan area shows an inverse J-type,and an S-type relationship with sunshine duration.The incidence and pressure of brucellosis in both places are inverted V-shaped.4.The random forest model based on historical cases and meteorological factors had better prediction effect than SARIMA model,which was more practical. |
PDF Full Text Request