2017—2023年青海省其他感染性腹泻病发病率的时空分析及预测模型比较  

The spatio-temporal analysis and prediction model comparison of incidence rate of other infectious diarrhea diseases in Qinghai Province from 2017 to 2023

作　　者：姜雨淇 龙江 赵金华 张华一[3] 邓萍[1] 姜文琦 JIANG Yuqi;LONG Jiang;ZHAO Jinhua;ZHANG Huayi;DENG Ping;JIANG Wenqi(Department of Public Health,Qinghai University Medical College,Xining 810001,China;Institute for Communicable Disease Control and Prevention,Chongqing Center for Disease Control and Prevention,Chongqing 400707,China;Institute for Communicable Disease Control and Prevention,Qinghai Center for Disease Control and Prevention,Xining 810007,China)

机构地区：[1]青海大学医学院公共卫生系,西宁810001 [2]重庆市预防控制中心传染病预防控制所,重庆400707 [3]青海省疾病预防控制中心传染病预防控制所,西宁810007

出　　处：《中华疾病控制杂志》2024年第11期1301-1307,共7页Chinese Journal of Disease Control & Prevention

基　　金：国家自然科学基金(12371503);“昆仑英才·高原名医”项目(青卫健办[2021]104号)。

摘　　要：目的分析青海省其他感染性腹泻病(other infectious diarrhea disease,OIDD)流行情况与变化特点,为2024年青海省OIDD发病率提供预测。方法以2017年1月—2023年12月青海省OIDD的月发病率和年发病率为原始数据,利用Arcgis 10.8软件对青海省年发病率进行地图可视化,使用GeoDa 1.16软件进行空间自相关分析,使用R 4.3.1软件建立青海省OIDD的季节性自回归积分滑动平均(seasonal autoregressive integrated moving average,SARIMA)模型、三次指数平滑法(Holt-Winters)模型、神经网络自回归(neural network autoregression,NNAR)模型、指数平滑空间状态(trigonometric seasonality,Box-Cox transformation,TBATS)模型、先知模型。根据均方根误差(root mean square error,RMSE)、平均绝对误差(mean absolute error,MAE)、平均绝对百分比误差(mean absolute percentage error,MAPE)评价模型拟合效果。结果除Holt-Winters模型之外,各种模型均能较好地捕捉发病率趋势,其中NNAR模型训练集的MAE为0.90、RMSE为1.25、MAPE为16.43,在TBATS等模型中表现最好;NNAR模型测试集除RMSE值大于SARIMA模型和TBATS模型外,MAE和MAPE值均小于其他模型,总体而言预测性能最佳。因此,可基于NNAR模型对2024年青海省OIDD发病率做出预测,为高海拔地区的疾病预防策略做出启示。结论2017—2023年青海省西宁市、海东市、黄南藏族自治州为OIDD的高发地区。模型预测中,NNAR模型的预测效果最好,但在实际情况中需要结合各地区时空特征和流行趋势制定相应的防治措施。Objective To analyze the epidemiological trends and characteristics of other infectious diarrheal diseases(OIDD)in Qinghai Province,and to provide predictions for these diseases in Qinghai Province for 2024.Methods Using monthly and annual incidence rates of OIDD in Qinghai Province from January 2017 to December 2023 as primary data,the study employed ArcGIS 10.8 software for map visualization of annual incidence rates in Qinghai Province,and GeoDa 1.16 software for spatial autocorrelation analysis.R 4.3.1 software was used to construct various models for OIDD in Qinghai Province,including seasonal autoregressive integrated moving average(SARIMA)model,triple exponential smoothing(Holt-Winters)model,neural network autoregression(NNAR)model,trigonometric seasonality,Box-Cox transformation(TBATS)model,and Prophet model.The models′fitting effects were evaluated using root mean square error(RMSE),mean absolute error(MAE),and mean absolute percentage error(MAPE).Results All models,except the Holt-Winters model,effectively captured the incidence rate trends.Among them,the NNAR model performed best in the training set,with MAE of 0.90,RMSE of 1.25,and MAPE of 16.43,outperforming models such as TBATS.In the test set,while its RMSE value was higher than those of the SARIMA and TBATS models,its MAE and MAPE values were lower than other models,indicating the best overall predictive performance.Therefore,the NNAR model can be used to forecast the incidence rate of OIDD in Qinghai Province for 2024,providing insights for disease prevention strategies in high-altitude regions.Conclusions From 2017 to 2023,Xining City,Haidong City,and Huangnan Tibetan Autonomous Prefecture in Qinghai Province were high-incidence areas for OIDD.Among the predictive models,the NNAR model showed the best performance.However,in practical applications,it is necessary to develop corresponding prevention and control measures by considering the spatiotemporal characteristics and epidemic trends of each region.

关 键 词：其他感染性腹泻病 神经网络自回归模型 模型预测 季节性自回归积分滑动平均模型 先知模型 

分 类 号：R183.4[医药卫生—流行病学]