基于不同空间权重设置下的空间滤波模型在手足口病发病数据中的应用  

Application of spatial filtering model based on different spatial weight settings in hand-foot-mouth disease incidence data

作　　者：陈希 李柯 殷韵 刘元桦 洪洁 石瑾 黄家祺 赵峥 徐珈瑶 袁睿 张志杰 CHEN Xi;LI Ke;YIN Yun;LIU Yuanhua;HONG Jie;SHI Jin;HUANG Jiaqi;ZHAO Zheng;XU Jiayao;YUAN Rui;ZHANG Zhijie(Department of Epidemiology,School of Public Health,Fudan University,Shanghai 200032,China)

机构地区：[1]复旦大学公共卫生学院流行病学教研室,上海200032

出　　处：《中华疾病控制杂志》2023年第11期1262-1267,共6页Chinese Journal of Disease Control & Prevention

基　　金：国家自然科学基金(81973102)。

摘　　要：目的研究不同空间权重矩阵设置下的空间滤波模型在华东地区手足口病(hand-foot-mouth disease,HFMD)发病数据中的应用,并比较不同空间模型的效果,评估其应用价值。方法收集华东地区2009年HFMD的发病数据和相关影响因素。通过空间滤波方法(eigenvector spatial filtering,ESF)将4种不同的空间权重矩阵分解,根据莫兰指数(Moran′s I,MI)和逐步回归筛选出特征向量作为空间滤波器带入模型,通过赤池信息准则(Akaike information criterion,AIC)、偏差信息准则(deviance information criterion,DIC)和均方根误差(root mean square error,RMSE)比较不同权重矩阵的影响,最后从模型系数的拟合值和标准差与置信区间的角度分析比较基于最优权重矩阵的空间滤波模型与贝叶斯空间模型。结果2009年华东地区共报告403607例HFMD,主要分布在山东省西部和浙江省东南部地区,MI显示华东地区HFMD具有空间相关性。普通的负二项分布模型加入空间滤波器后,空间滤波模型残差的空间相关性被有效去除(MI分别为-0.11、-0.15、-0.08和-0.09,均P>0.05)。Rook权重矩阵为最优权重矩阵,并且最优权重矩阵下的空间滤波模型和贝叶斯空间模型的回归系数值结果接近,但是空间滤波模型的标准差和置信区间小于贝叶斯空间模型。结论空间滤波模型计算简单、结果准确,能够反映从整体到局部在不同地理尺度下的地图模式,揭示疾病发病的潜在空间结构,可作为传统复杂空间模型的有效替代方法。Objective To study the application of spatial filtering model in the incidence data of hand-foot-mouth disease(HFMD)in East China given different spatial weight,and to determine its applicability by comparing the effects of different spatial models.Methods The incidence data of hand,foot and mouth disease in East China in 2009 were collected and the related influencing factors were identified.Four different spatial weight matrices were decomposed using the eigenvector spatial filtering method(ESF),and the eigenvectors were determined according to Moran′s I(MI)value and stepwise regression,which was introduced as the spatial filter into the model.The effects of different weight matrices were compared by Akaike information criterion(AIC),deviance information criterion(DIC)and Root Mean Square Error(RMSE).Finally,the spatial filtering model based on the optimal weight matrix was compared with the Bayesian spatial model in terms of the fitting value,standard deviation and confidence interval of the model coefficients.Results There were a total of 403607 HFMD cases reported in East China in 2009,most of which concentrated in the west of Shandong Province and the southeast of Zhejiang Province.According to MI test,HFMD exhibited spatial correlation in East China.After the spatial filter was introduced into the normal negative binomial distribution model,the residual of the spatial filter model ceased to show spatial autocorrelation(MI were-0.11,-0.15,-0.08 and-0.09,respectively,all P>0.05),and the spatial autocorrelation was effectively removed.The Rook weight matrix was considered the optimal weight matrix.Although,the regression coefficient of the spatial filtering model under the optimal weight matrix were comparable to that of the Bayesian spatial model,the spatial filtering model was still significantly outweighed by the Bayesian spatial model in terms of standard deviation and confidence interval.Conclusions The spatial filtering model demonstrates the advantages of simple calculation and accurate results.There

关 键 词：手足口病 空间权重矩阵 空间滤波模型 空间流行病学 

分 类 号：R183[医药卫生—流行病学] R188[医药卫生—公共卫生与预防医学]