重庆市手足口病多种预警模型的建立与评估

Establishment and evaluation of multiple warning models for hand，foot and mouth disease in Chongqing

摘要:

目的：构建多种重庆市手足口病（hand，foot and mouth disease，HFMD）疫情预测模型并评价模型的拟合和预测效果，为卫生部门提供准确的疾病动态及预警预告奠定基础。方法：对重庆市2009至2014 年HFMD月发病率资料分别建立3种单一预测模型（SARIMA、BPNN和Elman神经网络）和3种组合预测模型（SARIMA-BPNN、SARIMA-Elman、BPNN-Elman），利用2015年HFMD月发病率资料对各个模型进行拟合和评价，并选择最优预警模型预测2016至2017年重庆市HFMD发病率。结果：SARIMA模型、BPNN模型、Elman模型、SARIMA-BPNN组合模型、SARIMA-Elman组合模型、BPNN-Elman组合模型拟合及预测的平均绝对误差（mean absolute error，MAE）、平均误差率（mean error rate，MPAE）、预测准确度（prediction accuracy，P）、非线性相关系数（non-linear correlation coefficient，RNL）和平均相对误差（mean relative error，MRE）分别为0.595、0.515、0.485、0.551、0.493；0.265、0.229、0.771、0.673、0.029；0.234、0.202、0.798、0.737、0.142；0.248、0.215、0.785、0.714、0.101；0.328、0.284、0.716、0.620、0.103；0.229、0.198、0.802、0.745、0.071。综合各个评价指标，选出本次研究最优预测模型为BPNN-Elman组合模型，并采用BPNN-Elman组合模型预测出2016年和2017年重庆市HFMD年平均发病率分别为162.8/10万和160.9/10万，95%置信区间为-0.236~0.184。结论：BPNN-Elman组合模型预测效果最优，能较好预测重庆市HFMD发病情况，为HFMD防控提供科学依据。

Abstract:

Objective：To lay the foundation for the health sector and to provide accurate disease dynamics and early warning by con－structing a variety of prediction models of hand，foot and mouth disease（HFMD） in Chongqing and to evaluate its effect. Methods：Three kinds of single prediction models（SARIMA，BPNN and Elman neural network） and three combined prediction models（SARI－MA-BPNN，SARIMA-Elman，BPNN-Elman） were established for the monthly incidence of HFMD in Chongqing from 2009 to 2014. The models were evaluated by using the monthly incidence data of HFMD in 2015，and the optimal warning model was selected to predict the incidence of HFMD in Chongqing from 2016 to 2017. Results：The results showed that the mean absolute error（MAE），mean error rate（MPAE），prediction accuracy（P），non-linear correlation coefficient（RNL） and mean relative error（MRE） of the SARIMA model，BPNN model，Elman model，SARIMA-BPNN，SARIMA-Elman，BPNN-Elman combined model respectively were 0.595，0.515，0.485，0.551，0.493；0.265，0.229，0.771，0.673，0.029；0.234，0.202，0.798，0.737，0.142；0.248，0.215，0.785，0.714，0.101；0.328，0.284，0.716，0.620，0.103；0.229，0.198，0.802，0.745，0.071. Through comprehensive evaluation of various indicators，the optimal prediction model is BPNN-Elman combination model. The average annual incidence of HFMD of the year 2016 and the year 2017 in Chongqing were 162.8/10 million and 160.9/10 million based on BPNN-Elman combination model with 95% confidence interval of -0.236 to 0.184. Conclusion：The BPNN-Elman combined model has the best prediction effect and can better predict the incidence of HFMD in Chongqing，which provides a scientific basis for prevention and treatment of HFMD.