纳入气象因素的ARIMAX模型预测流行性感冒流行趋势

Influenza incidence prediction based on ARIMAX model including meteorological factors

目的评价纳入气象因素的多元差分自回归移动平均模型(ARIMAX)预测流感样病例(ILI)流行趋势的效果,为流行性感冒监测和预警提供参考。方法收集杭州市余杭区2014年第1周—2018年第26周4家监测哨点医院上报的ILI和同期气象资料;利用余杭区2014年第1周—2017年第52周ILI资料以及经Lasso回归模型筛选的气象变量,建立ARIMAX模型,预测余杭区2018年第1—26周流感样病例占门急诊病例的比例(ILI%),并与实际情况比较以验证模型的预测效果。结果2014年第1周—2018年第26周,余杭区监测哨点医院共报告ILI病例60419例,ILI%为1.29%。Lasso回归分析结果显示,周平均绝对湿度与ILI%呈正相关(r=27.769),周平均气温与ILI%呈负相关(r=-0.117)。纳入周平均绝对湿度和周平均气温建立最优模型ARIMAX(1,0,0)(1,0,0)12,贝叶斯信息准则(BIC)值为81.30,平均绝对误差百分比(MAPE)为15.77%。采用最优模型预测余杭区2018年第1—26周ILI%,MAPE值为43.75%。结论纳入气象因素的ARIMAX模型可对ILI流行趋势进行预测,但预测精度有待提高。

Objective To evaluate the feasibility of autoregressive integrated moving average with explanatory variables(ARIMAX)model including meteorological factors on the prediction of influenza-like illness(ILI),so as to provide a basis for the monitoring and early warning of influenza.Methods The ILI data reported by four sentinel hospitals in Yuhang District of Hangzhou from the 1 st week of 2014 to the 26 th week of 2018 was collected,as well as the meteorological data during the same period.The ARIMAX model was established using the percentage of ILI cases in total outpatients(ILI%)data from the 1 st week of 2014 to the 52 nd week of 2017 and the meteorological factors selected by Lasso regression model.The ILI%from the 1 st to 26 th week of 2018 was predicted and compared with the actual values to verify the ARIMAX model.Results From the 1 st week of 2014 to the 26 th week of 2018,a total of 60419 cases of ILI were reported by the four sentinel hospitals of Yuhang District,with ILI%of 1.29%.Lasso regression analysis showed that there was a positive correlation between weekly average absolute humidity and ILI%(r=27.769),and a negative correlation between weekly average temperature and ILI%(r=-0.117).The ARIMAX(1,0,0)(1,0,0)12 with weekly average temperature and absolute humidity was selected as the optimal model,with the Bayesian information criterion(BIC)value of 81.30 and the mean absolute percentage error(MAPE)value of 15.77%.The MAPE value of the ARIMAX model predicting the ILI%from 1 st to 26 th week of 2018 were 43.75%.Conclusion The ARIMAX model including meteorological factors can be used to predict the prevalence of ILI,but the accuracy needs to be promoted.