中国内地法定报告传染病预测和监测的ARIMA模型

Using ARIMA model to surveillance and forecast the incidence rate of notifiable infectious diseases in China's Mainland

目的通过对1995年1月～2004年4月中国大陆法定报告传染病逐月发病率数据的分析,研究其变化规律,建立预测与监测的ARIMA时间序列模型。方法利用时间序列模型中的自回归滑动平均混合模型ARIMA,考虑非季节效应和季节效应,分析中国法定报告传染病发病率的变化趋势和周期性,模型参数估计采用非线性最小二乘法,应用残差和赤池信息量准则(AIC)评价模型的优劣。1995～2004年我国内地法定报告传染病逐月发病率的数据用于建立模型,2005年1月～2006年4相应数据用于模型检验。结果分析结果显示,法定报告传染病发病以年为周期,一年中6～9月为高发月,尤其是8月和7月最为严重。ARIMA(0,1,0)(0,1,0)12模型是法定报告传染病拟合的最佳模型,其拟合残差的方差为2.28,外推预测的平均绝对误差为0.34。利用预测值的95%置信区间建立了我国内地法定报告传染病发病率变化的监测控制线,用于其发病情况的预测与预报。结论对传染病发病率历史数据进行时间序列分析是用于传染病监测的一个重要的工具。所建立的ARIMA模型适用于对中国大陆法定报告传染病发病率预测与监测。该模型具有一定的实用价值,并可以应用于其他传染病的监测和异常变化的检测。

Objective To develop the model for forecasting and surveilling the spreading of notifiable infectious diseases in China's Mainland. Methods Using time-series methods, ARIMA （0, 1, 0） （0, 1, 0）12 model was developed for purpose of forecasting and surveillance of notifiable infectious disease in China's Mainland. The model was based on the reporting data of these diseases in China's Mainland from 1995 to 2004, and it was tested by the data from January, 2005 to April, 2006. Results The changes of incidence rate of the notifiable infectious diseases in China's Mainland presented a yearly periodicity, and showed that the incidence rate from April to September exceeded the monthly average of it. The residuals sum of square of the ARIMA model was 2.28 for incidence rate of the notifiable infectious disease from 1995 to 2004, and the mean error of the model was 0.34, The model could further be used to draw a control chart for surveillance of the disease. Conclusions Time series methods applied to historical reporting data of infectious disease are an important tool for infectious disease surveillance. The ARIMA model is suitable to forecast report incidence rate of notifiable infectious diseases in China's Mainland, Our approach potentially has a high practical value for the notifiable infectious diseases in forecasting and surveillance, and it can be generalized to other diseases to develop automated surveillance system and capable of detecting anomalies in disease pattern.