曲线估计在流行性感冒发病预测中的应用

Application of curvilinear estimation in predicting incidence of influenza

目的探索曲线估计在流行性感冒发病预测中的应用并对预测效果进行评价,为该病的防治提供理论依据。方法采用移动平均法对2010—2014年安徽省流行性感冒分月发病数建立基础数据库,运用曲线估计对分月发病数开展模型的拟合并优选最佳模型,通过2015年实际发病数与模型预测结果比对进行拟合效果评价。结果 2010—2015年安徽省各级医疗、疾控机构报告流行性感冒病例38 523例,年发病率为4.35/10万~18.58/10万,各年发病率差异有统计学意义(P<0.01)。病例报告集中于12月和次年的1—3月,报告病例数占全部病例数的47.14%(18 160/38 523);其中8月份占8.58%(3 306/38 523),存在夏季病例报告小高峰。模型拟合结果显示4—7和9月适用三次方曲线模型;1、8和10月适用二次方曲线模型;11和12月适用线性模型;2月适用S型曲线模型;3月适用乘幂次方曲线模型。模型预测结果显示2015年流行性感冒病例数为11 938例,实际发病数为11 300例,预测误差率为5.34%。结论曲线估计作为流行性感冒发病预测方法可行,预测结果较为可靠。

Objective To explore the application of curvilinear estimation in the prediction of influenza incidence and evaluate the forecast effect, so as to provide a theoretical basis for prevention of influenza. Methods Moving average method was applied to establish a database of the monthly incidence of influenza from 2010 to 2014 in Anhui Province, curve estimation was used to conduct the model fitting and merge the best model on data of cases in 2015 to assess the fitting effect. Results Total of 38 523 influenza eases were reported by medical institutions and centers for disease control and prevention at all levels in Anhui Province from 2010 to 2015, with annual incidence rates ranging from 4.35 to 18.58 per 100 000. The differences of annual incidence rates were statistically significant （P 〈 0.01 ）. The most cazes were concentrated in December and January to March, accounting for 47.14% （ 18 160/38 523 ） of the total cases. A small peak in incidence of influenza was observed in August, accounting for 8.58% （3 306/38 523 ）of the total eases. Model fitting result showed that the cases in April to July, and September fitted for cubic curve model; January, August, and October fitted for quadratic curve model; November and December fitted for linear curve model; February fitted for S curve model; March fitted for power curve model. Model prediction showed that there were 11 938 cases of influenza in 2015, and the actual reported eases were 11 300, with a prediction error rate of 5.34%. Conclusion Curvilinear estimation can be used as a relatively reliableapproach for predicting the incidence of influenza.