Joinpoint模型与ARIMA模型在细菌性痢疾发病趋势分析的应用比较

Comparation between Joinpoint model and ARIMA model in bacillary dysentery incidence trend analysis

目的比较Joinpoint模型与ARIMA模型在深圳市盐田区细菌性痢疾(以下简称菌痢)发病趋势分析及预测中的效果。方法对1998—2017年盐田区菌痢发病率进行Joinpoint模型与ARIMA模型分析,同时比较两模型拟合结果的一致性、优劣性、稳健性。结果Joinpoint模型结果显示,2013年为趋势变化“折点”,将菌痢流行曲线分为2段。1998—2013年变化百分比(APC)为-2.0%(P>0.05),2013—2017年APC为-31.5%(P<0.05)。ARIMA最优拟合模型为ARIMA(1,1,0)。两模型拟合结果的相关系数ICC为0.74(95%CI:0.45-0.89)。Joinpoint模型拟合结果的平均绝对误差、平均绝对百分误差分别为9.26%,17.61%;ARIMA模型分别为11.31%,24.86%。离群值对Join-point模型无影响,对ARIMA影响较大,整体趋势表现为先上升后下降。结论两模型对菌痢发病数据拟合结果的一致性较好,Joinpoint模型比ARIMA(1,1,0)模型显示出更好的拟合精度。小样本情况下,Joinpoint模型受离群值影响小,方法更稳健,并能对长期趋势进行分段描述,有较高实际价值。

Objective Compared the effectiveness between Joinpoint model and ARIMA model on the incidence trend anlayses and the prediction of disease of bacillary dysentery in Yantian district of Shenzhen. Methods Joinpoint Regression model and ARIMA(auto-regressive integrated moving average)model were used to analyze the incidence rate of bacillary dysentery in Yantian district of Shenzhen from 1998-2017,and its goodness of model fit were evaluated.Results “Year 2013” was a statistically significant“joinpoint”,which cut the epidemic curve of bacillary dysentery from 1998-2017 into two line segments.The APC(annual percent change)for 1998-2013 was -2.0%(P>0.05),and -31.5% for 2013-2017 (P<0.05).The opitimal ARIMA model for bacillary dysentery in Yantian district was ARIMA(1,1,0).The interclass correlation coefficient was 0.74 (P<0.01),indicating significantly good agreement between two models.The MAE(mean absolute error)and MAPE (mean absolute percentage error)obtained by Joinpoint model were 9.26% and 17.61%;11.31% and 24.86% by ARIMA model.Outliers had no impact on Joinpoint model,but they had significant impact on ARIMA model. The overall trend of ARIMA model was ascending and then descending.Conclusion Joinpoint model and ARIMA model both showed good model fitness.However,the fitting precision of Joinpoint model had higher accuracy.In the case of small sample size,Joinpoint model is less affected by the outliers that provides a more robust approach compared to ARIMA model.Moreover, Joinpoint model can describe the long-term trend in segments,which has an outstanding practical value.