Dynamic surveillance for influenza like illness in northern China after pandemic of influenza A (H1N1) in 2009

Objective To discuss the dynamic warning model for influenza like illness (ILI) in northern China after the pandemic of influenza A (H1N1) in 2009. Methods By using software Eviews 6.0, seasonal autoregressive integrated moving average (ARIMA) model was established based on ILI sentinel surveillance data in northern China during 2010-2014, then the optimum time-internal for prediction was selected. Control chart was used to establish a warning model for ILI in northern China. After calculating sensitivity, specificity and describing receiver-operating characteristic curve (ROC), the optimal alert threshold was selected. The dynamic warning can be achieved by combining these two models. Results We established the multiple seasonal ARIMA (1, 0, 0)(1, 1, 0)52, the R2 value of the model fitting degree was 0.65. Among these different time-interval patterns, we found 2-week internal had a balance between the effectiveness and the timeliness, the root mean square prediction error was 0.37, the mean absolute error was 0.24, and the mean relative error percentage was 8.26%. Selecting P95 as the alert threshold line, the sensitivity was 100% and the specificity was 96%. By using 2-week time-interval pattern to alert ILI% of 1-32 week in northern China in 2015, the predicting result was consistent with actual data. Conclusion We established an early warning model by combining the ARIMA model with control chart, which would reflected the epidemiological trend of ILI cases in northern China and support the early detection and control of influenza outbreak.