Geographic divisions and modeling of virological data on seasonal influenza in the Chinese mainland during the 2006-2009 monitoring years

Abstract

Background: Seasonal influenza epidemics occur annually with bimodality in southern China and unimodality in northern China. Regional differences exist in surveillance data collected by the National Influenza Surveillance Network of the Chinese mainland. Qualitative and quantitative analyses on the spatiotemporal rules of the influenza virus's activities are needed to lay the foundation for the surveillance, prevention and control of seasonal influenza.

Methods: The peak performance analysis and Fourier harmonic extraction methods were used to explore the spatiotemporal characteristics of the seasonal influenza virus activity and to obtain geographic divisions. In the first method, the concept of quality control was introduced and robust estimators were chosen to make the results more convincing. The dominant Fourier harmonics of the provincial time series were extracted in the second method, and the VARiable CLUSter (VARCLUS) procedure was used to variably cluster the extracted results. On the basis of the above geographic division results, three typical districts were selected and corresponding sinusoidal models were applied to fit the time series of the virological data.

Results: The predominant virus during every peak is visible from the bar charts of the virological data. The results of the two methods that were used to obtain the geographic divisions have some consistencies with each other and with the virus activity mechanism. Quantitative models were established for three typical districts: the south1 district, including Guangdong, Guangxi, Jiangxi and Fujian; the south2 district, including Hunan, Hubei, Shanghai, Jiangsu and Zhejiang; and the north district, including the 14 northern provinces except Qinghai. The sinusoidal fitting models showed that the south1 district had strong annual periodicity with strong winter peaks and weak summer peaks. The south2 district had strong semi-annual periodicity with similarly strong summer and winter peaks, and the north district had strong annual periodicity with only winter peaks.

Figure 1. Control chart for the weekly total number of detected viruses (Beijing).

It's an example to illustrate the peak performance analysis method. The horizontal ordinate denotes the corresponding number of the 159 monitoring weeks in chronological order. As an example, the 13th week in 2006 is numbered as 1. The black line is the time series of virological data in Beijing. The green lines are the annual medians, and the red lines are the annual 2 control limits, that is, the thresholds. More information about the results of this method is given in Table S1, S2 in detail.

Figure 2. The Fourier harmonic extraction results of Hunan province.

It's an example to illustrate the Fourier harmonic extraction method. The horizontal ordinate denotes the corresponding number of the 159 monitoring weeks in chronological order. The black line is the time series of raw data. The red line is the data after the replacement procedure and the green line is the superposition results of the extracted Fourier harmonics. More information about the results of this method is given in Table S3 in detail.

Figure 3. Bar chart of the influenza virus subtypes in the southern area.

Different colors represents different influenza subtypes as is listed in the head.

Figure 4. Bar chart of the influenza virus subtypes in the northern area.

Different colors represents different influenza subtypes as is listed in the head.

Figure 5. Geographic division results of Mainland China based on the peak performance analysis method.

The provinces with the same color belong to the same district.

Figure 6. Geographic division results of Mainland China based on the Fourier harmonic extraction method.

The provinces with the same color belong to the same district.

Figure 7. The comparison between raw data and sinusoidal model fit results for the south1 district.

The black line is the time series of raw data, while the red line is the sinusoidal fitting curves.

Figure 8. The comparison between raw data and sinusoidal model fit results for the south2 district.

The black line is the time series of raw data, while the red line is the sinusoidal fitting curves.

Figure 9. The comparison between raw data and sinusoidal model fit results for north district.

The black line is the time series of raw data, while the red line is the sinusoidal fitting curves.