Comparing the similarity and difference of three influenza surveillance systems in China

Three main surveillance systems (laboratory-confirmed, influenza-like illness (ILI) and nationwide Notifiable Infectious Diseases Reporting Information System (NIDRIS)) have been used for influenza surveillance in China. However, it is unclear which surveillance system is more reliable in developing influenza early warning system based on surveillance data. This study aims to evaluate the similarity and difference of the three surveillance systems and provide practical knowledge for improving the effectiveness of influenza surveillance. Weekly influenza data for the three systems were obtained from March 2010 to February 2015. Spearman correlation and time series seasonal decomposition were used to assess the relationship between the three surveillance systems and to explore seasonal patterns and characteristics of influenza epidemics in Gansu, China. Our results showed influenza epidemics appeared a single-peak around January in all three surveillance systems. Time series seasonal decomposition analysis demonstrated a similar seasonal pattern in the three systems, while long-term trends were observed to be different. Our research suggested that a combination of the NIDRIS together with ILI and laboratory-confirmed surveillance is an informative, comprehensive way to monitor influenza transmission in Gansu, China. These results will provide a useful information for developing influenza early warning systems based on influenza surveillance data.

Influenza is associated with considerable mortality and morbidity worldwide. Influenza surveillance systems are important tools for monitoring and evaluating transmission of influenza [1][2][3][4][5] . Gansu is relatively poor province, which public health capacity is backward compared with the economically developed provinces. The influenza incidence rates in Gansu were similar to the national average level. Influenza epidemics in Gansu exhibit strong seasonal winter peaks. In 2009, an extended influenza surveillance system for influenza-like illness (ILI) and virologic data was established in Gansu Province 5 . The extended influenza system aimed to better understand the epidemiologic trend of influenza epidemic and virus variation, and eventually develop an influenza early warning system that could provide advantage for improving the timeliness of epidemic control and formulating scientific prevention strategy. Currently there are three main systems used for influenza surveillance in Gansu Province, China: laboratory-confirmed (i.e., detection of influenza virus nucleic acid), ILI and nationwide Notifiable Infectious Diseases Reporting Information System (NIDRIS). This study aims to evaluate and compare effectiveness of the three influenza surveillance systems and provide practical knowledge for improving the performance of influenza surveillance in Gansu, China.

Results
Patterns of seasonality. There was a distinct seasonality in influenza occurrence in Gansu Province (Tables 1-3, Fig. 1). Both four series of monitoring data were highest in winter, followed by spring.
Our results showed an influenza epidemic to appear as a typical single-peak, around January, in all three surveillance systems. Time series seasonal decomposition analysis demonstrated a similar seasonal pattern in the three systems (Fig. 2). However, long-term trends were observed to differ between the three systems.
Correlation analysis. Fitted lines with scatter plots were shown for positive correlation between any two sets of data (Fig. 3). Spearman correlation indicated that influenza cases of NIDRIS correlated significantly with ILI (r = 0.741, p < 0.01), ILI% (r = 0.442, p < 0.01), and positive rate (r = 0.486, p < 0.01) ( Table 4). Table 5 summarized the advantages and disadvantages for the three surveillance systems (Table 5).

Discussion
There is a high correlation among the three influenza surveillance systems data, suggesting that all three influenza surveillance systems in Gansu Province reflect the epidemic characteristics of influenza. The three influenza surveillance systems indicated that there were epidemics peaks in winter and similar seasonal pattern. The influenza epidemics typically appeared as a single-peak around January in Gansu; this finding is consistent with other northern provinces in China 1,6 . Seasonal decomposition, however, showed differences in long-term trends of the three systems. The increase in ILI cases may be due to surveillance sensitivity improvement, while the downward trend of Influenza-like illness consultation rate (ILI%) may be due to the increase in total visits to outpatient and/or emergency departments. At the same time, the increase in monitoring sensitivity, and the increase in laboratory-confirmed influenza cases resulted in an increase in reported cases of influenza from NIDRIS. Influenza virus positive rate, the percentage of swabs that are positive for influenza by real-time Polymerase Chain Reaction (PCR), can provide accurate information about influenza virus activity, while its monitoring results indicated that influenza epidemic peak didn't occur in the very same time (Fig. 2c), so it's very necessary to carry out continuous monitoring of influenza and set up a suitable early warning technology [7][8][9][10] .
Each of the three surveillance systems has its own advantages and disadvantages. Laboratory surveillance can provide more accurate information about the activity of influenza virus serotypes, so influenza virus positive rate could be used as an indicator for influenza epidemic; it is, however, time consuming and resource intensive and thus may delay the early warning and control of influenza epidemics 11 . By contrast, ILI and ILI% are better indicators for use in early warning systems as they tend to be more time efficient in the generation of data. Sentinel systems, however, rely upon individual doctors or clinics to collate data and report it to the relevant authority, this may introduce compliance or bias issues. Furthermore, illnesses other than influenza may present with ILI 12 ; sentinel hospitals report ILI by age group, without case information; increases in ILI and ILI% may be the result of increased healthcare-seeking behavior in response to media reports or the circulation of non-influenza respiratory viruses 13,14 . ILI, ILI% and positive rate are based on the sentinel surveillance network which is built on data sourced from urban areas; as such, this system may not detect influenza virus activity in rural areas 2,4,11,15,16 . By contrast, influenza cases of NIDRIS are reported by all medical institutions in Gansu Province, including secondary, tertiary, pediatric hospitals in urban areas, and primary hospitals and village clinics in rural areas. Weekly influenza data from the NIDRIS were observed to have larger fluctuations and during the long vacation (especially Chinese New Year), influenza cases greatly reduced ( Fig. 2d) and data stability were affected because of an obvious decrease of patients' visits. Some influenza patients may be misdiagnosed, because it's hard to differentiate influenza from other respiratory viruses without laboratory testing, especially in the non-epidemic season; in addition, as passive surveillance, influenza cases may also be misreported to NIDRIS. This demonstrated the need to improve mechanisms for sharing information between influenza surveillance systems 4,17 , to enhance capacities. This study was the first attempt to compare performance of all the three influenza surveillance systems in Gansu Province. The results of this study provide practical knowledge for developing early warning systems for influenza based on surveillance system with climate and socio-environmental data 18 .
As the influenza surveillance systems were extended in 2009, only five-year data time series were available. This limited analysis of long-term epidemiological characteristics 19,20 .

Conclusion
A combination of the NIDRIS together with ILI and laboratory-confirmed surveillance is an informative, comprehensive way to monitor influenza transmission in Gansu, China. We should consider combining the three surveillance systems in developing influenza early warning systems.   Weekly ILI data were obtained from the national sentinel hospital-based influenza surveillance network. This network consists of 19 sentinel hospitals in Gansu, dispersed through 18 counties and districts, including all 14 cities of Gansu. These 19 sentinel hospitals are the larger comprehensive hospitals, located in the densely populated districts of these cities, and are recognized by the National Ministry of Health to represent the medical-seeking people in Gansu. Each week, participating doctors in the outpatient and emergency departments of internal medicine and pediatrics were required to report the number of patients with ILI by age group (i.e., 0-4 years, 5-14 years, 15-24 years, 25-59 years, and >60 years), and total visits to outpatient and/or emergency departments to the Chinese Influenza Surveillance Information System, which is a centralized online system. Identification of patients with ILI was based on a standard case definition, including body temperature ≥38 °C with either cough or sore throat, in the absence of other laboratory-confirmed evidence 21 . In this study, surveillance data for the period of March 2010 (2010 week 10) to February 2015 (2015 Week 9) was used. ILI consultation rate (ILI%) = ILI/total visits to outpatient and emergency departments * 100%.
Laboratory-confirmed influenza cases were collected for same period from the Chinese Influenza Surveillance Information System. In each sentinel hospital, nasopharyngeal swabs were collected from patients who presented within 3 days of ILI onset. The specimens were transported in viral transport media at 4 °C to the diagnostic laboratories within 48 h of collection. Analysis for influenza virus nucleic acid was performed by real-time Polymerase Chain Reaction (PCR) within 1 week. Taking swabs is a mandatory practice for all sentinel hospitals. Each hospital was required to provide 5 to 15 specimens per month from April to September and 10 to 15 specimens per week from October to next March 21 . Participating doctors chose the patients with typical ILI symptom to collect their swabs, and ensure that the specimen was evenly distributed every week or month.
Influenza is a notifiable infectious disease in China 22 . Case definitions for influenza and diagnostic criteria are outlined by the National Health and Family Planning Commission of the People's Republic of China, which includes clinically diagnosed cases and laboratory-confirmed cases. Notifications of cases are made to NIDRIS, by all medical institutions in Gansu Province, including secondary, tertiary and pediatric hospitals in urban areas, primary hospitals and village clinics in rural areas. Weekly data for reported cases of influenza were obtained from NIDRIS for the period of March 2010 (2010 week 10) to February 2015 (2015 Week 9).
The three systems were used to validate the influenza surveillance quality each other, especially in epidemic period.
Four sets of monitoring data were obtained from the three influenza surveillance systems: ILI, ILI%, positive rate of nucleic acid detection and influenza cases of NIDRIS. The sets of ILI and ILI% were taken from influenza-like illness surveillance of Chinese Influenza Surveillance Information Systems, the set of positive rate came from the laboratory testing results of Chinese Influenza Surveillance Information Systems, while the set of influenza cases was obtained from reported influenza numbers of NIDRIS, including all the clinically diagnosed cases and laboratory-confirmed cases. All four series data were collected continuously from March 2010 (2010 week 10) to February 2015 (2015 week 9). Each series of data was up to five years.
Spearman correlation was used to assess the relationship between the three surveillance systems, as the data appeared to have non-normally distributed pattern. A seasonal decomposition analysis was conducted to assess whether there was a distinct seasonality in each series. In this analysis, each time series was decomposed into seasonality, long-term trend, and irregular factors. The long-term trend showed progression of the influenza series (secular variation). A trend exists when there was a persistent increasing or decreasing direction in the surveillance data 23 .   We confirm that all methods were performed in accordance with the relevant guidelines and regulations by including a statement in the methods section.
Ethics approval and consent to participate. The study was approved by the Research Ethics Committee of Gansu Provincial Center for Disease Control and Prevention. Informed consent is not relevant to the study as this study used aggregated data without personal information.
Availability of data and material. Please contact author for data requests.

ILI
Sentinel hospitals ILI is a better indicator in developing early warning system because it be reported more timely.
ILI-like symptoms may be caused by etiologies other than influenza. Sentinel hospitals report the number of ILI by age group, without case information. Increase in ILI and ILI% might be a result of increased healthcare-seeking behavior after media reports of influenza outbreak or the circulation of non-influenza respiratory viruses.
19 sentinel hospitals are located in urban areas, so the surveillance system may not detect influenza virus activity in rural areas. ILI% Sentinel hospitals ILI% can eliminate the impact of increased number of sentinel hospital.
Long-term trend or short-term changes of the total number of outpatients may impact ILI%, such as "long-holiday effect".
Positive rate Sentinel hospitals Laboratory surveillance can provide more accurate information about influenza virus activity.
Laboratory test sometimes consumes too much time and thus may delay the early warning of influenza epidemics.
Influenza cases of NIDRIS All hospitals Influenza cases are reported by all medical institutions in Gansu Province, include secondary, tertiary, pediatric hospitals in urban area, and primary hospitals, village clinics in rural area.
Some influenza patients may be miss-diagnosed, because it's hard to differentiate influenza from other respiratory viruses infection without laboratory testing, especially in the non-epidemic season. As passive surveillance, influenza cases may also be missing-reported.