Burden of influenza‐associated outpatient influenza‐like illness consultations in China, 2006‐2015: A population‐based study

Abstract Background Human influenza virus infections cause a considerable burden of morbidity and mortality worldwide each year. Understanding regional influenza‐associated outpatient burden is crucial for formulating control strategies against influenza viruses. Methods We extracted the national sentinel surveillance data on outpatient visits due to influenza‐like‐illness (ILI) and virological confirmation of sentinel specimens from 30 provinces of China from 2006 to 2015. Generalized additive regression models were fitted to estimate influenza‐associated excess ILI outpatient burden for each individual province, accounting for seasonal baselines and meteorological factors. Results Influenza was associated with an average of 2.5 excess ILI consultations per 1000 person‐years (py) in 30 provinces of China each year from 2006 to 2015. Influenza A(H1N1)pdm09 led to a higher number of influenza‐associated ILI consultations in 2009 across all provinces compared with other years. The excess ILI burden was 4.5 per 1000 py among children aged below 15 years old, substantially higher than that in adults. Conclusions Human influenza viruses caused considerable impact on population morbidity, with a consequent healthcare and economic burden. This study provided the evidence for planning of vaccination programs in China and a framework to estimate burden of influenza‐associated outpatient consultations.


| INTRODUC TI ON
Pandemic and seasonal influenza viruses cause substantial morbidity and mortality worldwide and in China. [1][2][3][4][5] The viruses are associated with a large number of excess deaths, hospitalizations, and outpatient visits as well as absences from work or school resulting from infections. Population-based burden estimation found persons of all ages were associated with substantial burden of outpatient visits and associated absenteeism. In the United States, the Influenza Incidence Surveillance Project reported a national estimate of around 10 per 1000 persons consulted because of influenza A(H1N1) in 2009/10 pandemic season, of which the highest age-specific rates occurred among children aged below 17 years. 6 Currently, population-based estimates of influenza-associated excess deaths and hospitalizations have been increasingly reported, 2,5,7,8 while few studies have been published on influenza-associated outpatient visits which may also bring considerable economic burden to the society. 3,6,[9][10][11] These studies were mostly limited to a single season or region (state/province/city/county), for a specific age group, or reported overall rather than region-specific estimates. 3,9,10 Meanwhile, few evidence on influenza-associated outpatient burden among Chinese population is available. 10,12,13 The challenges in estimating population-level influenza-associated outpatient burden may arise from the difficulty in distinguishing symptoms caused by influenza viruses from those by other respiratory virus (eg, RSV or rhinovirus). Compared with other clinical case definitions, the commonly used definition of influenza-like illness (ILI) is relatively broad to capture influenza-associated illness. Another challenge lies in determining the underlying population covered by surveillance networks. Studies from the United States or UK used the number of population registered with general practitioners as the underlying population; however, such data may not be available in developing countries, for example, China.
Influenza seasonality varies across countries and may exhibit different patterns within countries covering a wide range of latitudes. 14 Understanding regional influenza-associated ILI burden plays a key role in formulating and evaluating national control strategy for seasonal influenza. To determine province-specific and age-specific influenza-associated outpatient burden, and further guide influenza

| Influenza surveillance and virological data
The national sentinel hospital-based ILI surveillance system was initiated by the Ministry of Health in 2000, in line with the recommendations by the World Health Organization. 15  To account for the different sensitivity to influenza detection between RT-PCR and virus culture, we adjusted specimens that were tested positive by RT-PCR using an adjustment ratio based on a previous study. 5 The laboratories used the protocols and test kits 2018ZX10713001-005), the Harvard Center for Communicable Disease Dynamics from the National Institute of General Medical Sciences (grant no. U54 GM088558), and the China-US Collaborative Program on Emerging and Re-emerging Infectious Disease (6 NU2GGH000961-05-02). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of General Medical Sciences or the National Institutes of Health. The funding bodies had no role in study design, data collection and analysis, preparation of the manuscript, or the decision to publish.

Conclusions:
Human influenza viruses caused considerable impact on population morbidity, with a consequent healthcare and economic burden. This study provided the evidence for planning of vaccination programs in China and a framework to estimate burden of influenza-associated outpatient consultations.

K E Y W O R D S
disease burden, influenza, influenza-like illness, surveillance, vaccination released by the Chinese National Influenza Centre, a World Health Organization Collaborating Centre for Reference and Research on Influenza. The national surveillance system has been described elsewhere in detail. 2,14 The same surveillance protocol and ILI definition were used in all sentinel hospitals. As public health surveillance data and routinely collected specimens were used, according to the National Health Commission of China national influenza surveillance guidelines, this project was considered official public health surveillance activities and institutional review board approval was not required. 16 In our study, we retrieved data from the national surveillance system in China from 2006 to 2015. We performed data cleaning and weighting to improve data validity and representativeness (Appendix S1).

| Meteorological, geographic, and socioeconomic data
Daily and city-level temperature and relative humidity from 2006 through 2015 were obtained from the China Meteorological Administration. Province-level meteorological data were aggregated as the mean of city-level data and then averaged into weekly-level. Missing meteorological data (0.07%) were imputed by averaging the weekly data in the same province from other years.
We obtained data on administrative divisions and areas from the State Council, the People's Republic of China, 17 and age-specific annual population size from National Bureau of Statistics of China. 18 Annual number of outpatient (medicine and pediatrics) were obtained for each province from 2007 to 2014. 18 Missing data in 2015 and 2016 (20%) from annual provincial on population and outpatients were imputed by data from other years using linear regressions using year as predictor. Weekly number of the population and outpatients were interpolated from annual data by natural spline method.
To further explore potential socioeconomic factors associated with disease burden, we also collected data for each province on total and urban area in 2015, per capita gross regional product (per-

| Statistical methods
We assumed the proportion of ILI consultation among medicine and pediatric outpatient consultations in the surveillance hospitals is representative of the province. Hence, the weekly total influenza-associated ILI burden was calculated as the product of weekly total influenza-associated ILI consultation proportion (among surveillance departments including medicine and pediatrics) in surveillance hospitals and outpatient consultation proportion in the population.
A wide variety of pathogens may lead to ILI medical consultations.
Here, we adopted a well-established strategy by first linking the observed ILI medical consultations to influenza virus activity in a statistical model, and then, the influenza-associated ILI burden was estimated by quantifying the excess ILI consultations due to influenza virus. 19 Generalized additive regression models were used to estimate weekly influenza-associated ILI burden, accounting for the time-varying baseline and potential non-linear effects. 8,20 We assumed an additive association between influenza-associated ILI burden and influenza virus detection rate, and thus, linear model with an identity link was used. 20 Weekly province-specific virus detection rates of influenza A (H1N1), A(H3N2), A(H1N1pdm) and influenza B were predictors in the model. We also included potential confounding factors such as temperature and absolute humidity in the models, 21 and spline terms to allow for potential non-linear relations. 8 Dummy variable was used to allow for potential instantaneous effect due to changes in reporting and laboratory methods during influenza pandemic period from May 11, 2009, to January 17, 2010. 22 In the sensitivity analysis, we assumed that the above change would result in a long term effect and we also allowed for potential change due to expansion of the surveillance network after the pandemic. Further information on main and sensitivity analysis were described in Appendix S1.
We anticipated influenza-associated ILI burden may vary across provinces. To understand factors behind, we tested several socioeconomic factors which could be associated with population health-seeking behavior using a cluster analysis (further details in Appendix S1). 5 Analyses were conducted using R version 3.3.3.

| Socioeconomic status and healthcare resources
The national influenza surveillance system covered 30 provinces, and the surveillance hospitals provided an average of 3.1% (5.5% in 2009) of all medical consultations in the study period. Highest per capita GRP was observed in Tianjin, Beijing, and Shanghai municipalities, and Jiangsu and Zhejiang provinces (Table S1). Healthcare resources, indicated by number of hospitals, number of hospital beds, and number of outpatient department, varied by provinces (Table S1). We examined correlations between healthcare resources    (Table S1). Mean annual specimen tested and annual influenza-positive rate were also varied by provinces: Beijing, Shanghai, and Tianjin municipalities detected highest influenza-positive rates (Table S1). Figures S3 and S4  The pattern of ILI consultation rate was less clear between provinces but generally consistent with virological activity.   Figure 1). Highest influenzaassociated ILI burden was estimated in Beijing, Tianjin, and Shanghai municipalities, ranging from 6.2 to 8.9 per 1000 py, while lowest burden in Jilin, Ningxia, and Qinghai provinces (Table 1, Figure 1). The sensitivity analysis shows similar estimates in influenza-associated ILI burden (Table S2).  Figure 2).

| Influenza-associated ILI burden
We also estimated the burden ratio of patients aged 0-14 years vs ≥60 years (ratio 1 ) and patients aged 15-59 years vs ≥60 years (ratio 2 ) from 2005 to 2016. We observed higher post-pandemic ratio 1 , suggesting a shift of influenza-attributed outpatient burden to children since 2009 (Table 2).
We estimated the proportion of type/subtype-specific influenza-associated ILI burden by each province and for each year ( Figure S6). Though the pattern was inconsistent between provinces before 2009, the dominant burden was generally attributed to the same influenza type/subtype across 30 provinces after 2009. Most provinces had larger proportion of burden associated with influenza A(H1N1)pdm or A(H3N2) comparing to influenza B ( Figure S6).

| Cluster analysis on between-province variation
We level. We plotted perGRP, perDI, perUHCMS, perRHCMS, and population density against the estimated influenza-associated ILI consultations at each province ( Figure S7). We identified significant correlations (ρ = 0.52-0.88, all P < .05) for all of these variables, indicating moderate to strong positive associations. Further, a Euclidean distance matrix was constructed based on the above factors from 30 provinces and was compared with another distance matrix of the estimated influenza-associated ILI burden from 30 provinces ( Figure   S8). The Mantel correlation statistic was 0.70 (P < .001), implying that perGRP, perDI, perUHCMS, perRHCMS, and population density could explain more than half of the heterogeneity of influenza-associated burden across provinces.

| D ISCUSS I ON
Based on national sentinel hospital-based ILI and virological surveil- Comparing to the United States, China had a lower overall burden estimates but with a larger variation between provinces. 6 The annual overall influenza burden was more than 8  We observed increase of burden of post-pandemic period compared with pre-pandemic. However, we were not able to determine whether this elevation was associated the expansion of surveillance network.

| PUB LI C HE ALTH IMPLI C ATI ON S
Influenza vaccination is the most effective tool in preventing seasonal influenza. This study highlights provinces with higher influenza outpatient burden where influenza vaccination is likely to be more cost-effective. Our findings also suggest the importance of increasing vaccination coverage among children, of whom influenza-associated outpatient burden was the highest but was not included as priority group for most regions. Continued monitoring of influenzaassociated disease burden would be important for further planning and evaluating the cost-effectiveness of influenza vaccination program. [33][34][35] Furthermore, influenza surveillance should also extend to primary care settings including general practitioner clinics or community healthcare centers in the future. This could give a full picture of influenza-associated medical seeking behaviors, outpatient burden, and impact of influenza epidemics in the community, also necessary for pandemic influenza preparedness.
In conclusion, we found substantial influenza-associated outpatient burden in 30 provinces in China across 10-year period. The burden varied by time, increased sharply in 2009 pandemic, and was more pronounced in well-developed provinces and among children, and influenza A(H3N2) contributed to most outpatient burden after the 2009 pandemic. Our study informs the planning of influenza vaccination program in China.

ACK N OWLED G EM ENTS
We thank the staff members at the sentinel hospital network in each province in China for the administrative work and data collection.
We thank colleagues from St. John's College (The University of Hong Kong) for useful discussion.

CO N FLI C T O F I NTE R E S T
BJC has received research funding from Sanofi Pasteur for a study of influenza vaccine effectiveness. The authors report no other potential conflicts of interest.

E TH I C S A PPROVA L A N D CO N S E NT TO PA RTI CI PATE
This project was therefore determined by China CDC not to be subject to institutional review board approval.

DATA AVA I L A B I L I T Y S TAT E M E N T
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.