Mortality burden from seasonal influenza and 2009 H1N1 pandemic influenza in Beijing, China, 2007‐2013

Background Data about influenza mortality burden in northern China are limited. This study estimated mortality burden in Beijing associated with seasonal influenza from 2007 to 2013 and the 2009 H1N1 pandemic. Methods We estimated influenza‐associated excess mortality by fitting a negative binomial model using weekly mortality data as the outcome of interest with the percent of influenza‐positive samples by type/subtype as predictor variables. Results From 2007 to 2013, an average of 2375 (CI 1002‐8688) deaths was attributed to influenza per season, accounting for 3% of all deaths. Overall, 81% of the deaths attributed to influenza occurred in adults aged ≥65 years, and the influenza‐associated mortality rate in this age group was higher than the rate among those aged <65 years (113.6 [CI 49.5‐397.4] versus 4.4 [CI 1.7‐18.6] per 100 000, P < .05). The mortality rate associated with the 2009 H1N1 pandemic in 2009/2010 was comparable to that of seasonal influenza during the seasonal years (19.9 [CI 10.4‐33.1] vs 17.2 [CI 7.2‐67.5] per 100 000). People aged <65 years represented a greater proportion of all deaths during the influenza A(H1N1)pdm09 pandemic period than during the seasonal epidemics (27.0% vs 17.7%, P < .05). Conclusions Influenza is an important contributor to mortality in Beijing, especially among those aged ≥65 years. These results support current policies to give priority to older adults for seasonal influenza vaccination and help to define the populations at highest risk for death that could be targeted for pandemic influenza vaccination.


| INTRODUCTION
Influenza is a major cause of global morbidity and mortality each year, especially among older adults and those with chronic diseases. 1 Although several statistical models have estimated the mortality burden of influenza, 2 few have focused on mainland China, [3][4][5] and information on mortality burden in northern China is especially limited. Considering the diverse seasonality patterns, 6  The copyright line for this article was changed on 25 January 2018 after original online publication 10 844 laboratory-confirmed cases of influenza A(H1N1)pdm09 and 69 deaths were reported in Beijing. 7 These numbers solely represent laboratory-confirmed influenza deaths and not decedents who might have died because of pandemic influenza but were never tested. 3 A mathematical modeling study estimated that the likely number of influenza A (H1N1)pdm09 infections in Beijing was 1.8 million during the pandemic period. 8 To date, no study has estimated the likely number of influenza A(H1N1)pdm09-associated deaths in Beijing.
Since 2007, the Beijing Municipal Government has provided seasonal influenza vaccination to priority populations, including adults ≥60 years of age and primary and middle school students, free of charge between September and November each year prior to the start of typical influenza virus circulation. Beijing also conducted influenza A(H1N1)pdm09 vaccination campaigns during the pandemic period. 9 These programs were implemented in the absence of local data on influenza-associated mortality burden because there was concern that this novel virus could have substantial burden on the population's health. Influenza mortality burden data allow programs to determine the potential impact of influenza vaccination and inform investments in prevention strategies. In this study, we modeled mortality data obtained from a representative mortality register system, combined with weekly influenza virus surveillance data, to provide estimates of the mortality impact associated with seasonal influenza and the influenza A(H1N1)pdm09 virus during the pandemic period by specific death categories, age groups, and influenza virus type and subtype in Beijing from 2007 to 2013.

| Population data
We obtained annual population data from the National Population Census in Beijing, China. 10 In 2010, the total population of Beijing was nearly 20 million, of which 12-13 million were registered residents of Beijing. The remaining 7-8 million persons were classified as "migrants" who are not registered residents of Beijing. In this study, we used population data of registered populations only to calculate the annual mortality rates. respiratory (ICD-10 codes J00-99) and circulatory diseases (ICD-10 codes I00-99) (respiratory and circulatory: R&C), and all-cause deaths (including both respiratory and circulatory deaths which have been previously been associated with influenza-associated deaths). [2][3][4][5][6] Two age groups were considered: <65 years and ≥65 years.

| Influenza virology data
We obtained influenza virology data from the Beijing influenza surveil-

| Statistical model
A negative binomial model was applied to estimate influenza-associated excess mortality, using age-specific weekly mortality data as the outcome and proportions of specimens positive for influenza by type/subtype (not age-specific) as the predictor variables. [3][4][5] Models included virology surveillance time series data, as well as time terms to account for the weeks in the model and harmonic terms to account for the cyclical pattern in deaths. Because of the lack of respiratory syncytial virus virology data in Beijing, we did not include this pathogen in the model.
Negative Binomial models were fit separately for the two outcomes (all-cause and respiratory and circulatory deaths) and age groups (aged <65 years and ≥65 years) using a log link function. The final models were selected by evaluating the AIC/BIC values in combination with the statistical significance of the viral surveillance terms and the harmonic and time terms (for baseline estimation) for each of the models examined (Table S1). Further, we decided to choose the same model for each age group and outcome based on the evaluation of the separate models. We chose to use the same model for each age group for ease of interpretation and better comparison of estimates and methods. The final model selected for both all-cause and respiratory and circulatory deaths was as follows: Y t is the weekly variable for the time series from 2007/2008 to 2012/2013 (1-312) and represents age-specific weekly number of deaths, α is the offset term equal to the log of the age-specific population size, β 0 represents the intercept, β 1 and β 2 account for the linear and quadratic time trends, β 3 and β 4 account for cyclical secular trends, and β 5 through β 8 represent coefficients associated with the weekly proportion of specimens testing positive for each influenza subtype.
Further explanation of our model selection is available in Table S1.
The number of deaths attributed to influenza was calculated as the difference between the predicted influenza-associated deaths from the full model and the predicted baseline deaths (from the model when the co-variables for influenza subtype were set to zero).
Confidence intervals were generated for excess mortality estimates based on the confidence intervals of the model predictions. Agespecific excess mortality rates were calculated from the estimated number of deaths attributed to influenza divided by population from the National Population Census in Beijing, China. 10 All the statistical analyses were carried out using SAS University Edition (SAS Institute Inc., Cary, NC, USA). A P value < .05 was considered to be statistically significant.

| Ethics approval
Study approval was obtained from the Institutional Review Board and Human Research Ethics Committee of BJCDC.

| Mortality data reported from the Mortality Register and Surveillance System and influenza virus activity, from season 2007/2008 to 2012/2013
For the influenza seasons 2007/2008 to 2012/2013, the annual allcause mortality rate ranged from 581.7 to 602.6 per 100 000 persons for all ages, and 3238.9 to 3394.3 per 100 000 for adults aged ≥65 years. Adults aged ≥65 years accounted for 76% of all-cause deaths among all ages. R&C causes represented 58% of all deaths. The annual R&C mortality rate ranged from 335.8 to 346.9 per 100 000 persons for all ages and 2050.0 to 2221.0 per 100 000 for adults aged ≥65 years. Adults aged ≥65 years accounted for 84% of R&C deaths among all ages (Table 1).
During the 6-year study period, a total of 60 795 specimens were     Table 5 shows comparisons of mortality burden between seasonal epidemics and the influenza A(H1N1)pdm09 pandemic period from this  1 14 27 40 1 14 27 40 1 14 27 40 1 14 27 40 1 14 27 40 1    Our findings for the all-age mortality burden of seasonal influenza are comparable with previous studies derived from similar approaches in China (Table 5) [3][4][5] and other areas, such as Hong Kong, 14 Singapore, 16 USA, 18 and Mexico. 20   One study estimated that the vaccine coverage in the Chinese population overall was only 2% in 2009. 27 Second, as the largest proportion of influenza-associated deaths among older adults is among those aged ≥75 years, 18 different age distributions across regions may contribute to differences in influenza-associated mortality burden. Beijing is a large city with a very high population aging rate that has reached the average level of developed countries. 28 A higher proportion of adults aged ≥75 years in Beijing may contribute to a higher mortality burden among all adults aged ≥65 years. However, as Beijing is also one of the most developed regions in China, on average, its residents have higher socioeconomic status and better access to health care than residents in other Chinese regions which, in turn, may contribute to a lower mortality burden among older adults.

| Comparisons of mortality burden between seasonal epidemics and the influenza A(H1N1)pdm09 pandemic period
We estimated that the all-cause mortality rate associated with influenza A(H1N1)pdm09 was comparable to the mortality impact of seasonal epidemics during the non-pandemic seasons (19.9 vs 17.2 per 100 000). Although the influenza A(H1N1)pdm09 mortality rate in this study is similar to that estimated by another study in China,4 it is lower than the rate described in Mexico, 20 and higher than rates estimated in the United States. 29 The case fatality rate of pH1N1 infection during the pandemic was higher in Mexico than in Beijing Consistent with studies conducted in other countries, our study found that, compared with seasonal epidemics, there was an "age shift" in age-specific mortality burden toward persons <65 years of age during the influenza A(H1N1)pdm09 pandemic period. 4,20,29,[33][34][35][36] A previous meta-analysis from 27 studies showed that the influenza A(H1N1)pdm09 cumulative incidence varied significantly by age with much higher incidence in younger people. 37 In addition, a study comparing seasonal influenza and influenza A(H1N1)pdm09 demonstrated that younger people had 2-4 times the risk of severe outcomes from influenza A(H1N1)pdm09 than persons of the same ages with seasonal influenza. 38  significantly reduces the chance of overestimation. Prior to including these terms, we evaluated the surveillance data to ensure that it was robust and did not show instances of bias due to few specimens being collected. We also compared our viral surveillance data to the national system which showed a similar pattern in influenza virus circulation for the northern region of China. Third, we did not have the sample size available to obtain reliable estimates within more refined age categories. There were also several factors that may contribute to deaths in Beijing that we were unable to include in our model, such as air quality, which is particularly poor in Beijing especially when compared to cities outside of China. Finally, our estimates were based on an ecological study design, and therefore, we cannot confirm a causal relationship between influenza activity and our estimates of mortality burden.

| CONCLUSIONS
Our study demonstrates a substantial influenza-related mortal-