Influenza illness averted by influenza vaccination among school year children in Beijing, 2013‐2016

Background The benefit of school‐based influenza vaccination policy has not been fully addressed in Beijing. Objectives To evaluate the benefit of school‐based influenza vaccination policy launched in Beijing. Methods Using existing surveillance and immunization data, we developed a dynamic transmission model to assess the impact of influenza vaccination in school‐going children. The outcome was defined as the averted number of medically attended influenza illnesses and the prevented disease fraction to all children aged 5‐14 years for the 2013/14, 2014/15, and 2015/16 seasons. Results We estimated that during the three consecutive influenza seasons, the averted number of medically attended influenza illnesses among children aged 5‐14 years was around 104 000 (95% CI: 101 000‐106 000), 23 000 (95% CI: 22 000‐23 000), and 21 000 (95% CI: 21 000‐22 000), respectively. Corresponding prevented fractions to all children aged 5‐14 years were 76.3%, 38.5%, and 43.9%. Conclusions In Beijing, school‐based vaccinations reduced a substantial number of medically attended influenza illnesses despite seasonal variation in the prevented fraction. This is strong supportive evidence for the continuation of school‐based vaccination programs to reduce the influenza burden in this age group.

rare globally. 3,4 In Beijing, in 2009, a free influenza vaccination policy was launched in primary, secondary, and high schoolchildren who are 6-18 years old. 5 Temporary points of vaccination (POVs) are established at schools, and students with their parent's written consent are organized to receive the vaccination there. Moreover, students are able to receive the free vaccination at community POVs located if the temporary school POVs are closed. Despite 8 years of a funded school-based influenza vaccine policy, there has not been any research evaluating the impact of this vaccination policy.
In recent years, several studies assessed vaccine effectiveness (VE) in Beijing residents to demonstrate the benefit of influenza vaccination. Moreover, to understand the public health benefit of school-based vaccination policy, a long-term ecological overview study based on school outbreak data was conducted. The findings indicated that the reduction in school outbreaks by vaccination was significantly dependent on vaccination coverage (VC) and VE. 6 However, the finding was not quantitative. Of the existing data on

| Methods overview
In this study, the target outcome is the reduced number of influenzaassociated medically attended illnesses among schoolchildren aged 5-14 years. The target age group was 6-18 years; however, we were only able to obtain data on population size and medical attendances in the age group of 5-14 years. The outcome was estimated in several steps. First, we estimated the number of influenza infections in Beijing among children aged 5-14 years using surveillance data.
Then, the number of influenza infections was used in a susceptibleexposed-infectious-removed (SEIR) model to simulate transmission dynamic processes accounting for VC and VE. During this process, a series of parameters of the SEIR model were estimated. Next, the parameters were again used in the SEIR model to simulate the number of influenza infections without vaccination. The difference between the estimated numbers of influenza infections with vs without vaccination equals the influenza infections averted by vaccination. Lastly, the averted number of influenza-associated medically attended illnesses was calculated.

| Number of influenza infections
To determine the number of influenza illnesses averted, we needed an estimate of the total number of influenza infections in the Beijing population in children aged 5-14 years in each season. To do this, we progressively scaled up the number of infections detected by influenza-like-illness (ILI) surveillance, accounting for underreporting in ILI surveillance, as shown in the published study. 7 The number of infections detected by ILI surveillance was calculated using the number of ILI cases reported by hospitals, multiplied by the proportion of ILI cases that were positive for influenza in each week. ILI surveillance is conducted in outpatient and emergency departments in all 421 general hospitals in Beijing, reporting for weekly number of ILI cases by age groups. ILI is defined as a person presenting with fever (temperature >38°C) and cough or sore throat in the absence of other diagnosis. The ILI surveillance is enhanced through the collection of virological data from 24 sentinel hospitals, where the age group specific weekly rates of influenza among ILI cases were acquired. In this study, influenza-associated medically attended illness is defined as the number of infections detected by ILI surveillance.
We scaled up the number of influenza-associated medically attended illnesses among children aged 5-14 years to estimate the number of influenza infections, in reference to the published methods and parameters. 7 It is estimated that every influenza-associated medically attended illness represented 9.08 (95% CI: 7.21-12.21) influenza infections in the Beijing population aged 5-14 years old.

| Influenza VE
We reviewed observational studies with VEs in the previous three flu seasons by Beijing CDC [8][9][10][11] and found that the VE against medically attended illness for children aged 5-14 years was not always reported. Instead, we used the VE estimates that are either from the outpatients or subjects closest in age. Therefore, the VE values in the 2015-16 season; 8,11 therefore, we used a low effectiveness of 20% instead. As estimates of VE varied across studies for the same season, sensitivity analyses were performed to reflect the uncertainty of the assumptions, where the VE was assumed to be 10% lower and 10% greater, respectively, on the base of original ones in all seasons.

| Structures of SEIR model
The SEIR model can provide a basic description of the transmission dynamics of influenza using a simple parameterized set of ordinary differential equations. The total number of individuals in the popula-

| Averted number of influenza infections and averted number of influenza-associated medically attended illness
We first simulated the curve of the number of influenza infections with vaccination by fitting the model (formula 6-9) to initial data, and parameters including σ, γ, β 1 , β 2 , β 3 , T 1 , and T 2 were solved using Runge-Kutta methods during this process. We ran the fitting program for 100 times, and 95% confidence interval of each parameter was obtained. The estimated number of influenza infections with vaccination and its 95% CI was then obtained by adopting the parameters into the SEIR model. Afterward, we estimated the number of influenza infections for a scenario without vaccination by the SEIR model (formula 6-9, VE = 0). Averted number of influenza-associated medically attended illnesses was defined as the averted number of influenza infections by vaccination divided by 9.08 (see 2.2.1). The prevented fraction was defined as the number of averted influenza infections divided by the total influenza infections that would have been expected in an unvaccinated population.  (Table 1). The estimated model parameters from epidemiological data by subtype, season, and epidemic period are shown in Table S1.

| Averted number of influenza-associated medically attended illnesses and prevented fraction
The weekly number of persons vaccinated and the number of estimated influenza infections among children aged 5-14 years (1)  (Table 2).
We estimated the number of infections averted by direct and indirect vaccine effects (

| Sensitivity analysis
To reflect the uncertainty of this result, we performed a sensitivity analysis where VE was assumed to be 10% lower and 10% greater in all seasons, while VC was unchanged. Results of this sensitivity analysis are shown in Table 3. It is shown that the benefit of vaccination increased with improving VE. The overall averted number of influenza-associated medical visits were around 81 000, 12 000, and 9000 in three seasons, respectively, when VE decreased by 10%; and 137 000, 38 000, and 35 000 in three seasons, respectively, when VE increased by 10%.
We performed a sensitivity analysis using different influenza infection estimations to evaluate the impact of the number of influenza infections on the estimation of the prevented fraction among all children aged 5-14 years. The results showed that the higher estimation of influenza infections resulted in a lower prevented fraction, and vice versa, but the impact was small (Table 3). be prevented. 13 Our results suggested that vaccination can effectively prevent influenza illness among schoolchildren, even with mismatched vaccine. 10 Schoolchildren is a group that is easy to reach, and feasible to vaccinate, when compared with other vulnerable populations. Given these features, the current policy that encourages influenza vaccine uptake among schoolchildren by providing free vaccination should be supported.

| D ISCUSS I ON
Our study suggested that VE has a strong impact on vaccination outcome, and a small improvement of VE will result in a significant reduction in cases. The timing of vaccination relative to disease occurrence also affects the benefit of vaccination. 14  In our study, the VC was similar over years, so the impact of VC was not evaluated. In the United States, CDC estimated the averted number of influenza-associated outcomes by vaccination in consecutive seasons by age group, and the results suggested that a greater fraction of disease was prevented as greater fractions of the population were vaccinated. 14 It is estimated that a 20% increase in VC among schoolchildren corresponds to an 8% decrease in emergency department visits. 15 More effort is therefore needed to improve coverage among schoolchildren in the future.
Our study has some limitations. Firstly, we only assess the im- other unvaccinated individuals than with vaccinated individuals, and this may counteract herd immunity to some extent. 17 If the contact patterns in China are similar to those in the United States and this effect is present, our results may somewhat overestimate the impact of vaccination. Lastly, preexisting immunity in school-age children was not considered in this study due to a lack of reference data.
We used a dynamic transmission model to estimate the gains of the vaccination program, considering the influenza attack rate over times and the herd immunity effect. Furthermore, the immunization rates were obtained from the registration system rather than a sampled survey. The influenza surveillance data covered all general hospitals in Beijing, improving the representativeness of the database.
Moreover, we estimated the season-, subtype-and stage-specific transmission parameters, considering different transmission periods, allowing a more complete estimate of the value of influenza vaccination in Beijing. We used prevented fraction to show the impact of vaccination policy. The index has been used by researches from CDC, 18 and in a similar modeling study. 14 Unlike absolute numbers of averted outcomes, which are associated with the influenza attack rate, the prevented fraction is mainly affected by VE and VC, making it a good method for evaluating impact of vaccination policy.

| CON CLUS IONS
Vaccination against influenza among schoolchildren has a substantial impact on reductions in medical encounters for influenza in Beijing.
The study indicates that improvement in VE and coverage will lead to greater protection. This approach could be used to estimate the averted number of influenza-associated hospitalizations and death, while our estimation could be used to analyze the cost-effectiveness of the free-of-charge policy.