Exploring the effect of previous inactivated influenza vaccination on seasonal influenza vaccine effectiveness against medically attended influenza: Results of the European I‐MOVE multicentre test‐negative case‐control study, 2011/2012‐2016/2017

Background Results of previous influenza vaccination effects on current season influenza vaccine effectiveness (VE) are inconsistent. Objectives To explore previous influenza vaccination effects on current season VE among population targeted for vaccination. Methods We used 2011/2012 to 2016/2017 I‐MOVE primary care multicentre test‐negative data. For each season, we compared current season adjusted VE (aVE) between individuals vaccinated and unvaccinated in previous season. Using unvaccinated in both seasons as a reference, we then compared aVE between vaccinated in both seasons, current only, and previous only. Results We included 941, 2645 and 959 influenza‐like illness patients positive for influenza A(H1N1)pdm09, A(H3N2) and B, respectively, and 5532 controls. In 2011/2012, 2014/2015 and 2016/2017, A(H3N2) aVE point estimates among those vaccinated in previous season were −68%, −21% and −19%, respectively; among unvaccinated in previous season, these were 33%, 48% and 46%, respectively (aVE not computable for influenza A(H1N1)pdm09 and B). Compared to current season vaccination only, VE for both seasons' vaccination was (i) similar in two of four seasons for A(H3N2) (absolute difference [ad] 6% and 8%); (ii) lower in three of four seasons for influenza A(H1N1)pdm09 (ad 18%, 26% and 29%), in two seasons for influenza A(H3N2) (ad 27% and 39%) and in two of three seasons for influenza B (ad 26% and 37%); (iii) higher in one season for influenza A(H1N1)pdm09 (ad 20%) and influenza B (ad 24%). Conclusions We did not identify any pattern of previous influenza vaccination effect. Prospective cohort studies documenting influenza infections, vaccinations and vaccine types are needed to understand previous influenza vaccinations' effects.


| INTRODUC TI ON
Constant evolution of influenza viruses requires possible reformulation of the influenza vaccine every season. In February each year, the World Health Organization (WHO) organises a technical consultation to decide which influenza strains to be included in the Northern Hemisphere seasonal influenza vaccines. 1 Most groups for whom the seasonal influenza vaccine is recommended may receive a trivalent or quadrivalent inactivated seasonal influenza vaccine annually irrespective of their previous influenza virus infections or influenza vaccination history. In children less than 9 years old, one dose of inactivated influenza vaccine is recommended for children vaccinated in previous season and two doses for those previously unvaccinated. 2 Several observational studies and meta-analyses have reported inconsistent results of the effect of previous vaccination on current season influenza vaccine effectiveness (VE). [3][4][5][6][7][8][9][10][11][12] Some suggest that previous vaccination may reduce the effectiveness of vaccination in the current season. 3,5,6 Various explanations were proposed. The original "antigenic sin" hypothesis suggests that vaccination primarily boosts pre-existing antibody responses that cross-react with the vaccine strain rather than producing a de novo response to the vaccine or infecting strain. 13 The "antibody block" hypothesis suggests that previously vaccinated individuals do not have the crossprotective immunity provided by natural infection. 14,15 According to the "antigenic distance" hypothesis, in seasons when similar strains are included in the subsequent vaccines but are different from the circulating strain, previous vaccination may negatively interfere with current vaccination. 16 Finally, because the influenza vaccine is recommended to individuals with high-risk conditions (eg pregnant women, persons with chronic conditions, older adults aged >59 or >64 years), characteristics of individuals repeatedly vaccinated may result in a poorer observed response to the vaccine if we fail to control for negative confounding. 4 Due to the frequent changes of the genetic and antigenic characteristics of the circulating influenza strains and of those included in the vaccines, data from multiple seasons are needed to measure the potential effects of previous vaccinations and to explore the possible mechanism(s) that may explain such effects. 11 In this article, using the I-MOVE primary care multicentre casecontrol study (MCCS) data, we present influenza type-/subtypespecific VE stratified by previous season vaccination among the target groups for vaccination for each of the 2011/2012 to 2016/2017 seasons. Additionally, using those unvaccinated in both seasons as a reference group, we calculated VE for different combinations of previous/current vaccination among the target population for vaccination (indicator analysis).
Both analyses measure the effect of current vaccination among those not vaccinated in the previous season. The indicator analysis additionally gives information on the potential residual protection of the previous season vaccination and the combined protection of current and previous season vaccination, compared to the reference group. This reference group "unvaccinated in both seasons" may consist of a population that is quite different from the other categories of individuals who have been vaccinated, making controlling for confounding challenging. 11 Therefore, we present also the stratified analysis, where the stratum of those with previous vaccination indicates the extra protection the current vaccination may have, compared to residual effects of previous vaccination, while controlling adequately for previous vaccination history.

| ME THODS
Each season we conducted a primary care-based test-negative design MCCS. The methods were described previously and are based on the same generic study protocol. 17 In summary, practitioners from twelve participating sites interviewed a systematic sample of patients consulting for influenza-like illness (ILI; EU case definition 18 : sudden onset of symptoms and at least one of the following systemic symptoms: fever or feverishness, malaise, headache, myalgia and at least one of the following respiratory symptoms: cough, sore throat, shortness of breath) and collected nasopharyngeal specimens for virological analyses. The data collected include ILI symptoms and date of onset, age, sex, presence of chronic conditions and hospitalisation for the chronic conditions in the previous 12 months.
Vaccination status for current (including date of vaccination and type of vaccine used) and previous season were documented either through patients' self-report or extracted from practitioners' vaccine registries.
We defined patients as vaccinated in the current season if they had received at least one dose of influenza vaccine more than 14 days before symptom onset. All others with information on current vaccination status were defined as unvaccinated. Patients who received at least one dose of influenza vaccine in the previous season were defined as vaccinated in the previous season. For this analysis, we used the population for which influenza vaccination is recommended every season, as they are likely to be a more homogeneous group in terms of vaccination practices than those for whom vaccination is not recommended. They include older adults (aged over 54, 59 or 64 years depending on study site), individuals with chronic conditions and, where available, other groups for whom the vaccine was recommended in a given country (eg pregnant women, healthcare workers and other professional groups, depending on the study site). We excluded children aged less than 9 years as the definition of their current season vaccination status depends on their previous season vaccination.
Cases were ILI patients testing positive for any type-/subtypespecific influenza virus using real-time reverse-transcription PCR (RT-PCR). Controls were those testing negative for all influenza viruses.
Study periods depended upon season-and site-specific influenza virus type/subtype circulation and vaccination campaigns.
We included ILI patients who consulted their practitioner more than 14 days after the start of national or regional seasonal influenza vaccination campaign, who were swabbed less than 8 days after ILI symptom onset and who did not receive influenza antivirals before swabbing. We excluded patients with missing information for current or previous season vaccination status.
We used logistic regression to compute the odds ratio (OR) of being vaccinated in cases and controls. We estimated the type-/subtypeadjusted influenza VE as (1 − OR)*100. Study site was modelled as a fixed effect and always included in the crude and adjusted analysis models. We measured VE carrying out a complete case analysis excluding patients with missing values for any of the variables in the model measuring VE. We included age, sex, presence of chronic conditions, pregnancy and obesity, where applicable, and date of symptom onset as a priori confounding variables in the model. All other potential confounders were included in the model if they changed the VE point estimate by 5% (absolute percentage). Age and onset time were modelled as a restricted cubic spline with knots specified according to Harrel. 19 We measured current season VE for each type/subtype and season. To study the effects of previous vaccination on current season vaccination, we conducted two analyses.
First, we conducted a stratified analysis in which we measured current season VE among individuals vaccinated and unvaccinated in the previous season.
In a second analysis, we used unvaccinated in both seasons as a reference to compare VE between vaccinated in current season only, vaccinated in previous season only and vaccinated in both seasons, using a mutually exclusive indicator variable. We refer to this analysis as the indicator analysis.
For each influenza type/subtype, we used seasons and sites for which I-MOVE MCCS included at least four cases of this influenza type/subtype per analysed stratum. We deemed sample size too small to attempt an analysis if there were fewer than 50 cases in the pooled study analysis. If the 10 events per variable (EPV) rule was TA B L E 1 Vaccine effectiveness a against influenza type/subtype overall and stratified by previous season vaccination, among the target group for influenza vaccination, aged 9 y or older, I-MOVE multicentre case-control study, influenza seasons 2011 /12-2016/2017 (patients with missing data on previous vaccination status excluded) violated, 20 we carried out a sensitivity analysis using Firth's method of penalised logistic regression to check for small sample bias.
In each analysis, among cases and controls, less than 6% were vaccinated in the current season only and less than 8% in the previous season only. More than 60% of cases and controls were unvaccinated in both current and previous season and less than 27% were vaccinated in both seasons (

| Vaccine effectiveness
We included four influenza seasons for the influenza A(H1N1)pdm09 and A(H3N2) analysis and two for the influenza B analysis (Table 4).

| Indicator analysis, using those unvaccinated in both seasons as reference
Influenza A(H1N1)pdm09 In 2013/2014, the VE point estimate for vaccination in both current and previous seasons was higher than that for vaccination in the current season only (Table 5 (Table 5).

Influenza A(H3N2)
In 2013 (Table 5). In TA B L E 2 Characteristics for influenza A(H1N1)pdm09, A(H3N2) and influenza B cases and controls belonging to the target group for influenza vaccination, aged 9 y or older, I-MOVE 2011I-MOVE /2012I-MOVE -2016I-MOVE /2017 Characteristics Number of test-negative controls a /total n (%)

Number of influenza B cases c,d /total n (%)
Median age (

Characteristics
Number of test-negative controls a /total n (%)  (Table 5).

| Sensitivity analysis
Where the EPV was <10, penalised and standard logistic regression VE estimates did not differ by more than 6.5% absolute, with an average of 3.4%, for both the stratified and indicator analyses, indicating little bias due to sparse data.  In a hospital-based study in Japan, a negative effect of previous vaccination on current season vaccine effectiveness was observed in individuals not infected but not in individuals infected with influenza A virus in the previous season. 8  While the test-negative design attempts to control for bias due to differential healthcare-seeking behaviours, 29 the study is observational and subject to the usual limitations, in particular adequately controlling for confounding.

| Influenza A(H3N2)
We observed that in all seasons, A(H3N2) VE point estimates were lower among those vaccinated in the previous season than in those unvaccinated in previous season.
Using unvaccinated in both seasons as reference, VE of var-

| Influenza B
The low precision of the VE against influenza B did not allow us to identify any pattern of the effect of previous vaccination. However  So far, the studies presenting the effect of previous season influenza vaccinations neither collected all this information nor were powered enough to provide precise results. The modifying or confounding effect of those parameters would no longer be relevant in terms of public health if the influenza vaccine had a high effectiveness.

| CON CLUS ION
Only well-powered, population-based, long prospective studies would allow understanding the immunological response to influenza vaccinations and infections. They could document influenza infections, vaccinations and type of vaccines used, cellular and humoural immune status before and during each season in vaccinated and unvaccinated, infected and not. A European mechanism of funding such large studies is needed to ensure powerful studies conducted independently from funding sources.
In the context of universal influenza vaccine development, the question is whether results of these proposed prospective studies would be available before the next generation of vaccines is accessible.