2015/16 seasonal vaccine effectiveness against hospitalisation with influenza A(H1N1)pdm09 and B among elderly people in Europe: results from the I-MOVE+ project

We conducted a multicentre test-negative case–control study in 27 hospitals of 11 European countries to measure 2015/16 influenza vaccine effectiveness (IVE) against hospitalised influenza A(H1N1)pdm09 and B among people aged ≥ 65 years. Patients swabbed within 7 days after onset of symptoms compatible with severe acute respiratory infection were included. Information on demographics, vaccination and underlying conditions was collected. Using logistic regression, we measured IVE adjusted for potential confounders. We included 355 influenza A(H1N1)pdm09 cases, 110 influenza B cases, and 1,274 controls. Adjusted IVE against influenza A(H1N1)pdm09 was 42% (95% confidence interval (CI): 22 to 57). It was 59% (95% CI: 23 to 78), 48% (95% CI: 5 to 71), 43% (95% CI: 8 to 65) and 39% (95% CI: 7 to 60) in patients with diabetes mellitus, cancer, lung and heart disease, respectively. Adjusted IVE against influenza B was 52% (95% CI: 24 to 70). It was 62% (95% CI: 5 to 85), 60% (95% CI: 18 to 80) and 36% (95% CI: -23 to 67) in patients with diabetes mellitus, lung and heart disease, respectively. 2015/16 IVE estimates against hospitalised influenza in elderly people was moderate against influenza A(H1N1)pdm09 and B, including among those with diabetes mellitus, cancer, lung or heart diseases.


Background
Elderly populations, defined as those aged 65 years and above, and, more specifically, elderly people with underlying conditions, are at increased risk for hospitalisation due to influenza [1]. Influenza may also increase the severity of underlying chronic lung diseases, probably through inflammatory processes [2]. Viral pneumonia due to influenza seems to predispose to myocardial infarction, and congestive heart failures are more common during influenza seasons [3]. Patients with cancer treated with chemotherapy [4] and diabetic patients are more vulnerable to influenza. Their impaired immune response [5] could also affect host response to vaccination [6,7]. Evidence of the effectiveness of influenza vaccination in preventing severe clinical outcomes was recently described as low or very low among elderly people [8], and among patients with cancer [9], diabetes mellitus [10], lung diseases [11] [12], or cardiovascular diseases [13].
Despite the Council of the European Union and the World Health Organization's (WHO) recommendations to annually vaccinate elderly people [14,15], influenza vaccine coverage among elderly people remains below the 75% target in most European countries [16].
In this context, post-marketing studies to estimate the influenza vaccine effectiveness (IVE) among elderly people are needed to inform about vaccination benefits for vaccinees, detect subgroups in which the vaccine performs less well and identify vaccine types that perform best. In 2015, to address this issue, the Integrated Monitoring of Vaccines in Europe plus (IMOVE+) consortium initiated a network of hospitals across Europe to measure IVE against laboratory-confirmed hospitalised influenza among elderly people.
In the 2015/16 influenza season in Europe, influenza A(H1N1)pdm09 and influenza B (mainly Victoria lineage) viruses predominated [18]. We conducted a multicentre hospital-based test-negative design (TND) case-control study to measure the 2015/16 seasonal IVE against hospitalisation with influenza A(H1N1)pdm09 and influenza B among elderly people in Europe, by risk groups and for specific vaccine types. Each dot represents one location and there may be more than one hospital in one location.

Study sites and design
We set up a European network of 27 hospitals in 11 countries (Croatia, Finland, France, Hungary, Italy, Lithuania, the Netherlands, Poland, Portugal, Romania and Spain) (Figure 1), organised in 12 study sites (in Spain, Navarre region hospitals had their own coordination centre). Each study site adapted a generic protocol to their local setting [19,20]. Monitoring visits were organised to ensure the study was done similarly across hospitals. We conducted a multicentre TND case-control study.

Study period
In each study site, the study period started at least two weeks after the beginning of the vaccination campaign in the respective countries and lasted from the week of the first detection of a laboratory-confirmed case of influenza to the week of the last laboratory-confirmed case of influenza. We defined different study periods for influenza A(H1N1)pdm09 and B.

Study population
Our study population included all community dwelling patients aged 65 years and above who had no contraindication for influenza vaccination or previous laboratory-confirmed influenza in the season and agreed to participate. In the participating services of each hospital, patients admitted for clinical conditions that could be related to influenza were screened for eligibility. Study physicians, nurses or collaborating medical staff asked patients about onset of symptoms compatible with the definition of a severe acute respiratory infection (SARI) in the previous 7 days.
We defined a SARI case as a hospitalised patient with at least one systemic (fever or feverishness, malaise, headache, myalgia or deterioration of general or functional condition) and at least one respiratory sign or symptom (cough, sore throat or shortness of breath) at admission, or within 48 hours after admission.

Data collection
The hospital study teams swabbed patients meeting the SARI case definition. Specimens were tested by RT-PCR and patients classified as influenza A(H1N1) pdm09 cases, influenza B cases, other influenza cases or controls if their specimens tested negative for any influenza virus.
The hospital study teams collected information on patients' age and sex, influenza vaccination status including date and brand of the 2015/16 vaccine and the status in two previous seasons and underlying conditions listed for clinical risk groups recommended for influenza vaccination [21]. The underlying conditions included diabetes mellitus, obesity (defined as body mass index ≥ 30 kg/m2), cardiovascular conditions (such as congenital heart disease, congestive heart failure and coronary artery disease), lung diseases (such as chronic obstructive pulmonary disease, cystic fibrosis, asthma), renal and rheumatologic diseases, cancer, stroke, dementia and cirrhosis. Information on number of hospitalisations for underlying conditions in the previous 12 months, number of general practitioners (GP) visits in the previous three months, smoking status and functional impairment (based on Barthel index score [22]) was also collected.
Information on demographics and underlying conditions were collected from interviews with patients (or their relatives) and hospital and/or primary care databases. In study sites with no vaccination register, vaccination status was collected through interview with patients. For patients vaccinated or unable to provide their vaccination status, study sites called patients' GP or pharmacists to retrieve vaccination status, date and brand (Table 1).
We defined patients as vaccinated with the 2015/16 influenza vaccine if they had been vaccinated at least 14 days before symptoms onset. Otherwise, they were considered as unvaccinated.

Data analysis
We computed the IVE as (1 minus the odds ratio (OR) of vaccination between cases and controls) x 100. We performed a pooled one-stage analysis using the study site as a fixed effect and estimated IVE stratified on the presence of underlying conditions. All IVE estimates were adjusted for study site, date of SARI symptom onset and age modelled as restricted cubic splines with four knots (initial model). To adjust for additional potential confounders (sex, each group of underlying conditions, hospitalisation in the past year, more than one GP visit in the past 3 months, functional impairment, current smoking), we performed a multivariable analysis using an onward step by step modelling and analysing them as dichotomous variables. Patients with missing covariates were excluded from the analyses adjusted for these covariates. We retained in the model (full model) all covariates that changed the IVE estimate by 10% of more (relative change).
We grouped the vaccine brands in split virion, subunit or adjuvanted vaccines. To compute vaccine typespecific effectiveness, we restricted our analyses to  countries with at least one patient vaccinated with a specific type.
We also computed a pooled IVE with a two-stage model, adjusting study site-specific IVE for study site-specific confounders (same as listed above) when sample size allowed. We quantified the heterogeneity between site estimates using the I-square [23].
To minimise the inclusion of false influenza-negatives in the control group, we carried out sensitivity analyses by restricting population to (i) patients swabbed up to three days after symptom onset and (ii) patients not treated with antivirals until the day before swabbing.
Of the 52 cases of influenza B with a known lineage, 47 (90%) were Victoria and 5 (10%) were Yamagata. The 42 cases positive for influenza A(H3N2) did not allow us to compute IVE against this subtype.
The maximum number of influenza A(H1N1)pdm09 cases were recruited in weeks 5 to 8 of 2016 and the maximum number of influenza B and A(H3N2) cases in week 10 ( Figure 2).

Discussion
Our results suggest that the seasonal IVE against hospitalised influenza among elderly people was moderate during the 2015/16 influenza season in Europe for influenza: 39% overall, 42% against influenza A(H1N1) pdm09 and 52% against influenza B . These estimates did not vary between categories of underlying conditions.
Data from European virological surveillance reported that most of the characterised influenza A(H1N1)pdm09 viruses belonged to the emerging subclade 6B.1, defined by haemagglutinin amino acid substitutions S162N and I216T [18]. Despite these genetic evolutions, A(H1N1)pdm09 viruses were considered antigenically similar to the northern hemisphere vaccine component A/California/7/2009. IVE estimates against hospitalised A(H1N1)pdm09 was consistent with the results we reported in 2012/13 and 2013/14 among hospitalised elderly people [24,25].
In 2015/16, the circulating influenza B Victoria lineage was distinct from the Yamagata vaccine component [26] and there was no quadrivalent vaccine used in our study population. IVE against influenza B was close to what we reported, using the same generic protocol, in 2012/13 (66% in the 65-79 year-olds and 44% in the 80 year-olds and older) in a season with co-circulation of B Victoria and Yamagata lineages and a Yamagata vaccine component [24,27]. These results suggest some cross-lineage protection and they are in line with previously reported data in GP-based studies [28,29] and a meta-analysis of eight randomised controlled trials with mismatched B viruses resulting in a VE of 52% (95% CI: 19 to 72) among healthy adults [30]. Further studies are needed to increase the understanding of mechanisms of cross-lineage protection for influenza B and better guide policy makers in terms of recommendations for using trivalent or tetravalent seasonal vaccines.
We observed higher point estimates of IVE for the inactivated split virion vaccines compared with inactivated subunit vaccines, although the 95% CIs of the point estimates were widely overlapping. The completeness of data on vaccine type was high (1% of missing vaccine type among those vaccinated), thus these results, concurring with published data [31][32][33], could be due to differences in T-cell responses conferred by the two vaccine types [34]. However, they should be interpreted with caution as they may be due to random variation. Further evidence, and pooling of several years of data would be required to obtain precise vaccine type specific effectiveness. Higher adjuvant vaccine coverage would be needed to compute adjuvant vaccine specific IVE. This would be useful information to adapt influenza vaccination strategies among elderly people.
Recent reviews underlined the need for further evidence of seasonal IVE against laboratory-confirmed influenza in elderly people and patients with underlying conditions [8][9][10][11][12][13]. We were able to collect high quality data from 1,802 elderly patients hospitalised with SARI, making our study one of the largest hospitalbased studies on IVE. The large number of participants, and a vaccine coverage close to 50%, enabled us to compute IVE against type/subtype-specific influenza among patients with specific underlying conditions. Our results suggest that, in 2015/16, the seasonal influenza vaccine provided protection against hospitalised influenza A(H1N1)pdm09 and B in the elderly with diabetes mellitus, heart and lung disease. We were unable to refine the underlying conditions categories further. To better guide vaccine recommendations, IVE among patients receiving specific treatment (e.g. statins [35,36], chemotherapy [9,37]) or with more specific conditions (e.g. asthma, chronic obstructive pulmonary disease) would be needed. A larger sample size would be required for such studies.
We collected information related to access to care, health conditions and smoking status. Recruited cases and controls were similar. We adjusted our estimates for study site, onset week and age. Further adjustment for potential confounders (underlying lung, heart, renal disease, diabetes mellitus, cancer, obesity and hospitalisations in the past year) did not change the estimates. However, as in any observational study, we cannot exclude unmeasured confounding leading to over-or under-estimation of the IVE.
The contribution to the pooled dataset was different between study sites. The two Spanish sites recruited 44% of the patients. The viruses circulating and vaccines used in Spain were similar to the other countries. Consequently we do not expect the over-representation of Spanish sites to have biased our overall estimates. Variations in the number of recruited individuals may be explained by differences in local influenza activity or number and size of participating hospitals/ services. We believe that access to hospitalisation in case of severe influenza is similar across participating European countries. A common generic protocol and the monitoring of its implementation through on-site visits contributed to ensuring comparability of patients recruited and data collected across study sites. We measured low statistical heterogeneity based on I-square values. However, small number of estimates and large study-specific CIs may hinder adequate quantitative assessment of heterogeneity between studies [38]. True differences between study site specific IVE could be related to different vaccines used during this season or different immunological profiles of recruited patients including their past vaccination histories [39]. Larger study site-specific sample sizes are required to ensure that the differences in IVE across study sites are not due to chance. Currently, multicentre studies are necessary to obtain precise IVE estimates.
A recent publication by Foppa et al. suggested that measuring IVE against laboratory-confirmed influenza SARI hospitalisation using the TND was subject to biases if the test-negative controls were hospitalised because of an exacerbation of underlying lung disease unrelated to a respiratory infection [40]. In our study, cases and controls had similar prevalence of underlying lung disease. Underlying lung disease did not appear to confound IVE estimates, even when combined with a proxy of its severity (hospitalisation because of underlying conditions in the past 12 months). Cohort and TND-based IVE estimates against laboratory-confirmed hospitalised influenza in Navarre repeatedly showed similar estimates, reassuring on the appropriateness of TND at hospital level [41].
Several studies suggest that past influenza vaccinations may decrease or enhance current vaccine effectiveness depending on previous and current vaccine and circulating strains as well as past exposure to the virus [24,32,[42][43][44]. A large proportion of our vaccinated population had been vaccinated in the previous season(s) but the very small number of patients with varying repeated vaccination status over the years did not allow us to measure the effect of previous vaccinations. To understand the effect of repeated vaccinations on IVE, large cohorts of individuals with different vaccination patterns and symptomatic (and asymptomatic) influenza infection status over the years would be needed.

Conclusion
Our multicentre test-negative case-control study estimated that in 2015/16 the seasonal influenza vaccination prevented approximately half of the cases of hospitalisation with laboratory-confirmed influenza among vaccinated elderly people. Our results suggest that vaccination provided similar protection to elderly patients with underlying diabetes mellitus, cancer, lung and heart diseases. Because vaccination remains the most effective preventive measure against severe influenza among elderly people, increasing the vaccine coverage in this group should be a priority. This pilot season of the hospital-based I-MOVE + project proved that obtaining precise estimates of IVE against a severe influenza outcome among elderly people was feasible. Enlarging our network and its sample size will enable us to better guide vaccination strategies against severe influenza cases by comparing the performance of different vaccine types and identifying risk groups for poor response to vaccination. IVE network and contributed to the writing of the research article.