The results for the 2011–2012 season of the cycEVA study show a low-to-moderate protective effect for the 2011–2012 seasonal trivalent influenza vaccine against medically attended, laboratory-confirmed influenza in the target groups for vaccination, consistent with Spanish and European estimates [11–14]. In a late influenza season with a limited match between vaccine and circulating strains, we may suggest waning protection of the influenza vaccine 2011–2012 in the elderly with time since vaccination, although the trends were not statistically significant [12–14].
Although this study runs within the framework of the current SISSS, it is an observational study that followed a common protocol to be part of the European multicentre case–control study I-MOVE. Using the EU ILI case definition, GPs performed systematic sampling and collected high quality information on the main confounding factors described in the literature, thus reducing possible confounding bias.
By restricting the study to the epidemic period we also reduced the possible bias resulting from the inclusion of ILI patients when influenza viruses are not circulating . During this period of intense influenza activity, influenza positivity was higher than 50%, such that a higher number of cases than controls was included in the analysis.
The test-negative study design is becoming an increasingly well-established approach to measuring influenza VE and generates the highest estimates of VE [18–22]. This design avoids confounding by the propensity to seek care within the control group, which is negative for influenza, thus providing better comparability with the confirmed cases [23–27]. In addition, study participants were selected by practitioners according to a systematic sampling procedure before either the patient or the physicians knew the case and control status of the patients. This protocol should minimise selection bias . By restricting our analysis to ILI patients swabbed less than eight days after the onset of symptoms, we tried to minimise the possibility of misclassification because of false-negative RT-PCR results that could contribute to an underestimation of influenza VE. In addition, adjusting for swabbing week helped to overcome another possible limitation of the test-negative design by controlling the analysed data for calendar time . Nevertheless, as with any observational design, we cannot rule out residual bias and confounders [23, 28].
The 2011–2012 influenza season was characterised by several noteworthy aspects in Spain. First, this season was notably late, as in most of the northern hemisphere’s temperate zone with the exception of North Africa . The epidemic peak was not reached in Spain until mid-February 2012, whereas peaks are usually in late December or early January . Second, there was a predominant circulation of A(H3N2) influenza virus with a minimal contribution of the A(H1N1)pdm09 influenza subtype, which has been the predominant subtype since the 2009 pandemic. Third, there was a limited match between the vaccine and circulating A(H3N2) strains [29, 31].
Our global adjusted influenza VE estimate against A(H3N2) influenza infection, 45% (95% CI, 0–69), was consistent with the VE estimated in Australia for the 2011 season against influenza A(H3) (58%, 95% CI, -53 to 89)  and with the results of previous studies in years with a predominant circulation of seasonal influenza A(H3N2) virus, which determined influenza VE to range from 10% to 68%, depending on the degree of antigenic match [20, 21, 24, 33]. Although the effectiveness of the influenza vaccine is often less pronounced during seasons with antigenic mismatch between the vaccine and the circulating strains , in certain influenza season’s, antigenic changes occur without resulting in any apparent loss of influenza VE .
Several factors might have contributed to the low to moderate protective effect of the 2011–2012 trivalent influenza vaccine. Firsly, the circulating A(H3N2) influenza viruses in Spain clustered into several genetic groups that were reported to be antigenically and genetically distinct from the vaccine virus A/Perth/16/2009 . This limited match was observed globally in the northern hemisphere , resulting in a change in the WHO recommended A(H3) vaccine strain for 2012–2013 in the northern hemisphere .
Second, in an unusually late influenza season [29, 30], with a time lag between the vaccination campaigns and the start of the epidemic that was longer than in previous seasons, our results suggested a decrease in the protective effect of the 2011–2012 trivalent influenza vaccine with time since vaccination. This pattern was also observed in other studies [12–14].
A decreasing influenza VE over time could be related to increasing changes in circulating viruses towards the end of the season and/or potentially waning immunity in the months following vaccination [12, 29]. Phylogenetic analyses of the circulating influenza viruses in Spain did not support the hypothesis of an increased emergence of antigenically drifted A(H3) variants during the influenza season in Spain. There was a presence of mismatched influenza viruses since the beginning of the 2011–2012 Spanish influenza season, with a similar weekly proportion of circulating changed A(H3N2) influenza strains throughout the entire study period.
Re-analysing influenza VE data for the preceding season 2010–2011  we also observed decreasing influenza VE against A(H1N1)pdm09 in target groups for vaccination, from 62% (95% CI, -1 to 85) to 44% (95% CI, -122 to 86) for those vaccinated within three months or more than three months before the onset of symptoms, respectively. This lower decrease in VE over time (26%) observed in the previous season relative to the VE in our study for the 2011–2012 season (46%, data not shown) highlights the possibility that even in a usually timed season with a predominant circulation of a well-matched A(H1N1)pdm09 influenza virus, the influenza vaccine exhibits a degree of waning immunity.
To further explore the decrease in the protective effect of the vaccine with time since vaccination, we determined influenza VE by the phase of the season. We observed a decreasing influenza VE with time since vaccination during the early phase, from a high influenza VE of 95% (95% CI :45–99) at three months since vaccination, to a lower influenza VE point estimate of 36% (95% CI: -71 to 76) at more than 3.5 months since vaccination.
In the late phase, influenza VE estimates were compatible with null vaccine protection since shortly after vaccination. This finding is consistent with the finding that patients included in the late-phase subgroup had a median time since vaccination that was nearly one month longer that that in individuals in the early phase (128 days, range: 39–166 days vs 103 days, range: 45–135 days, respectively).
Together, these findings could reinforce the hypothesis of the possibly waning protection of the influenza vaccine. However, these results must be interpreted with caution because the study was limited by its small sample size. Therefore, although point estimates showed a substantial decrease with time since vaccination in the early phase, we could not demonstrate a significant decreasing influenza VE trend over time.
By age group, a decline in influenza VE with time since vaccination was also observed only in patients ≥65 years, although interpretation is limited by small sample size, which likely precluded the observation of a significant trend with time since vaccination .
A significant reduction in antibody titres 5–6 months after vaccination and, therefore, waning immunity following seasonal vaccination has been demonstrated in the elderly [38, 39].
However, in our study, we also obtained a higher influenza VE point estimate shortly after vaccination among the elderly compared with aged <65 years patients: 71% (95% CI: 6–91) and 45% (95% CI: -114 to 87), respectively . Conflicting results have been reported concerning the association between older age and the response to influenza vaccines. Several authors have found a reduced response in aged subjects, but others have reported no difference or even better results compared with younger control subjects [15, 41]. Other studies showed that subjects vaccinated in every epidemic season for several years were protected against influenza despite low titres of anti-haemagglutinin antibodies . Recently, studies performed during 2010–2011 influenza season demonstrated a higher influenza VE in patients vaccinated with both the current 2010–2011 and the previous 2009–2010 influenza vaccines in all age groups [7, 43] and in a population with major chronic conditions . In our study, a higher proportion of the elderly population (59%) compared with individuals <65 years (15%) was vaccinated with both the 2011–2012 vaccine and previous 2010–2011 seasonal influenza vaccines (p = 0.000). This factor could explain the higher influenza VE estimate obtained in the elderly relative to the younger group.
It is worth noting that currently there are few studies analysing influenza VE according to the time when the vaccination was given [12–14]. In addition, although levels of antibodies to seasonal inactivated influenza vaccine decline in the months following vaccination, this phenomenon does not necessarily reflect clinical VE . Consequently, although limited by their statistical power, our results contribute to the currently available scientific evidence on influenza VE, which have been poorly studied so far. Nevertheless, how antigenic drift in circulating influenza strains could affect influenza VE in a late season remains unclear. Further studies are needed to elucidate the impact of these and other possible factors on the protective effect of the influenza vaccine.
Our results were also limited by low vaccine coverage (VC), especially in individuals <65 years. In addition, we cannot extrapolate influenza VE estimates for older people to all elderly populations  because influenza VE in the ≥65-years-old test-negative controls was higher than VC in the same age group belonging to the GPs’ catchment area (70% vs. 56%). Another limitation of our study arises from the recommendation of swabbing to all ILI patients ≥65 years, a subgroup of the population targeted for vaccination that was included in our study, what could have introduced a selection bias that affected the influenza VE estimates. However, we believe that the target group for vaccination is a homogeneous study population with regard to vaccination, the main exposure of interest, because the study participants had more equal access to vaccination than the total population.
Because annual influenza vaccination is recommended by public health authorities, it is crucial to annually evaluate influenza immunisation programs and issue recommendations. This study has fulfilled this mission over the past four years by developing and implementing a sustainable system for annually assessing of influenza VE vaccination in Spain and Europe, as part of the I-MOVE project. Over two consecutive years, preliminary and end-of-season influenza VE estimates supported the feasibility of generating and disseminating preliminary influenza VE estimates while virus circulation is ongoing [11, 45].
The low-to-moderate protective effect of the 2011–2012 influenza vaccine in Spain that was observed in this study is in line with evidence from trials and observational studies to date . This finding highlights suboptimal vaccine performance in most years within the presently available influenza vaccines, with performance seldom exceeding 60%, thereby underscoring the urgent need for better and longer-lasting protective vaccines [22, 46]. Moreover, results on influenza VE, together with virological studies, should contribute to decision-making for the annual selection of influenza vaccine strains.
After five editions, the test-negative design of the cycEVA study has provided reliable information on annual influenza VE in Spain and may have important implications for the design of control influenza strategies.