The impact of repeated vaccination on influenza vaccine effectiveness: a systematic review and meta-analysis

Seasonal influenza vaccination is the predominant strategy for preventing influenza-related morbidity and mortality. Annual vaccination is recommended because of waning immunity and because influenza strains undergo antigenic drift, necessitating reviewing and, in most seasons, changing the vaccine to better match the upcoming season’s strains [1]. Because of the frequently changing vaccine, influenza vaccine effectiveness (VE) is assessed annually.

With increasing numbers of people being immunized against influenza annually, the impact of repeated vaccination has gained significant interest. Of particular interest are older adults (65 years and older), who tend to have more comorbidities as they age, as both age and co-morbidities increase their risk of influenza-associated complications [2]. If repeated vaccination negatively impacts current VE, then having been repeatedly vaccinated in earlier years may be detrimental to the protection of older adults when they need it most. Studies from the 1970s and 1980s found inconsistent results regarding the impact of repeated vaccination [3, 4]. In 1999, a systematic review and meta-analysis of field studies, trials, and serologic studies found no evidence of negative impacts of repeated vaccination [5]. More recently, some studies have found VE to be reduced in those who received repeated prior influenza vaccinations [6‐8].

Since most VE studies now report estimates taking into account vaccination status for both current and prior seasons, we sought to evaluate the impact of repeated vaccination on VE through a systematic review and meta-analysis. We aimed to assess the impact of repeated vaccination to provide evidence to support patient and clinician decision-making about receiving the current season’s influenza vaccine. We considered two patient-relevant scenarios: (1) for someone who received last season’s vaccine, should he/she also receive this season’s vaccine? (vaccination in both seasons versus prior season only); and (2) for someone who did not receive last season’s vaccine, should he/she receive this season’s vaccine? (vaccination in current season only versus neither season). We also considered a policy-relevant scenario, comparing VE for vaccination in both seasons versus the current season only. This latter scenario is not relevant to patients because they cannot alter their vaccination history; however, the findings may influence policy decisions regarding whether or not to offer annual vaccination to the entire population if there was evidence suggesting that repeated vaccination could negatively impact future VE.

The results presented below have been corrected from our previously published review on this topic. The previous article included one study that did not meet the inclusion criteria, omitted one estimate from an included study, and included two estimates from populations that were already included in other estimates (i.e., estimates from a subset of a population that were also captured in other included estimates).

We identified 3435 unique articles from the database searches (Fig. 1). After screening titles and abstracts, we selected 634 articles for full-text review. Of these, 26 studies met the inclusion criteria for the qualitative synthesis, and 20 were included in the meta-analysis [6‐8, 15‐37]. We observed excellent agreement between reviewers for the title and abstract screen (kappa, Ƙ = 0.94) and for the full-text review (Ƙ = 0.98). No additional studies were identified from hand-searching references. One study was excluded from the qualitative synthesis and meta-analysis because, while it included persons with laboratory-confirmed influenza in the vaccination groups of interest, the study provided VE estimates only for severe or fatal influenza outcomes rather than for any laboratory-confirmed influenza [38]. We excluded six studies from the meta-analysis but included them in the qualitative synthesis: four studies because they only provided VE estimates for any influenza rather than by influenza type/subtype [15, 23, 31, 32], and two studies because they provided VE estimates for any influenza and “prior season vaccination” was not restricted to the immediate year prior to the study season [16, 26].

The 26 included studies captured influenza seasons between 2004–2005 and 2014–2015, with most reporting estimates for the 2010–2011 to 2014–2015 seasons (Table 1). One study was from the Southern hemisphere [32], one study was restricted to pregnant women [35], and two studies were in pediatric populations [17, 34]. Most studies featured outpatient data, but two studies used solely inpatient data [25, 37], and two studies used data from both settings [15, 16]. All studies used reverse-transcriptase polymerase chain reaction (RT-PCR) testing to confirm influenza infection.

For 24 of the 26 studies, we extracted the variables included in the multivariable regression models used to obtain VE estimates (Additional file 2: Table S1); the remaining two studies did not clearly report these variables. All 24 studies with available information adjusted for age, and the majority adjusted for presence of high-risk conditions or comorbidities (n = 17; 71%) and calendar time (n = 15; 63%). Many studies also adjusted for time between illness onset and sample collection (n = 12; 50%), and sex (n = 10; 42%).

All of our included test-negative design studies were deemed to be at low risk of bias, and all included calendar time in their adjusted models [6‐8, 15, 16, 18‐21, 24‐30, 32, 34, 36, 37]. The remaining case-control studies were also categorized as being at low risk of bias [17, 35], as were all the included cohort studies [22, 23, 31, 33]. Details of the evaluation of the included studies are provided in Additional file 3: Figure S1.

Among the 20 articles included in the meta-analysis, there were 16 analyses for influenza H1N1, 14 for H3N2, and 13 for B that compared VE among those vaccinated in both seasons to those vaccinated in the prior season only. When compared to vaccination in the prior season only, VE was higher for vaccination in both seasons for influenza H1N1 (∆VE = 25%; 95% CI 14%, 35%; I² = 0%) and B (∆VE = 18%; 95% CI 3%, 33%; I² = 26%), but not H3N2 (∆VE = 7%; 95% CI – 7%, 21%; I² = 4%) (Table 2, Figs. 2, 3, and 4). When stratified by influenza season, the results for all seasons were consistent with the overall results (Additional file 4: Table S2).

Sixteen analyses for influenza H1N1, 14 for H3N2, and 14 for B compared VE among those vaccinated in the current season only to those vaccinated in neither season. VE was higher for vaccination in the current season compared to neither season for influenza H1N1 (∆VE = 62%; 95% CI 51%, 70%; I² = 26%), H3N2 (∆VE = 45%; 95% CI 35%, 53%; I² = 0%), and B (∆VE = 64%; 95% CI 57%, 71%; I² = 0%) (Table 2, Figs. 5, 6, and 7). The results for individual seasons were consistent with the overall results (Additional file 4: Table S2).

For the comparison of those vaccinated in both seasons to those vaccinated in the current season only, 16 analyses for influenza H1N1, 14 for H3N2, and 14 for B compared VE among from the 20 included articles. We observed no statistically significant VE difference between vaccination in both seasons and vaccination in the current season only for influenza H1N1 (∆VE = 3%; 95% CI – 8%, 13%; I² = 0%). We observed statistically significant VE differences between vaccination in both seasons and vaccination in the current season only for influenza H3N2 (∆VE = − 20%; 95% CI – 36%, − 4%; I² = 35%), and B (∆VE = − 11%; 95% CI – 20%, − 2%; I² = 0%) (Table 2, Figs. 8, 9, and 10). For the H3N2 comparison, the results for individual seasons were inconsistent with the overall result except for the 2014–2015 season (i.e., ∆VEs were not statistically significant in the comparison of both seasons versus current season for any season except for 2014–2015) (Additional file 4: Table S2).

To address overlapping patients in the 2010–2011 season analyses in Spain (one conducted in the Navarra region [20] and one that included eight regions in Spain [19]), we performed a sensitivity analysis removing the Navarra region study [20] from the H1N1 meta-analyses and found that it had no effect on the results for any of the three comparisons. When we removed the Navarra region study, we found that for H1N1 VE_both-VE_{prior only} was 25% (95% CI 14%, 35%); VE_{current only} was 62% (95% CI 51%, 71%); and VE_both-VE_{current only} was 2% (95% CI− 9%, 13%). The original estimates for H1N1 were 25% (95% CI 14%, 35%), 62% (95% CI 51%, 70%), and 3% (95% CI− 8%, 13%), respectively.

Among the studies included in the qualitative synthesis but not the meta-analysis, two studies presented results using a definition of “prior season vaccination” that included multiple prior seasons and therefore did not meet the inclusion criteria for the meta-analysis [16, 26]. One of these studies considered vaccination history over two consecutive seasons using data from nine influenza seasons (2000–2001 to 2008–2009); those vaccinated in the current season only had the highest VE [26]. Finally, a study from the Navarra region in Spain that assessed vaccination over the current and two prior seasons showed a range of results [16]. Residual VE without current vaccination was noted if vaccinated in both the prior two seasons. For both influenza H3N2 and B, vaccination in the current season and one prior season resulted in quite low VE, whereas vaccination in the current and both prior seasons resulted in higher VE. VE against influenza B was highest among those vaccinated in the current season only compared to the other vaccination groups, whereas this group had the lowest VE against H3N2 [16].

Six studies [15, 16, 23, 26, 31, 32] presented results for any influenza rather than by influenza type/subtype, two of which were summarized above because they also used multiple prior seasons [16, 26]. There were five estimates from the four remaining studies not summarized above. Of these, three favored vaccination in the current season only, and two favored vaccination in both seasons. None of the estimates favored vaccination in the prior season only. In the one study that presented two VE estimates, seasonal differences were apparent. In the Southern Hemisphere 2011 season, the highest VE was observed among those who had been vaccinated in both current and prior seasons, but in the Southern Hemisphere 2010 season, the highest VE was observed among those who had received the current season vaccine only [32].

We found that irrespective of a patient’s vaccination status for the prior season, current season vaccination is associated with greater protection against laboratory-confirmed infection by influenza H1N1 and B. This was evident comparing vaccination in both seasons to vaccination in the prior season only. Furthermore, compared to no vaccination for either season, individuals who received the current season’s vaccine had greater protection against all three influenza types/subtypes. Therefore, vaccination in the current season is generally the best option for the patient. Recent studies have raised questions about the impact of repeated vaccination [6‐8], which is of concern to policy-makers with regard to annual influenza vaccination recommendations. Of relevance to the policy-maker (but not of pragmatic relevance to the patient, who cannot alter his/her vaccination history or predict future influenza strain circulation), we observed lower VE in those vaccinated in both seasons compared to only in the current season for influenza H3N2 and B, but not for H1N1. The 2014–2015 influenza season influenced the overall significance, where pooled VE across three studies was lower for those vaccinated in both the current and prior season compared to those vaccinated in the current season alone. However, this comparison does not account for increased susceptibility to influenza during the prior season due to being unvaccinated for that season. Based on the NOS, we assessed that the studies included in this review had a low risk of bias. However, the theoretical underpinnings of the test-negative design are still in the process of explication [39‐41], and there has not yet been a theoretical assessment of the potential biases in evaluation of repeated vaccine effects using the test-negative design.

The results of this review are not consistent with those found by Beyer and colleagues in 1999 [5]; their meta-analysis of seven field studies and 12 serologic studies found no significant difference between the single and multiple vaccination groups for all influenza subtypes combined. Our study differs from Beyer et al. by including studies that feature contemporary laboratory testing methods, influenza subtype-specific analysis, and study designs with consistent vaccination comparison groups. A recently published meta-analysis reported pooled VE estimates for the same vaccination status groups as our study (prior only, current only, both seasons) [42]. Similar to our study, that study found VE to be consistently lowest among those vaccinated during the prior season only. As well, for the 2014–2015 season, VE against H3N2 tended to be higher for those vaccinated in the current season only compared to those vaccinated during both seasons. However, that study did not examine the differences in VE as presented in this study.

In our review, the comparison groups used in the meta-analysis provided a more refined calculation of VE that accounted for recent vaccination history. Standard VE calculations (those that do not account for prior vaccination history of the vaccinated group) are comparing those vaccinated in the study season (a mixture of subjects vaccinated in the current season only and those with current and prior vaccination) to a reference group of those not vaccinated in the study season (which includes both those vaccinated in neither season and those vaccinated in the prior season only). Our study allowed for these vaccination groups to be analyzed separately to understand the impact of prior season’s vaccination on current season VE.

Our study was further strengthened by aligning the VE comparisons with patient and policy perspectives, in order to aid decision-making by patients, practitioners, and policy-makers. Additionally, by calculating the differences in VE between the various vaccination groups within each study, we controlled for any methodological biases unique to a particular study, such as study design or study settings (e.g., outpatient vs. hospital-based studies), since these biases would apply equally to each vaccination group. Thus, rather than first pooling the VE estimates from each vaccination group across studies and subsequently taking the difference, we pooled the differences obtained from VE estimates within each study. Finally, because VE can vary by age group and influenza type/subtype, this study was strengthened by the detailed stratification of results by type/subtype, as well as by using VE estimates for the most specific patient groups available (e.g., age-stratified groups rather than “all ages”).

This study also has some limitations. First, the analysis accounts only for vaccination status in one prior season. Results might differ when considering a patient’s vaccination history over a greater number of seasons, which is particularly important when considering the importance of influenza VE in older adults who potentially have received many years of consecutive vaccinations. McLean and colleagues found no difference when exploring VE over two consecutive seasons, but when they used a reference group with no vaccination over six seasons, those vaccinated in the current season only and not in the previous five seasons had the highest VE against influenza H3N2 and B, with progressively lower VE with increasing vaccines received over the previous five seasons [6]. Few studies reported on vaccination history beyond prior and current seasons, and they did not group history consistently, so further analysis incorporating the effects of serial vaccination from these studies was not possible, but is an important analysis to conduct in the future when more data are available. Second, our study did not account for past influenza infection, which may have provided some protective effect against laboratory-confirmed influenza in subsequent seasons [43]. A patient’s first exposure to influenza vaccination or infection can impact subsequent responses to vaccination or infection (referred to as original antigenic sin or back-boosting), which was not accounted for in this study [44]. Third, this study did not differentiate the types of influenza vaccines used for vaccination (e.g., live attenuated or inactivated; quadrivalent or trivalent; adjuvanted or unadjuvanted; high dose or standard dose). Given the differing types of immune response induced by these various products, different impacts of prior vaccination on current season VE may result. Fourth, we evaluated the absolute difference in VE, instead of looking at a ratio, which could be considered more appropriate given the scale on which VE is calculated. However, reporting ratios introduces other challenges, such as accommodating negative values and estimating confidence intervals. Since deriving practical conclusions for annual vaccine decision-making was the goal, we reported more intuitive differences in VE, as others have done previously [14, 45]. Finally, based on the limited available information in each study, we could not adjust for the match between the current season’s vaccine and the circulating strains, the prior season’s vaccine and the current season’s circulating strains, or changes in vaccine strains from one season to the other, all of which may affect VE from 1 year to the other as noted by Smith et al.’s antigenic distance hypothesis [46]. Skowronski et al. recently examined VE for influenza H3N2 in Canada using this framework and concluded the effects of repeated vaccination were consistent with the antigenic distance hypothesis [47]. We attempted to assess VE based on antigenic distance in the included articles by considering the vaccine strain and circulating strain match where possible, but not all studies provided detailed strain information. In the articles with sufficient information, the variation of vaccine and circulating strain matches were too few and were grouped by season, and as seasonal analysis was already included in our meta-analysis, no further information was gained. However, consistent with Skowronski et al.’s findings, we did find significant negative interference in the 2014–2015 influenza season, which supports the antigenic distance hypothesis which predicts that this would occur when vaccine strains are homologous from 1 year to the next but the prior season’s vaccine does not match the current circulating strain. Future VE studies should continue to incorporate vaccination status in prior seasons and provide as much detail as possible to allow assessment of the match between vaccine and circulating strains and the changes in vaccine strains over time. Future studies should also assess the impact of vaccination over multiple past seasons.

In conclusion, from the patient’s perspective, vaccination in the current season is generally the best option regardless of prior season vaccination. From a policy perspective, although VE was lower against H3N2 and B for individuals vaccinated in both seasons compared to those vaccinated in the current season only, this result may vary from season to season; in addition, it should be noted that past vaccination history cannot be altered and this comparison disregards susceptibility to influenza during the prior season among those vaccinated in the current season only.