Protection Against Influenza After Annually Repeated Vaccination:  A Meta-analysis of Serologic and Field Studies FREE

VACCINATION strategies should result in effective protection. Current efforts with respect to influenza rightly focus on the prevention of pandemic influenza (pandemic planning),¹ but the effects on mortality and morbidity between pandemic periods should not be neglected. The number of deaths attributed to annual epidemic influenza during the past 60 years is many times greater than that attributed to pandemic influenza.² Recommendations for the use of inactivated influenza vaccine in humans proceed from the necessity to administer the vaccine every year³ because the antigenic properties of wild influenza viruses change frequently, and antibody titer levels may decline to nonprotective levels within a year after vaccination. Hoskins et al⁴ challenged this common policy by claiming that protection after annual influenza vaccination would successively decrease. Although doubts about the validity of the findings of Hoskins et al had been articulated,^5- 8 uncertainty about the issue has remained, which may contribute to suboptimal vaccine use in people at risk.⁹

Evidence of influenza vaccine efficacy in humans is derived from 3 types of clinical studies: the experimental study, the field study, and the immune-response study (serologic study). In experimental studies, volunteers are challenged by live influenza viruses under strictly controlled circumstances. This approach is scientifically most satisfying but cannot be applied in populations at risk for serious complications from influenza infection. Field studies register morbidity or mortality during naturally occurring influenza outbreaks. Despite certain drawbacks and limitations of this approach,^10- 11 field studies have convincingly proved the efficacy of single influenza vaccine.⁷ Challenge studies in healthy children and young adults have established that a high serum antibody level can prevent infection.^12- 14 It has, therefore, been widely accepted as a surrogate marker for protection against influenza and vaccine efficacy in serologic studies.

We attempted to identify all field and serologic influenza studies published during the past 3 decades that compare appropriate outcome measures in subjects vaccinated for the first time (single vaccination) with subjects vaccinated also in previous years (multiple vaccinations). Pooling of these data allows a quantitative analysis of whether the effect of influenza vaccination decreases when it is repeated annually.

Titles and abstracts of articles published from January 1966 to December 1997 and included in the MEDLINE computerized library system (National Library of Medicine, Bethesda, Md) were checked for the combination of "influenza" and "vaccine" or "vaccination." The search strategy was limited to human studies (including articles in English, Dutch, German, French, Spanish, Italian, and Russian). A total of 2391 references were found to comply to this strategy. Next, all titles and abstracts were evaluated. Articles were selected for further reading if there was a reference to a study design comprising more than 1 year or if the words "revaccination," "annually repeated vaccination," "preimmunized," or "hemagglutination inhibition" (HI) were used. This selection produced 431 references. Also, cross-references in already identified articles were included. These articles were then read in search of information on the effect of annually repeated vaccination.

In the case of field studies, any article was accepted that presented data on influenza-related morbidity or mortality in groups with appropriate single vaccination and multiple vaccinations. "Appropriate vaccination" was defined as vaccination just before the influenza season wherein the outbreak under study occurred, in subjects who had not been vaccinated in the years before (single), or in subjects who also had been vaccinated in 1 or more consecutive years before (multiple). In the case of serologic studies, any article was accepted that included the sampling of 2 blood specimens, 1 before and a second 2 to 6 weeks after vaccination, in groups with single and multiple vaccinations. The assessment of serum antibody levels should have been done by a microtiter HI test. The most meaningful serologic variable is the postvaccination geometric mean antibody titer,¹⁵ but virtually no article presented a measure of dispersion for these values, so that variable could not be used for a meta-analysis. Instead, we used the proportion of subjects exceeding a certain antibody titer threshold after vaccination, conventionally referred to as "serologic protection rate." Other commonly used variables, like mean fold increase or response rate, may show mathematical flaws¹⁵ and were, therefore, not considered here.

Studies may contain 1 or more trials, according to the year of influenza outbreak and the subtypes involved (influenza A-H3N2, A-H1N1, and B). The individual trial was the unit of meta-analysis.

Per trial, the difference between protection rates in groups with single (S) and multiple (M) vaccinations was calculated as follows: field protection-rate difference (PRD)=(S_f/N_S − M_f/N_M) × 100%, with N_S and N_M indicating the number of vaccinated subjects and S_f and M_f, the number of protected subjects after exposure (cases without influenza-related morbidity or mortality); and serologic PRD=(S_s/N_S − M_s/N_M) × 100%, with S_s and M_s indicating the number of subjects exceeding the protective antibody threshold after vaccination.

A PRD of greater than 0 favors single vaccination, and a PRD of less than 0 favors multiple vaccinations in preventing influenza infection after challenge (field protection) or in achieving high antibody titers after vaccination (serologic protection).

Protection-rate differences were subjected to meta-analysis according to Yusuf et al¹⁶ (fixed-effects model) and DerSimonian and Laird¹⁷ (random effects accounting for a possible heterogeneity of treatment effects). When a heterogeneity of treatment effects occurred, we first attempted to reduce the heterogeneity by forming subpopulations. The meta-analysis was repeated with odds ratios and relative risks to check the robustness of the pooled results in measuring effects. In some instances, more than 1 field or serologic trial within a study referred to identical groups of vaccinees. Therefore, a sensitivity analysis was performed by excluding the trials in question 1 at a time and observing whether the pooled result changed essentially.

For calculations, a software program (Meta-Analyst, version 0.991/1997) provided by Joseph Lau, MD, New England Medical Center, Boston, Mass, was used. The significance level for all calculations was .05.

Eight articles describing field studies were identified by the literature search and cross-references. One article, by Hoskins et al,⁴ lacked appropriate groups with single and multiple vaccinations and was not included.¹⁸Table 1 shows some relevant properties of the 7 remaining articles.^19- 27 Nine trials were conducted in subjects of various age classes and different health states. "Influenza-related cases" were defined as incidents of influenzalike illness with or without laboratory confirmation in 6 articles^19- 23,25 or as occurrences of death clinically related to influenza.²⁴ Two of the trials cumulated 3 observational years. The comparability of single- and multiple-vaccination groups in the risk of influenza-related morbidity (age, proven efficacy, and other factors) was controlled by a randomized study design in most articles.^{20- 21,23,25} In total, 13 influenza outbreaks occurred in 7355 subjects with single or multiple vaccinations (multiple counts of subjects allowed). Most outbreaks concerned influenza subtype A-H3N2. The antigenic match between the vaccine strain and the epidemic virus was excellent or sufficient in all outbreaks.

Of 13 trials, 12 supported the hypothesis that annually repeated vaccination provides a protection at least as good as that of single vaccination, but 2 of them did not provide exact quantitative information needed for meta-analysis (F-1, F-5, Table 1), and for 1 trial (F-6, a case-control study), the analysis was made on the basis of odds ratios. On the remaining 10 trials with 5117 subjects, all using laboratory-confirmed influenzalike illness as the clinical end point, a meta-analysis was performed (Table 2). The test of heterogeneity showed that homogeneity among the trials could not be rejected (P>.05). The pooled field PDR was calculated as 1.1% (95% confidence interval, −0.2% to 2.4%; random-effects model), ie, multiple vaccination had no effect on the field protection rate. The result was similar when using the fixed-effects model or other effect measures or performing the sensitivity analysis as described in the "Calculations" subsection of the "Methods" section.

Twelve articles^5,28- 38 presenting data on seroresponse after single and multiple vaccinations were identified (Table 3). Of 4 articles,^{32- 33,37- 38} the original raw data were used. The articles covered 53 trials with 12,468 postvaccination titers (multiple counts allowed in subjects receiving bivalent or trivalent vaccines). All trials were performed in young or elderly adults or both. Vaccine doses were constant throughout the trials (10 or 15 µg of hemagglutinin per dose, except for 1 trial²⁸ wherein doses between 100 and 400 IU were used), but other details—health state, vaccine types, HI thresholds of subjects with high titers—varied considerably among studies. Hemagglutination-inhibition tests were performed by either methods similar to those described by Dowdle et al³⁹ (Keitel,⁵ Howells,²⁸ Powers,²⁹ Peters,³⁰ Gross,³¹ McElhaney,³³ Glathe,³⁴ Pyhälä,³⁶ and their colleagues) or the method of Masurel et al⁴⁰ and Beyer et al⁴¹ (Beyer,^32,37 Govaert,³⁵ de Bruijn,³⁸ and their associates).

Of all 53 trials, 44 (83%) supported the working hypothesis that multiple vaccinations provided a protection at least as good as that of single vaccination, and 9 did not. One trial (S-4, n=129) (Table 3) that did not support the working hypothesis presented geometric mean antibody titers only, and no serologic PRDs could be calculated. Table 4 shows the results for the remaining 52 trials. The serologic PRDs varied in a large range around 0 (−25.4% to 26.6%). Eight serologic PRDs were significantly greater than 0, and 7 were significantly less than 0. The estimated pooled rate difference was close to 0: 1.7% (95% confidence interval, −1.3% to 4.8%; random-effects model) or 0.8% (95% confidence interval, −0.4% to 1.9%; fixed-effects model). This result should be interpreted with caution because it was based, in part, on multiple observations (the same volunteers were vaccinated with bivalent or trivalent vaccines, producing 2 or 3 results per trial), and the test of heterogeneity clearly indicated the absence of homogeneity between trials (P<.001). Subdividing according to influenza subtypes suggested slight differences between A-H3N2 (pooled rate difference <0) and A-H1N1 and B (pooled rate difference >0) and did not reduce heterogeneity. Similarly, subdividing according to age classes (younger adults vs elderly), study design types (cohort vs 1-year design), HI assay methods, or vaccine types (not shown) did not reduce heterogeneity. On the other hand, robustness and sensitivity analysis did not reveal any essential changes of the result. Therefore, the true pooled serologic PRD, despite considerable heterogeneity between trials, is close to 0, and there is no evidence for assuming a generally lower seroresponse of annually vaccinated subjects than that of subjects vaccinated for the first time.

The state of previous vaccination does not influence field or serologic protection against influenza. Concerns about a decreasing field protection after a number of annual vaccinations, as expressed by Hoskins et al,⁴ could not be substantiated in a review of 7 articles involving 13 field trials in a total of 7355 vaccinees. Of these 13 field trials, 10 could be subjected to a meta-analysis because they had been based on the same clinical case definition (influenza-related morbidity, ie, clinical influenzalike illness confirmed by laboratory means). These trials showed a sufficient intertrial homogeneity (P=.30, Table 2) and produced a pooled field PRD virtually equal to 0 (ie, no difference in field protection between groups with single and multiple vaccination). This result remained stable in robustness and sensitivity analyses. A selection bias that may have influenced the change of a vaccinee to receive either single or multiple vaccination could not be detected.

Of 53 serologic trials with 12,468 observations in 12 articles, 44 (83%) confirmed the results of the field studies. Fifty-two trials could be subjected to meta-analysis that, again, produced an overall PRD virtually equal to 0. However, the interpretation of this result deserves comment. The total range of serologic PRDs and the 95% confidence intervals of most individual serologic PRDs were large. Several trials even "contradicted" each other by significantly favoring either single vaccination or multiple vaccinations (Table 4). Considerable heterogeneity of treatment effects was present. This variability between serologic studies is well known and has been described earlier.^{11,37,42- 43} Although the postvaccination HI titer is a valuable surrogate marker for real protection, it may depend on a number of cofactors that are imperfectly controlled by the study design, among others, prevaccination titer and study population characteristics (age; history of previous exposure to natural influenza, especially the effect of the "original antigenic sin"⁴⁴; health state; etc).

The present heterogeneity among serologic studies does not argue against our conclusion that repeated vaccination is not associated with a decrease in effectiveness. The pooled serologic PRD (calculated by a meta-analysis method that accounts for heterogeneity) and the huge majority of individual serologic trials did not detect any meaningful difference between groups of single and multiple vaccinations. This accords with the results of the field studies and may be taken as additional evidence that annual influenza vaccination does not affect vaccine-provided protection against natural influenza.

On the other hand, large heterogeneity among serologic studies is undesirable and should stimulate further efforts to standardize study designs between study centers. The HI test should be standardized among laboratories, eg, by introducing internationally accepted reference serum, and the end points of serologic studies should be predetermined.¹⁵ In particular, the state of previous vaccination, besides the prevaccination titer,³⁷ should be included as a possible source of effect modification in the statistical analysis of influenza vaccine trials. It could be shown (Table 4) that in several individual trials, the state of previous vaccination was indeed a significant confounder, although its general effect was virtually 0.

For the past 20 years, the articles by Hoskins et al^4,19- 20 have caused doubts about the common policy of vaccinating populations at risk on an annual basis (the Hoskins paradox). Their results can be disputed, particularly because of serious shortcomings in study design. Hoskins et al⁴ described a vaccination campaign in adolescents and 4 influenza outbreaks from 1970 to 1976. During the first outbreak in 1972-1973 (influenza A-H3N2), no adverse effect of repeated vaccination could be detected (see also Table 1, F-1), but exact data were not given. During the second outbreak in 1973-1974 (influenza A-H3N2), the attack rate was indeed higher in groups with multiple vaccinations than in those with single vaccination, but the difference was not significant (F-2, Table 1 and Table 2). A simultaneous influenza B outbreak could not be evaluated because of the lack of a group with multiple vaccinations. The last influenza A outbreak in 1975-1976 also lacked an appropriate group with multiple vaccinations. In conclusion, the Hoskins paradox cannot be substantiated by Hoskins's own data.¹⁸

Taken together with the results of this review, the Hoskins paradox has no basis in reality and should not influence the decision whether to consider annual influenza vaccination.

Another clinically relevant issue is whether there may be differences between vaccine-related reactions or adverse events between the first and annually repeated vaccinations. In the articles reviewed here, no such data were given. In a previous article,⁴⁵ no such differences were found in 1800 adult vaccinees (16-94 years of age). Thus, it appears that the current inactivated influenza vaccines are safe and well tolerated when used annually.

Accepted for publication April 13, 1998.

This study was supported by the Foundation of Respiratory Virology, notably Influenza, Rotterdam, the Netherlands.

We thank Janet E. McElhaney, MD, PhD, Edmonton, Alberta, for kindly providing original data; Nic Masurel, MD, PhD, Leiden, Jan C. de Jong, MSc, PhD, Bilthoven, and Rob Diepersloot, MD, PhD, Utrecht, the Netherlands, for their valuable comments; and Conny Kruyssen for help in preparing the manuscript.

Reprints: Albert D. M. E. Osterhaus, DVM, PhD, Institute of Virology, Erasmus University of Rotterdam, PO Box 1738, 3000 DR Rotterdam, the Netherlands (e-mail: osterhaus@viro.fgg.eur.nl).