Although inactivated influenza vaccines are recommended for many populations, estimates as to their true efficacy vary widely. It is commonly held that the efficacy of TIVs is 60 to 80% in healthy younger adults and similar in children over the age of 2 years, but less in the elderly (particularly the chronically-ill or institutionalized) [25–29]. The degree of antigenic match between the vaccine and circulating virus strains, the exact illness endpoint used, and the surveillance methodology in clinical trials can influence efficacy estimates. Efficacy data concerning children under 2 years of age are very limited.
This placebo-controlled, randomized study was conducted to assess the efficacy, safety, and immunogenicity of a TIV over the 2005 2006 (Season 1) and 2006 2007 (Season 2) influenza seasons in healthy adults in the US. The primary endpoint was protection against VMCCI, as requested by US regulatory authorities, rather than the frequently cited endpoints of CCI or LCI. Although the LCI endpoint has been questioned by some authors , it is still widely used, and constitutes the primary endpoint for the majority of the published TIV efficacy experience.
Despite a good degree of 'vaccine-match' among the influenza isolates in the trial (75.5%), and a strong immune response to the vaccine in both years, the average efficacy over Seasons 1 and 2 of TIV against VMCCI was only 46.3%, with a one-sided 97.5% CI lower bound of 9.8%. While the vaccine was clearly efficacious relative to placebo (one-sided 97.5% CI lower bound for efficacy excluded 0), these results did not satisfy the pre-defined criterion for success, i.e. exclusion of efficacy of ≤ 35% (as evidenced by a lower 97.5% CI of >35%). The results were similar for CCI, with an average efficacy of 49.4%, with one-sided 97.5% CI lower bound of 20.3%, the CI becoming narrower due to the greater number of cases.
Efficacy was higher for the LCI endpoint, at 63.2%, with one-sided 97.5% CI lower bound of 48.2%; this is generally consistent with previously reported TIV efficacy estimates in studies using this endpoint [12, 14, 30]. While an intrinsic bias of the LCI endpoint in favor of the vaccine has been noted [15, 16] and, therefore, limits its utility for regulatory purposes, it should also be noted that this endpoint may improve sensitivity relative to culture. In addition, the endpoint of LCI has been used in many prior clinical trials of TIV, yielding estimates of TIV efficacy in adults of about 70% [14, 15]. Thus, the estimated vaccine efficacy against LCI in this trial is consistent with that estimated in prior trials of other TIVs.
A key factor influencing efficacy findings in influenza vaccine studies is the disease attack rate, and the unpredictable nature of epidemic intensity poses a challenge when planning influenza vaccine efficacy trials [12, 13, 25, 31]. Our study was powered on the assumptions that vaccine efficacy would approximate 70%, based on the prior literature reporting culture-confirmed influenza endpoints, and that the attack rate for CCI in the placebo group would be 2.0% across both seasons. The attack rate assumption was adjusted in Season 2 to 1.6% based on the regulatory request to adopt VMCCI as the primary endpoint, and the sample size correspondingly increased. The actual VMCCI attack rate in placebo recipients in Season 1 was close to our prediction, but in Season 2 the attack rate was less than half of that predicted, and this had a large negative impact on the precision of the average efficacy estimate. Despite this, it was also clear that the efficacy observed did not differ markedly between the seasons, and might not have sufficed to exclude the 35% lower bound of the primary hypothesis even if the VMCCI attack rates in Season 2 had matched those in Season 1.
Similar to the findings of our study, the Centers for Disease Control and Prevention surveillance data summaries for the 2005-2006 and 2006-2007 influenza seasons in the US suggest that these were relatively mild influenza seasons [23, 24]. The weekly percentage of outpatient visits to US sentinel providers for ILI peaked twice in the 2005-2006 season at 3.3% and 3.2% (baseline 2.2%), and peaked twice in the 2006-2007 season at 3.0% and 3.5% (baseline 2.1%) [23, 24]. In the prior three seasons, the peak percentage of outpatient visits for ILI had ranged from 2.3% to 7.6% [23, 24].
A negative impact of low attack rates in a given season on point estimates of efficacy has been well documented. Previous studies of a TIV product, involving the same investigators, methods, and source populations over successive seasons, have reported dramatic differences in efficacy estimates between seasons, without obvious explanation based on circulating strain match to the vaccine [12, 13]. In these studies, the low efficacy estimates were found in seasons with lower overall attack rates [12, 13]; a similar experience has occurred in pediatric trials . While a clear reason for this phenomenon is not apparent, it can be speculated that low background attack rates may result from either relatively high pre-existing population resistance to the circulating strain(s), or modest intrinsic virulence of the circulating strains, such that only intense exposures lead to disease. In either case, the attainable impact of a vaccine might be blunted. Primary detection of influenza virus infections in this study was by means of viral culture; with the use of molecular methods confined to characterization of isolates. Other reports have suggested that the use of RT-PCR as a primary detection tool may improve the rate of ascertainment of influenza virus infections by 20 to 30% [12, 13, 31, 32], but this maneuver does not appear, in and of itself, to have a major impact on point estimates of efficacy.
As suggested previously, a further factor that may compromise vaccine efficacy is the extent of antigenic drift and the degree of match between vaccine and circulating viruses [11, 12, 33]. In our study, the majority of cases in both seasons were due to influenza A/H3N2 viruses, and antigenic drift was modest and had little apparent impact. However, only 5/17 cases of culture-confirmed influenza B disease were due to viruses of the same lineage as the vaccine strain in the relevant season (data not shown), and average efficacy against non-matching B strains was low (~16%); this result clearly reduced the estimate of efficacy against all CCI.
The definition of influenza-like illness (ILI) in our study required that illness impeded normal daily activities, with cough plus ≥ 1 other symptom from a panel including both respiratory and systemic complaints, so this allowed for a diagnosis without need for a systemic symptom (i.e. fever, myalgia/arthralgia). Cough has been previously found to be one of the most sensitive clinical indicators of influenza, but has little specificity, whereas fever is a classically-cited characteristic of influenza, with substantial positive predictive value, but lower sensitivity . Fever was not made a requirement for ILI in our study in order to maximize sensitivity. However, our ad hoc analyses showed that narrowing of the case definition to require fever reduced the case rate but notably improved the apparent efficacy of TIV. A potential interpretation of this finding may be that TIV is most effective against severe manifestations of influenza, and less so against illnesses with predominantly respiratory symptoms. Although efficacy against severe disease is desirable, the relatively sensitive but non-specific case definition in our study may have allowed for the inclusion of respiratory illness caused by mixed infections in which influenza viruses may have been minor or coincidental. An additional factor that could have compromised the results is the highly decentralized source of the specimens, and the transportation of specimens to a centralized laboratory may not have been optimal for the recovery of influenza viruses from nasopharyngeal swabs.
An unexpected finding of the study was the difference between men and women for the primary endpoint: men, 89.0%; and women, 19.4%. Women were slightly older than men (33.3 and 31.7 years, respectively), and were more likely to have had previous recent influenza vaccination (21% and 16%, respectively). These between-gender differences were reflected in both treatment groups, which were well matched for all baseline characteristics. In addition, all three viruses were culture-confirmed in men and women, there was no differential temporal clustering of cases, and serum HI titers did not differ meaningfully between the sexes (data not shown); this is in contrast to a previous large study of a TIV, which reported significantly higher GMT responses in women than men, regardless of age or dose or influenza strain . The attack rate of VMCCI among placebo-treated men and women was similar (1.27% and 1.14%, respectively), which suggests that both sexes had a similar level of exposure. At present, the difference in values between men and women remains unexplained and must be viewed with caution because a further subdivision of the already-small case numbers results in very broad CIs about the efficacy estimates.
In terms of seroconversion and rates of attainment of post-vaccinal HI reciprocal titers ≥ 40 ("seroprotection" rates), the TIV in this study fulfilled the immunogenicity criteria for the accelerated approval of seasonal influenza vaccines established by the US FDA Center for Biologics Evaluation and Research for all strains in both seasons . In addition, manufacturing consistency was confirmed for three consecutive TIV lots in the 2005-2006 season, based on GMT values. Vaccine immunogenicity as measured by induction of hemagglutination-inhibiting antibodies provides a marker of potential vaccine effectiveness in the absence of viral circulation and has therefore been, and continues to be, a useful tool for the registration of influenza vaccines. Nevertheless, definitive demonstration of a single antibody titer which constitutes an absolute correlate of protection for all virus strains remains elusive, and the current study was not designed to demonstrate this.
The reactogenicity events reported in the study were consistent with those commonly reported with TIVs. Reactogenicity events that were significant in the TIV versus placebo group included injection site pain, injection site redness and swelling, myalgias, arthralgias, fever and fatigue; the majority of events were mild (grade 1) in severity. There was a slightly higher incidence of spontaneous AEs in the TIV versus placebo group reported up to 21 days post-vaccination, and this was primarily due to the persistence of injection site pain and redness. Overall, the results suggest that the safety profile of TIV was acceptable, and consistent with the historical performance of similar products.