Challenge of conducting a placebo-controlled randomized efficacy study for influenza vaccine in a season with low attack rate and a mismatched vaccine B strain: a concrete example

Influenza is a highly contagious infectious disease resulting in acute respiratory illness in people of all ages. Annual epidemics occur worldwide and cause substantial morbidity and mortality [1, 2]. Influenza poses a particular risk to the elderly and also to people suffering from conditions such as chronic heart or pulmonary disease. The causative agents are influenza A and influenza B viruses and the main virulence factors are the virus surface coat proteins hemagglutinin (HA) and neuraminidase (NA). There are several antigenic forms of HA and NA for influenza A which is classified into subtypes based on different combinations of these antigens [1, 3, 4]. Only a few of these influenza A subtypes have ever been associated with human disease and the subtypes currently in circulation in human hosts are H1N1 and H3N2 [5]. The influenza B virus currently belongs to two evolutionary lineages that are distinct at the genetic and antigenic levels and which are represented by B/Yamagata/16/88-like and B/Victoria/2/87-like viruses that have co-circulated in the population since the mid-1980s [5‐8]. In order to evade the host immune system, the HA and NA proteins of both influenza A and influenza B viruses undergo continuous mutation and by this mean evade the host immune system. This is known as antigenic or genetic drift [1, 5, 9, 10].

Influenza vaccination has been employed for many years as the primary tool to prevent influenza virus infection and its complications [2]. As recommended by the World Health Organization (WHO), vaccines are trivalent containing two influenza A strains (H1N1 and H3N2) and one influenza B strain [1]. However, to ensure efficacy against new drift variants the vaccine strains must be updated on an annual basis for both the Northern and the Southern hemisphere [11]. Based on epidemiology and phylogenetic analysis of HA and NA sequences of the circulating human strains detected though a global influenza surveillance network, WHO recommends the three strains that are anticipated to become dominant during the next influenza season [11]. Although in most years the recommendations accurately predict a close antigenic match between the vaccine and circulating strains, occasionally a predominant circulating strain turns out to be antigenically different from the corresponding vaccine strain. As two influenza B virus lineages co-circulate, the current recommendation to include only one lineage in each year's TIV poses a particular risk for a mismatch. Vaccine strain mismatch can have a negative impact on vaccine efficacy [9, 10, 12, 13].

In this paper we describe an efficacy study conducted with a trivalent inactivated split-virus influenza vaccine (TIV) manufactured by GlaxoSmithKline Biologicals which has been available since 1987 and is now used in over 100 countries [4, 14, 15]. As for other inactivated influenza vaccines, the ability of TIV to induce HA antibody levels (as measured in hemagglutination-inhibition (HAI) assay) that meet regulatory authority criteria has been accepted as a surrogate marker for efficacy [4, 14, 15]. Recently, this TIV vaccine was approved under the United States' accelerated approval regulation [4, 15]. To fulfill the requirements of this regulation, post-approval demonstration of clinical benefit in an adequate and well-controlled clinical trial is required.

We designed a placebo-controlled randomized efficacy study with a primary clinical endpoint based upon rates of culture-confirmed influenza A and/or B disease in a target population of healthy adults with a broad age range of 18 to 64 years. All influenza vaccine efficacy studies are subject to the unpredictable nature of the intensity of the influenza season in which they are conducted as well as the extent of the antigenic match between the circulating and vaccine strains [16]. As illustrated by the outcome of this present study, these factors can have a substantial negative impact on the assessment of vaccine efficacy.

This paper describes the outcome of a placebo-controlled randomized efficacy study with a trivalent inactivated split virus influenza vaccine (TIV) in healthy adults aged 18 to 64 years. The immunogenicity of TIV was consistent with previous studies in adults [4, 15] and seroprotection rates exceeded regulatory authority criteria. Despite this, the vaccine efficacy was not confirmed.

Influenza vaccine efficacy in a particular region depends on the nature of the influenza season. This can vary from year to year and as it cannot be predicted in advance, poses a particular challenge for the conduct of a prospective efficacy study. The optimal circumstances in which to evaluate the efficacy of influenza vaccines are i) high viral circulation (i.e. high attack rate) and ii) when the vaccine strains are antigenically-matched to the circulating virus types [16, 24]. The 2005–2006 influenza season in the Czech Republic was exceptional in that both of these factors turned out to be suboptimal for the conduct of an efficacy study.

The first issue faced by the study was the very low attack rate. The study was powered based on the assumption that the attack rate of culture-confirmed influenza in the placebo group would be 4% while the actual attack rate during the study was 0.9%. Literature data are limited on the attack rates of culture-confirmed influenza cases in an unvaccinated adult population and serology has been very often used alone or combined as a diagnostic tool to confirm the influenza virus etiology of a clinical case. The attack rate of 4% was hence selected as a conservative estimate based upon data from randomized placebo influenza vaccine studies in healthy adults with laboratory (by serology or a mixture of serology and culture) confirmed endpoints reviewed in the Cochrane data base [3]. These indicated an attack rate range of 6% to 16% in the placebo groups. More recently, prospective placebo-controlled trials have reported rates of culture-confirmed cases in the adult setting [25, 26] as well as in children [27]. These studies clearly confirmed the high year-to-year variability of documented influenza cases.

The low number of culture-confirmed influenza cases found during our study was unlikely to be due to inadequate monitoring of prospective cases, as active surveillance for ILI involving contact of subjects by phone on a biweekly basis was employed. Also, all nasal and throat swab specimens were transported to the laboratory with an adequate cold chain within 24 hours to avoid false culture negatives due to loss of virus viability. Furthermore, the low attack rate observed in the study area in 2005–2006 was confirmed by the European Influenza Surveillance Scheme (EISS) [28]. Specific data for the Czech Republic demonstrate that the weekly morbidity rates for both acute respiratory infections [29] and for ILI [30] were lower in the 2005/2006 influenza season compared to both the previous 2004/2005 season and also the following 2006/2007 season. The EISS also reported that the circulation of influenza B virus in the winter 2005–2006 was exceptional compared with European data accumulated during the last decade, as it was the only season in which influenza B viruses were dominant [28]. It is possible that this may have contributed to the low attack rate of clinical illness as influenza B is known to cause milder disease than influenza A [28, 31].

The significant impact that a lower attack rate can have on a clinical trial to evaluate influenza vaccine efficacy has previously been illustrated in a clinical trial conducted by Hoberman and colleagues in young children over two consecutive influenza seasons in Pittsburg [27]. In the first season (1999/2000) when the attack rate for culture-confirmed influenza was 15.9% in the placebo group, vaccine efficacy was estimated to be 65% (95% CI 34% to 82%) while in the second season (2000/2001) when the attack rate for culture-confirmed influenza was 3.3% in the placebo group, vaccine efficacy was estimated to be -7% (95% CI -247% to 67%). A similar finding was recently reported in a study conducted over 2 years in an adult population in Michigan [25, 26]. The attack rate of culture-confirmed influenza in the placebo group was reported to be 5.8%, whereas in the second year, a rate of 1.8% was found. The VE estimates for the inactivated influenza vaccine were found to be 77% (37% to 92%) in year 1 and 23% (-153% to 73%) in year 2. The latter result appears to be similar to the outcome of our study in adults where with an attack rate in the placebo group of 0.9% for influenza A and/or B cases and 0.2% for influenza A cases, the respective estimate of efficacies were 22.3% (95% CI -49.1% to 58.5%) and 25.1% (95% CI of -260.9% to 82.2%).

These data indicate that the possibility of a lower than predicted attack rate needs to be factored into the design of influenza vaccine efficacy studies. Apart from increasing the sample size or allowing for an extension of the study over more than one influenza season, another factor to be considered is the clinical case definition used. There is no universally accepted clinical case definition of influenza [32, 33]. We employed a definition from the Centers for Disease Control and Prevention (CDC) US based on the presence of fever plus either cough or sore throat, [17]. This combination of symptoms is recognized as having a high positive predictive value for influenza but is not very sensitive [34]. For seasons with low attack rates it may be preferable to improve case capture by using less restrictive clinical criteria particularly when culture confirmation is being employed to confer specificity. The use of reverse polymerase chain reaction (RT-PCR) assays techniques, which have been shown to be more sensitive than culture, may also improve detection of influenza virus in nasal/throat specimens [35, 36]. Serological analysis of acute and convalescent sera, which is also often used to confirm influenza cases, would however have specific limitations in a vaccine efficacy study. In the vaccine group, the presence of vaccine-induced antibodies could mask differences between paired sera in levels of antibodies induced by infection, thereby introducing a bias s in favor of the study vaccine.

The low attack rates encountered in the adult population during some influenza seasons may raise the question about the need for adult vaccination. It should be kept in mind that currently the main target populations for adult vaccination (apart from the elderly persons, which are outside the scope of this study) are adults who have medical conditions that place them at risk for complications from influenza and also, in order to limit influenza virus transmission to those at risk in all age groups, health-care personnel or caregivers [2]. As the attack rate for any coming influenza season is unpredictable at the time of annual influenza vaccination, it is not feasible to modify vaccination routines and put susceptible people at risk.

The second issue faced by the study was a mismatch between the dominant circulating influenza strain and the vaccine strains. The majority of culture-confirmed cases detected in our study were caused by an influenza B strain that was of a different lineage to the vaccine influenza B component which was from the B/Yamagata/16/88-lineage. This mismatch was confirmed by the EISS who reported that 90% of the characterized influenza B viruses in Europe during the influenza season 2005–2006 were similar to the B/Victoria/2/87 lineage of influenza B viruses [28]. As a consequence, the World Health Organization (WHO) substituted a virus from the B/Victoria/2/87 lineage in the 2006–2007 vaccine recommendations [37].

The co-circulation of these two antigenically distinct B lineages which form distinctly divergent genetic groups based on their hemagglutinin genes where there are some 27 amino acid differences [8] has raised the question of whether they should both be represented in the annual influenza vaccine formulation [38]. To date, this option has not been considered due to the constraints it would impose on an already limited influenza antigen manufacturing capacity. Also, although there is no hemagglutinin cross reactivity between the two B lineages in ferrets [7] or unprimed children [39], serological evidence from WHO [37] and another recent study [40] indicates that vaccine B/Yamagata/16/88-lineage components may provide reduced protection against the B/Victoria/2/87 lineage. Furthermore, reassortment between the two influenza B lineages has led to practically all B/Victoria-lineage viruses containing a B/Yamagata-lineage neuraminidase which may also provide some cross protection [13, 38]. Other workers have managed to demonstrate vaccine efficacy for influenza B despite mismatch between vaccine and circulating strains [13, 25]. In the Michigan study, Ohmit et. al demonstrated efficacy against influenza B during a mismatch of B strains in the 2004/2005 season [25]. However, the attack rate for culture-confirmed influenza in the placebo group in that study was considerably higher (5.8%) than in the study reported here and also a greater proportion of the influenza B isolates did match the vaccine strain (7 out of 18).

Although the exceptional nature of the influenza season in which this study was conducted did not allow us to establish the efficacy of TIV it is likely that, as demonstrated by Ohmit et al for another inactivated vaccine [25], TIV efficacy could be demonstrated under more typical influenza season conditions.

Due to the exceptionally low attack rate during the 2005/2006 season in the Czech Republic and the predominance of a circulating B strain of a different lineage to the vaccine strain, our study was unable to demonstrate the efficacy of TIV in healthy adults. This experience illustrates the challenge of conducting an influenza vaccine efficacy trial during one season when variations in disease attack rates and circulating strains cannot be predicted in advance.