Using Surveillance to Evaluate Influenza Vaccine Effectiveness

Influenza vaccine is unique in that it is reformulated annually to include the viral strains predicted to be the most likely to circulate during the coming influenza season. During years when the predominant circulating strain is antigenically dissimilar from the corresponding vaccine strain, the effectiveness of trivalent inactivated influenza vaccine (TIV) may be diminished, but this variation is unpredictable, and during those years the benefit derived from TIV is difficult to estimate reliably. The question of vaccine effectiveness (VE) against mismatched strains is important, because in the United States the influenza A/H3N2 vaccine strain has represented fewer than 25% of the circulating H3N2 strains during 4 of the past 10 years, and for B strains there has been a lineage-level mismatch between the vaccine and circulating strains during 4 of the 7 years since 2001, when both B lineages (Victoria and Yamagata) have circulated in North America.

In addition, although TIV has been used routinely for many years, a number of important questions have not been addressed completely. For example, there is limited information from placebo controlled trials on the efficacy of TIV in key groups, such as infants, older adults, and persons with chronic illness, or its efficacy against serious outcomes, such as pneumonia. In theory, additional clinical trials could be conducted, but in reality such trials are costly, are necessarily limited in size and scope, and, at least in the United States, can only involve healthy adults 18–49 years of age who are not targeted for routine vaccination.

In this issue of the Journal, there are 2 reports of a methodology that shows promise as a means to provide relevant information regarding VE against laboratory-confirmed influenza infections on an ongoing basis [1, 2]. The 2 studies have important differences, but both used surveillance systems to identify persons who had influenza infection confirmed by testing of respiratory specimens. Influenza VE was then estimated by comparing the vaccination status of that group with that of “test-negativecontrols,” defined as persons with a respiratory illness who tested negative for influenza virus.

The first year of the 3-year study reported by Belongia et al., conducted among residents of Marshfield, Wisconsin, was a mismatch mismatch year [1]. The predominant circulating strain during the 2004–2005 influenza season was theA/H3N2strain A/California/7/2004-like, which was characterized by the Centers for Disease Control and Prevention (CDC) as not antigenically similar to the A/Fujian/411/2002-like(H3N2) component of the 2004–2005 vaccine [3], and this drift strain accounted for 95% of the influenza viruses recovered from the Wisconsin study participants (table 1). During that year, the prevalence of vaccination among the influenza-positive case subjects did not differ significantly from that among the test-negative controls, and VE was not documented.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Table 1.

Summary of recent studies reporting influenza vaccine effectiveness (VE) against laboratory-confirmed influenza infection.

Coincidentally, a randomized placebo controlled trial of TIV among healthy adults in Michigan was also conducted during the 2004–2005 and 2005–2006 seasons [4, 5]. During the 2004–2005 season, the A/California/7/2004-like strain represented 44% of influenza viruses recovered from study participants, with the remainder accounted for by B strains. In that trial, TIV VE was documented for both influenza A and B strains (table 1).

The VE estimates reported by the Wisconsin observational study and the Michigan randomized trial for the 2004–2005 season are substantively different, in that the Michigan trial documented VE against the drift influenza A/H3N2 strain, whereas the Wisconsin study did not. Because the Wisconsin study was not randomized, one explanation is that bias influenced those findings. It is worth noting, however, that there were other potentially important differences between the Wisconsin study and the Michigan trial. The Wisconsin study population included persons ⩾6 months of age with an indication for influenza vaccination; 24% of participants were <5 years of age, and 70% of participants had a chronic medical condition. In contrast, the Michigan trial enrolled healthy adults. The Wisconsin study identified persons with respiratory illness who presented for medical care within 10 days after illness onset and who consented to participate, whereas the Michigan trial conducted active surveillance to identify trial participants with a respiratory illness to be targeted for testing. Thus, if VE varies in relation to factors such as age, the presence of chronic medical conditions, clinical presentation, or illness severity, these differences between the 2 studies could potentially account, at least in part, for the variation in the reported results. This suggests that influenza VE estimates obtained from surveillance studies could vary from those obtained from randomized clinical trials because of factors other than residual bias.

In the 2005–2006 season, the results reported by the Wisconsin study and the Michigan trial were more concordant, despite substantial differences in the distribution of influenza strains recovered in the 2 studies. Vaccine effectiveness was not documented in either study (table 1). In the Wisconsin study 62% of the influenza isolates were B/Victoria, a lineage-level mismatch to the B/Yamagata vaccine strain for that year, and most of the remainder were A/Wisconsin/67/2005-like(H3N2), a drift strain characterized by the CDC as not antigenically similar to the A/California/07/2004 -like(H3N2) vaccine strain [6]. In contrast, nearly all (96%) of the strains recovered in the Michigan trial were A/Wisconsin/67/2005-like(H3N2). So, in this year, VE was not documented against drift strains, although influenza activity was relatively low, limiting the number of cases of influenza infection identified in either study.

The 2006–2007 season offered the opportunity to compare the results from the Wisconsin study with those from the study by Skowronski et al. [2], which are based on data from a Canadian sentinel physician surveillance system, although this comparison also involves assessment of VE against very different distributions of influenza viral strains. The extent of the geographic variation in influenza strain distributions—between the Wisconsin study and the Michigan trial, between the Wisconsin and Canadian surveillance studies, and between the surveillance studies and the characteristics of strains identified by United States national surveillance—is one of the more surprising findings to emerge from the detailed viral analyses reported by the surveillance studies. During the 2006–2007 season, the predominant strain recovered in the Wisconsin study was the vaccine H1N1 strain, but this strain accounted for only 18% of strains recovered in the Canadian surveillance study and ∼44% of strains recovered by US surveillance [7]. In contrast, H3N2 strains accounted for 71% of isolates in the Canadian study, approximately half of which were A/Brisbane/10/2006 strains that were not antigenically similar to the vaccine strain, but H3N2 strains accounted for only a minority of isolates identified in the Wisconsin study and by US surveillance.

Despite differences in the strain distribution, the estimate for VE against the predominant vaccine-matched strains identified from the Wisconsin study (52%) was similar to the overall estimate of VE reported by the Canadian study (46%). The number of cases of confirmed influenza infection identified in the Canadian study allowed estimation of component-specific vaccine effectiveness (table 1), and the variation in estimated effectiveness across the 3 vaccine components suggests that geographic variation in influenza strain distribution could also result in geographic variation in the proportion of influenza cases prevented by vaccination. In the Canadian study, VE was documented for the H3N2 strains that included the drift strain, but was not documented for the lineagemismatched B strains.

VE against influenza infection, as confirmed by detection of virus in respiratory specimens, is of particular interest with respect to older individuals, given the paucity of data from randomized trials on this endpoint in older adults, the discussion surrounding estimationof VE against nonspecific end-points in this population [8], and recent evidence that vaccination does not appear to reduce the risk of community-acquired pneumonia for adults ⩾65 years old [9]. Unfortunately, relatively few cases were identified among persons ⩾65 years of age in either surveillance system, which precluded age group-specific estimates. Older adults may have been relatively underrepresented in these studies if influenza attack rates are lower in this group, if viral shedding is relatively limited in magnitude or duration, or if older adults tend to present for care later in the course of illness.One approach that has successfully identified a relatively large number of older people with confirmed influenza illness is the establishment of influenza surveillance in a large hospital network, as recently reported by McGeer et al. [10]. In that 21-hospital network in Ontario, Canada, approximately 227 older adults with influenza infection were identified over 2 influenza seasons, and 82% of that group had received influenza vaccination. The study was designed to evaluate the impact of antiviral therapy and an assessment of influenza VE was not reported, but the high vaccination coverage among the case subjects suggests that VE may not be confirmed in a formal comparison.

The methods used for estimating VE that have been established in Wisconsin and Canada provide useful tools for gaining important information on vaccine benefit in the face of ongoing variation in circulating strains and changes in the populations targeted for vaccination. The results available to date suggest that VE against antigenically mismatched strains is variable and that overall VE may differ by geographic region if there are significant differences in the distribution of influenza strains. There are limitations to the surveillance approach, one of which is that relatively large sample sizes are needed to allow analyses to be stratified by potentially important factors, such as age and influenza type. In the absence of such stratification, findings in one subgroup may drive results that are inappropriately generalized to all groups. Lastly, it should be recognized that if more widespread use of influenza vaccines leads to reductions in viral transmission, influenza attack rates will decrease among both vaccinated and unvaccinated groups, which will impair the ability of studies conducted at the individual level to identify the protective effect of vaccination, if present. In that circumstance, alternative approaches may be needed if researchers are to continue to estimate the benefit of vaccination.

• Received October 8, 2008.
• Accepted October 8, 2008.

• © 2008 by the Infectious Diseases Society of America