Children have high rates of seasonal influenza infection and illness, and children less than 2 years of age are at high risk of influenza-associated hospitalization, with infants being at the highest risk.1 In addition, preschool children have been recognized as a major reservoir for influenza virus circulation in the community.2 Although many countries now recommend routine vaccination of children with seasonal influenza vaccine, data on the efficacy of trivalent inactivated influenza vaccine (TIV) in young children are limited. It would be very useful if protective efficacy in children could be predicted from the immune response induced by influenza vaccines.
Antibodies directed against hemagglutinin as measured in the hemagglutination inhibition (HI) assay are generally recognized as being an essential component of protection against influenza infection. That there appeared to be a correlation between HI antibody to the circulating strain and risk of infection was noted by Salk and Suriano in 19453 in a comparison of 2 vaccines in military recruits and again by Meikilejohn et al in 1952 in a study of hospital personnel.4 Although an HI titer of 1:40 has been determined as corresponding to a 50% reduction in the risk of contracting influenza in a susceptible population and has been used as an immunologic correlate of protection,5,6 this relationship was defined in adults largely based on controlled challenge studies and may not be generalizable to children. Children have a reduced capability for cellular immunity and may lack prior experience with influenza virus infection or previous vaccination; therefore, a correlate may be different in children.
To evaluate potential correlates of protection in children, studies evaluating efficacy against proven influenza infections during the pediatric age are required. Novartis Vaccines has conducted a randomized, double-blind, 3-arm, comparative trial of adjuvanted influenza vaccine (Fluad, Novartis Vaccines, Siena, Italy), TIV, and placebo in more than 4700 children between 6 months and 6 years of age. Included in this trial was a subset of 777 children in whom pre- and postvaccination HI titers against influenza were measured. This subset was used to evaluate the relationship between HI antibody titer and clinical protection from influenza in this age group.
The detailed methods for the comparative efficacy trial are presented elsewhere.7 Briefly, 4707 influenza vaccine–naive healthy children between 6 and 72 months of age were randomized in a ratio of 2:2:1 into 3 groups to receive 2 doses 1 month apart of MF-59–adjuvanted inactivated influenza vaccine (ATIV, Fluad), subunit TIV (TIV control, Influsplit, GSK, Rixensart, Belgium), or a saline placebo during the 2007 to 2008 and 2008 to 2009 influenza seasons. During the 2007 to 2008 influenza season, the influenza vaccines contained influenza strains A/Solomon Island/3/2006(H1N1), A/Wisconsin/67/2005(H3N2), and B/Malaysia/2506/2004. During the 2008 to 2009 influenza season, the influenza vaccines contained influenza strains A/Brisbane/59/2007(H1N1), A/Brisbane/10/2007(H3N2), and B/Florida/4/2006. Cases were identified through active surveillance for influenza-like illness and subsequent case confirmation by reverse transcription polymerase chain reaction testing for influenza. Serologic results were obtained for a subset of 777 children at enrollment, on day 29 when the second dose was administered, at day 50 (approximately 3 weeks after dose 2), and at day 180 after receipt of 2 doses of vaccine or placebo. The HI for seasonal influenza strains was performed according to standard procedures as described by Palmer et al.8 All sera were pretreated with a receptor-destroying enzyme, heat-inactivated and diluted to a starting dilution of 1:10. On V-bottom–shaped, 96-well microtiter plates, pretreated sera were then serially diluted 2-fold. The same volume of virus suspension (whole active virus, 4 hemagglutination units [HAU] as determined directly before use) were added to all wells and incubated at room temperature for 1 hour. Subsequently, chicken red blood cells (0.5% suspension in phosphate-buffered saline) were added to each well, and plates were incubated again for 1 hour at room temperature. For titer reading, plates were tilted on a light screen. The highest dilution in which a “nose” could be seen was read as the titer. The HI titer was reported as the last dilution step, in which agglutination was still completely inhibited. As all sera were tested in duplicate, the final titer was the geometric mean of 2 readings. Reference sera from the National Institute for Biologic Standards and Control were used as control standards.
The availability of efficacy data for both TIV and ATIV as well as the day 50 HI immunogenicity data after 2 doses allowed evaluation of a correlate of protection for HI titer in this age group. In this analysis, the efficacy data from the entire cohort was used, whereas immunogenicity data only came from the subset of children on which immunogenicity data was available. As almost all influenza cases were caused by A/H3N2 strains, the analysis was confined to antibodies against A/H3N2.
The Prentice criterion is generally used to determine whether an immunologic correlate of protection can be determined.9 The criterion requires that the observed vaccine effect be explained in a statistical model using immunologic data. Accordingly, a linear logistic regression model was fitted with vaccine group only included as independent predictor and occurrence of influenza as dependent variable. A second logistic regression model was fit, which controlled for log2 day 50 antibody titer and vaccine group to determine the effect of antibody titer on the occurrence of influenza.
Then, the relationship between the occurrence of influenza and H3N2 antibody titer level was modeled using the logistic regression model advocated by Dunning that accommodates both antibody titers and factors independent of antibody titers.10 In this model, the probability that a subject develops influenza is the probability that the subject is susceptible multiplied by the probability that susceptible individuals develop disease. Susceptibility is characterized by the probability λ, and the probability that a subject with titer t is protected is represented by a 2-parameter logit function, with α and β denoting the location and the scale parameters of interest, respectively.
In the past, the HI correlate of protection has usually been defined as the antibody level at which the probability of clinical protection is 50%.11 However, as 50% protection is lower than the desired public health effect, further analyses were conducted to determine the level of antibody associated with 60%, 70%, 80%, and 90% protection.
Finally, as Hobson et al defined the immunologic correlate in adults in challenge studies, which used the HI antibody titer at the time of virus exposure to define a correlate, we developed a corresponding model to estimate a correlate of protection at the time of “challenge” in this population. For the subjects in the immunogenicity subset in whom an influenza A/H3N2 infection was identified, the day 50 titer was used to estimate the titer at the time the diagnostic viral swab was obtained using antibody decay kinetics derived from study children. For each of these cases, up to 2 noncase controls were randomly chosen from the immunogenicity subset who were of the same age (±28 days), same sex, and had the same day-50 blood draw date (±28 days). Using a half life of 111 days (95% confidence interval [CI], 90–123) and the measured titer at day 50, the titer at the time the swab was obtained was estimated assuming a logarithmic decay of H3N2 antibodies. The same was done for controls using the date of the swab of the matched case as the index date. These data were then used to fit a protection curve at the day of swab again using the method of Dunning. This model was run 1000 times for each half life (for the point estimate and upper/lower bounds). Resampling was performed each time with up to 2 controls being randomly selected, and then the protection curve was fit. Titers related to 50% and 80% protection were then estimated.
The age, sex distribution, and ages at vaccination of the immunologic subset did not differ significantly from the overall cohort. Younger children were much more likely to be seronegative before vaccination than older ones. The HI titers, vaccine efficacies, and the corresponding 95% CIs for the 3 study groups are shown in Table 1. Of note is that the geometric mean titer of antibody attained in the ATIV group of 1:746 was substantially higher than the 1:92 titer in the TIV-control group. Also of note is that the titer of 1:92 in the TIV-control group was associated with less than 50% protection in contrast to what would have been expected if the 1:40 adult correlate held for children. In all, 6 placebo recipients, 14 TIV-vaccinated, and 2 ATIV-vaccinated subjects experienced influenza caused by a A/H3N2 strain in the immunologic subset. The distribution of influenza cases by day 50 titer is shown later in the text (Table 2).
To evaluate compliance with the Prentice criterion, the regression models fitting the log2 day 50 antibody titer and vaccine group as independent variables and polymerase chain reaction–confirmed clinical influenza as the dependent variable revealed that the titers were highly inversely (P = 0.0071) related to infection incidence, indicating that the higher the HI titer the lower the risk of acquiring an infection. The effect of the vaccine group, however, changed from significant (P = 0.03) to nonsignificant (P = 0.1349) after controlling for antibody titer, indicating that the antibody titer mediates most of the vaccine effect on incidence of infection (Fig., Supplemental Digital Content 1, http://links.lww.com/INF/A981). Thus, according to the Prentice criterion, HI antibody titer can be considered as correlate of protection based on the data from this study. In addition, the fact that antibody titer alone was a significant predictor of risk of developing influenza and that the relationship between vaccine group and risk of disease was statistically nonsignificant allowed us to combine all of the vaccine groups to estimate the correlate of protection (Table, Supplemental Digital Content 2, http://links.lww.com/INF/A978).
Having established that antibody titer may serve as a correlate of protection, an antibody cutoff level for clinical protection against influenza infection was determined for 50%, 60%, 70%, 80%, and 90% protection using Dunning model. Fitting the data to the H3N2 antibody levels and the respective influenza cases observed in our immunogenicity subset gave the parameter estimates shown in Table, Supplemental Digital Content 3, http://links.lww.com/INF/A980. The fitted curve from the model including 95% CIs are shown in Figure 1, and the titer values associated with 50%, 60%, 70%, 80%, and 90% protection from the model are shown in Table 3. A cutoff HI titer of 1:110 was associated with the conventional 50% clinical protection rate with titers of 1:215, 1:330, and 1:629 predicting protection rates of 70%, 80%, and 90%, respectively. Using the same method, we calculated that the conventional 1:40 HI titer was only associated with 22% protection from clinical infection in these children.
Although the titer measured at a fixed time (50 days) following initial vaccination was used to estimate these correlates, it was recognized that the titer at the time the child was exposed to influenza was more likely to be the important biologic phenomenon. We estimate that a titer of 1:85 (95% CI, 35.6–137.9) at the time of exposure was associated with 50% protection in this model, and a titer of 1:302 (95% CI, 176.9–439.2) was associated with 80% protection at the time of exposure.
The development of a correlate of protection in large population studies in which serologic information is not available on all participants has been undertaken in several situations in the past. Such an approach is necessary in the situation where the population size is large and the disease attack rate is relatively low making serologic assessment in all participants impractical. One example is the evaluation of PRP polysaccharide efficacy in the Haemophilus influenzae type b trial conducted in Finland in the 1970s. In this trial, more than 100,000 children were randomized to receive PRP vaccine or control, but serology data was available on a subset of 514 children, with none of the children in the subset actually having Hib disease. However, by using the efficacy determination in the full cohort and looking at the distribution of serology results in PRP vaccine recipient and controls, it was possible to determine the protective correlate of 1.0 μg/mL of anti-PRP antibody in this study.12 Similarly, in studies of vaccination against invasive disease due to Streptococcus pneumoniae in infancy, a population-based serologic correlate was developed using immunologic data on a subset of patients even though no cases of invasive disease occurred in this subset.13 The development of a protective correlate for pneumococcal conjugate vaccination was also approached by a World Health Organization workgroup using a similar methodology across multiple trials.14 Thus, there are multiple precedents for using immunologic subsets to estimate a correlate of protection in large trials.
Another issue that should be addressed is the difference between a population correlate and an individual correlate of protection.15 Care must be taken to differentiate the two. Here, we are interested in a population correlate based on population estimates of efficacy and immunologic results in a subset of the efficacy population. Such population correlates are useful epidemiologic tools but do not predict individual protection. This is because in the individual situation, a subject with a titer above the protective threshold of the population may be exposed to a very large inoculum of the infecting agent or may have other risk factors that make them more subject to infection. In fact, in development of a population serologic correlate for protection against diphtheria, vaccine failures occurred in which individuals had very high titers against diphtheria toxin.16 This did not negate the validity or utility of the population protective correlate.17 Hence, the fact that vaccine failures occurred in our population in individuals with high titers does not negate the validity of the population correlate of protection. In addition, it should be noted that one might anticipate a population correlate requiring higher levels of antibody in young children because in this age group, there is a reduced capability for cellular immunity as well as a lack of influenza-specific prior immunity due to a lack of prior experience with influenza virus through vaccination or infection.18
Our analyses support the assumption that H3N2 antibody titers may serve as a correlate of protection for influenza infection. Although this analysis was limited to response to H3N2 virus, this approach is the same as was used originally in the seminal study by Hobson et al to propose a correlate from a study in which most cases were caused by a single strain of virus.6 The Hobson et al study data were obtained from a challenge model in which the interval between vaccination and exposure to the virus was controlled. A recent meta-analysis of the relationship of HI and clinical protection in adults supported the use of a 1:40 titer as a correlate of 50% protection in adults, although the supporting data from clinical trials in the meta-analysis were weak.3 However, our data do not support the use of a 1:40 titer as a correlate of 50% protection against influenza infection in children less than 6 years of age. A cutoff of 1:110 measured 21 days after the second vaccine dose may be used to predict a 50% clinical protection rate in this age group.
Although the reason that more antibody was required for protection in children is not known, several factors might contribute to the need for more antibody to provide protection in children. The most important is probably the fact that the younger the age of the individuals the lower the probability that a child has had immunologic experience (induction of cell-mediated immunity, specific memory) with influenza either through vaccination or infection. In young children, most, if not all, immunity detected is likely mediated by the antibodies induced by vaccination. At older ages, other effector mechanisms such as cell-mediated immunity primed by influenza infection and directed against proteins other than those present in a vaccine and hence a lower cutoff antibody levels may be sufficient to confer protection, at least as detected serologically by HI. Of note is that in this study, children were influenza vaccine–naive before enrollment, so any HI antibody present would have been from prior infection. In fact, of the cases of disease in the immunologic subset, all had HI titers <1:10 before vaccination as compared with 75.7% of controls. It is possible that in influenza-naive individuals, a higher titer is required for protection, whereas in individuals with a prior history of influenza infection, other protective factors may come into play that would require less associated HI antibody for a given level of protection. Other factors may include differences in the avidity of antibodies and importantly the likelihood that young children are exposed to higher inocula of virus through close contact and day care center exposure.19 We have also calculated correlates of protection for higher levels of protection than the 50% customarily used. Although LD50, MIC50, and other parameters are commonly used for description of biologic phenomenon, they do not make sense in assessing a public health intervention. Titers of 1:215, 1:330, and 1:629 predicted protection rates of 70%, 80%, and 90%, respectively. This has important vaccine policy implications because these higher levels HI antibody can be attained through vaccination with adjuvanted vaccine, but not TIV, in this age group.
Immunization of young children against influenza is performed both to protect this high-risk group against infection and hospitalization and to reduce the circulation of influenza virus in the population reservoir of infection. If the goal is to protect children against infection, it makes much more sense from a public health perspective to target 80% or 90% protection rather than 50%. To reduce influenza virus circulation in young children and hence to reduce the risk of infection in the elderly and other risk groups, targeting a high level of protection is especially important. Although the efficacy trial on which our estimate is derived did not evaluate shedding of influenza virus but rather clinical disease, studies of influenza infection in children have shown that the amount of shedding (and hence likelihood of transmission) is correlated with the severity of disease.20 Hence, it is interesting to speculate on the impact of the targeted level of protection on transmission. If one achieves 70% coverage for influenza vaccination in young children and the targeted level of protection for the vaccine is only 50%, one would have 65% of the population still at risk for infection and circulating the virus. If one targeted 90% protection and still had 70% vaccine coverage, one would only have 37% or approximately half the number of children at risk for infection and circulating the virus. Thus, targeting a higher level of protection has important public health implications both for the vaccinees and for influenza control in general. However, even if one wishes to evaluate influenza vaccines for use in children using a 50% protection level for comparative purposes, the use of a 1:110 rather than a 1:40 titer would be appropriate.
We have also attempted to estimate the titer associated with protection at the time children were exposed to influenza virus. We estimate that a titer of 1:85 was associated with 50% protection at the time of infection. It is interesting to note that in their article, published in 1982, describing influenza epidemiology in Seattle, Fox et al21 postulated that a titer of 1:80 was required for 50% protection in children. In our study, we estimated that a titer of 1:302 was associated with 80% protection at that time. These estimates are consistent with the variable time between the measurement of the titer at day 50 and exposure, the assumption that antibody levels decay as time passes, and our estimates for a protective correlate at day 50. Although it is of interest to estimate the titer associated with protection at the time of challenge, we recognize that for the evaluation of annual influenza vaccines, the use of a titer obtained at a fixed interval after vaccination, as in Table 3, is standard.
Although the protective correlate defined by Hobson et al was based on a protective correlate defined for a short fixed interval in a challenge study in adults, it has been used by regulatory agencies as a criterion for use of annual influenza vaccines. Even for adults, this is potentially problematic for 2 reasons. The first is that the commonly used correlate was developed to predict 50% protection. This clearly may not be the desirable public health target for protection. Secondly, as the Hobson et al correlate was defined from challenge studies performed shortly after immunization, it does not contemplate that a higher level of antibody would be required to protect the population throughout the influenza season especially in seasons when the onset of influenza in the community occurs late in the season.
We recognize that there are limitations to the use of these data. First, the derived correlate of protection has been derived using only an H3N2 strain. However, data in the initial Hobson et al study was also based on the circulating A strain at that time. It is possible that the correlate for influenza B or even for some A strains may be different. In addition, data from this study were derived only using inactivated influenza vaccines. It is very likely that different correlates, which still need to be defined, may apply for live attenuated influenza vaccines.22 Additionally, there may be subtle differences in the protective correlate for adjuvanted and unadjuvanted vaccines. Increased avidity of antibodies has been observed with MF-59–adjuvanted influenza vaccine as compared with unadjuvanted vaccine in one study.23 Such antibodies would likely provide better protection at a lower concentration and could lead to the underestimation of the correlate for unadjuvanted vaccines in this study. However, that differences between a correlate for adjuvanted and unadjuvanted vaccines, if they exist, were likely to be small in this study is supported by the fact that the Prentice criteria were met and that antibody titer alone was significantly associated with risk of disease whereas vaccine type was not when both variables were included in the same model. Additionally, our estimates were derived from a European population with a specific age distribution and influenza exposure history. It is possible that the correlate may vary by population.
In summary, the analyses above support the assumption that H3N2 antibody titers may serve as a correlate of protection for influenza infection. These data also indicate that the conventional use of the 1:40 adult correlate of protection is not appropriate in children. Although a cutoff titer of 1:110 may be used to predict the conventional 50% clinical protection rate, a titer of 1:330 would predict an 80% protective level, which would seem to be more desirable from a public health perspective.
1. Izurieta H, Thompson WW, Karmaranz P, et al.. Influenza and rates of hospitalization for respiratory disease among infants and young children. N Engl J Med. 2000;342:232–239.
2. Hemingway CO, Poehling KA. Change in recommendation affects influenza vaccinations among children 6 to 59 months of age. Pediatrics. 2004;114:948–952.
3. Salk JE, Suriano PC. The importance of antigenic composition of influenza vaccine and protecting against influenza disease. Am J Hygiene. 1949;39:345–355.
4. Meikilejohn G, Kempe CH, Thalman WG, et al.. Evaluation of monovalent influenza A vaccines: observations during an influenza A epidemic. Am J Epidemiol. 1952;55:12–21.
5. Coudeville L, Bailleux F, Riche B, et al.. Relationship between haemagglutinin-inhibiting antibody titres and clinical protection against influenza: development and application of a Bayesian random effects model. BMC Med Res Methodol. 2010;10:18. Available at: http://www.biomedcentral.com/1471–2288/10/18
6. Hobson D, Curry RL, Beare AS, et al.. The role of serum haemagglutination-inhibiting antibody in protection against challenge infection with influenza A2 and B virus. J Hyg (Lond). 1972;70:767–777.
7. Vesikari T, Knuf M, Wutzler P, et al.. Efficacy of MF59-adjuvanted seasonal influenza vaccine in 6-<72 months old children. N Engl J Med. In press.
8. Palmer DF, Dowle WR, Coleman MT, et al.. Advanced laboratory technicals for immunological diagnostics. In: Immunology Series No. 6. Part 2: Haemagglutination-Inhibition Test. US Department of Health, Education, and Welfare, and PHS: Atlanta, GA; 1975:25–62.
9. Prentice RL. Surrogate endpoints in clinical trials: definition and operational criteria. Stat Med. 1989;8:431–440.
10. Dunning AJ. A model for immunological correlates of protection. Stat Med. 2006;25:1485–1497.
11. Nauta JJ, Beyer WE, Osterhaus AD. On the relationship between mean antibody level, seroprotection and clinical protection from influenza. Biologicals. 2009;37:216–221.
12. Kayhty H, Peltola H, Karanko V, et al.. The protective level of serum antibodies to the capsular polysaccharide of Haemophilus influenzae type b. J Infect Dis. 1983;147:1100.
13. Black S, Shinefield H, Fireman B, et al.. Efficacy, safety and immunogenicity of heptavalent pneumococcal conjugate vaccine in children. Pediatr Infect Dis J. 2000;19:187–195.
14. Jodar L, Butler J, Carlone G, et al.. Serologic criteria for evaluation and licensure of new pneumococcal conjugate vaccine formulations for use in infants. Vaccine. 2003;21:3265–3272.
15. Plotkin S. Immunologic correlates of protection induced by vaccination. Pediatr Infect Dis J. 2001;20:63–75.
16. Ipsen J. Circulating antitoxin at the onset of diphtheria in 425 patients. J Immunol. 1946;54:325–347.
17. Bjorkholm B, Bottiger M, Christenson B, et al.. Antitoxin antibody levels and the outcome of illness during an outbreak of diphtheria among alcoholics. Scand J Infect Dis. 1986;18:235–239.
18. Zepp F, Schmitt HJ, Kaufhold A, et al.. Evidence of induction of polysaccharide specific B cell immunity in the first year of life: plain Haemophilus influenzae type b PRP boosters in children primed with Hib-DTPa-hepatitis B combined vaccine. Eur J Pediatr. 1997;156:18–24.
19. Hung IF, To KK, Lee CK, et al.. Effect of clinical and virologic parameters on the level of neutralizing antibody against pandemic influenza A H1N1 2009. Clin Infect Dis. 2010;51:274–279.
20. Hall CB, Douglas RC Jr, Gelman JM, et al.. Viral shedding patterns of children with influenza B infection. J Infect Dis. 1979;140:610–613.
21. Fox JP, Cooney MK, Hall CE, et al.. Influenza virus infections in Seattle families, 1975–1979. Am J Epidemiol. 1982;116:228–242.
22. Lee MS, Mahmood K, Adhikary L, et al.. Measuring antibody responses to a live attenuated influenza vaccine in children. Pediatr Infect Dis J. 2004;23:852–856.
23. Khurana S, Verma N, Yewdell JW, et al.. MF59 adjuvant affects human epitope repertoire, augments affinity maturation, and neutralization titers to pandemic influenza in all age groups. Science Transl Med. 2011;3:85ra48.
influenza; vaccine; correlate of protection; children
© 2011 Lippincott Williams & Wilkins, Inc.