MAJOR ARTICLE
Concentration of Antipneumococcal Antibodies
as a Serological Correlate of Protection:
An Application to Acute Otitis Media
Jukka T. Jokinen, Heidi Åhman,a Terhi M. Kilpi, P. Helena Mäkelä, and M. Helena Käyhty
Department of Vaccines, National Public Health Institute, Helsinki, Finland
Background. For the licensing of new pneumococcal vaccines, it is vital to be able to predict their protective
efficacy on the basis of immunogenicity. However, the serological correlates of protection have not been established
for pneumococcal diseases.
Methods. A total of 1666 children were immunized with the pneumococcal conjugate vaccine. Acute otitis
media (AOM) events were identified, and middle-ear fluid was cultured for pneumococci. The association between
the concentration of antibodies against serotypes 6B, 19F, and 23F and the risk of AOM caused by the homologous
serotypes or by the cross-reactive serotype 6A was assessed. An association model was used to predict efficacy at
different geometric mean concentrations (GMCs).
Results. An association between antibody concentration and risk of AOM was found, but with large differences
between serotypes. On the basis of the association, the predicted efficacy for 19F was negligible up to the highest
GMC tested. In contrast, 6B was found to be highly efficacious (165%) at a GMC of 0.5 mg/mL.
Conclusions. The results challenge the view that a new vaccine candidate should always induce antibody
concentrations that are not inferior to those produced by the licensed vaccine. Furthermore, the differences between
serotypes caution against defining a common correlate of protection that is applicable to all serotypes.
The first pneumococcal conjugate vaccine was recently
licensed in the United States and in the European Union.
In a clinical efficacy trial in the United States, this 7valent conjugate vaccine showed excellent efficacy, 97%
(95% confidence interval [CI], 81%–100%), against invasive disease caused by the 7 serotypes targeted by the
vaccine [1]. When the same vaccine was studied in Finland, in the Finnish Otitis Media (FinOM) Vaccine Trial,
with acute otitis media (AOM) as the end point, the serotype-specific efficacy was 57% (95% CI, 44%–67%) [2].
Because the first pneumococcal conjugate vaccine has
been licensed and proven to be efficacious, clinically de-
Received 8 July 2003; accepted 22 February 2004; electronically published 2
July 2004.
Presented in part: 3rd International Symposium on Pneumococci and
Pneumococcal Diseases, Anchorage, Alaska, 5–8 May 2002 (abstract XX).
Financial support: Aventis Pasteur; Merck; Wyeth-Lederle Vaccines.
a
Present affiliation: Wyeth Lederle Nordiska, Vantaa, Finland.
Reprints or correspondence: Dr. Jukka Jokinen, Dept. of Vaccines, National
Public Health Institute, Mannerheimintie 166, FIN-00300 Helsinki, Finland (jukka
[email protected]).
The Journal of Infectious Diseases 2004; 190:545–50
2004 by the Infectious Diseases Society of America. All rights reserved.
0022-1899/2004/19003-0018$15.00
fined efficacy trials of subsequent vaccines will meet practical and ethical difficulties. Instead, immunogenicity
studies are relatively easy to perform, and, on the basis
of our knowledge that antibody-mediated phagocytic
killing is the primary mechanism of protection [3–5],
they are attractive candidates for predicting efficacy. The
current regulatory requirement for the licensing of new
pneumococcal conjugate vaccines is the demonstration
of noninferiority in immunogenicity, compared with the
licensed vaccine. However, this approach lacks a direct
relationship to vaccine efficacy (VE).
The data from efficacy studies with invasive infections as the end point are often not useful for defining
the correlates of protection, because of the paucity of
breakthrough cases (i.e., vaccinated subjects developing
the disease). There would be less of a handicap with
AOM as the end point, whereas the antibody-mediated
phagocytic killing as the mechanism of protection would
be similar. We therefore considered it to be of interest
to study the question of serological correlates of protection against AOM, expecting the study to provide
us with new insight into the question. We used data
available from the FinOM Vaccine Trial, a combined
Serological Correlate of Protection • JID 2004:190 (1 August) • 545
efficacy and immunogenicity study with 2 7-valent (serotypes
4, 6B, 9V, 14, 18C, 19F, and 23F) pneumococcal conjugate
vaccines, PncCRM and PncOMPC. The primary end point of
the study was an AOM event caused by serotypes targeted by
the vaccine, leading to a moderate number of children with
breakthrough cases. In addition, postvaccination serum samples
were available from practically all participants [2, 6], making
it possible to establish a direct relationship between the disease
and immunogenicity.
The data from the FinOM Vaccine Trial were analyzed by
fitting a model to determine the association between the individual postvaccination antibody concentration and subsequent
risk of AOM. On the basis of this association model and of the
baseline risk estimated from the group of nonimmunized children, predictions of VE, at a range of geometric mean concentrations (GMCs) of antibody, were obtained. Because of the small
number of AOM events caused by the other serotypes, we focused
on the 3 most common vaccine serotypes—6B, 19F, and 23F—
and on the common cross-reactive nonvaccine serotype 6A.
SUBJECTS AND METHODS
The FinOM Vaccine Trial estimated the efficacy of 2 7-valent
conjugate vaccines, PncCRM and PncOMPC (pneumococcal serotypes 4, 6B, 9V, 14, 18C, 19F, and 23F conjugated to CRM197
protein or outer-membrane protein complex of Neisseria meningitidis serogroup B, respectively), for the prevention of pneumococcal AOM. The pneumococcal vaccines were used in parallel with a control vaccine (hepatitis B virus [HBV] vaccine;
Merck Sharp & Dohme). The main results on VE have been
published [2, 6]. The FinOM Vaccine Trial was conducted in
accordance with the Declaration of Helsinki (as amended in
Hong Kong, 1989). The protocol was approved before the start
of the trial by the ethics committee of the National Public
Health Institute of Finland, by the National Agency for Medicines, and by the relevant local health authorities. Written,
informed consent was obtained from a parent/guardian of all
children before enrollment in the study.
Study Children, Vaccinations, and Collection of Samples
for Serologic Testing
A total of 2497 children were recruited to this study; 831 were
randomized to receive the PncCRM vaccine (Wyeth Lederle
Vaccines), 835 received the PncOMPC vaccine (Merck Sharp
& Dohme), and the remaining 831 received the HBV vaccine.
A primary series of 3 doses of the pneumococcal vaccines or
the control vaccine was given at 2, 4, and 6 months of age [2,
6]. A booster dose, using the same vaccine as that used during
the primary series, was given at 12 months of age to all children,
with the exception of 106 children in the PncOMPC group,
who received the polysaccharide vaccine Pneumovax (Merck
546 • JID 2004:190 (1 August) • Jokinen et al.
Sharp & Dohme). Serum samples for determination of antibody
concentration were scheduled to be obtained from each child
at either 7 or 13 months of age. The number of children from
whom samples were obtained was 365, 368, and 376 at 7
months of age and 407, 411, and 421 at 13 months of age, in
the HBV, PncCRM, and PncOMPC vaccine groups, respectively.
Follow-up
The follow-up period for this particular analysis differed from
those reported earlier [2, 6], because of the timing when postvaccination serum samples were obtained for determination of
antibody concentration. For those children from whom serum
samples were obtained at 7 months of age, the follow-up period
was 5 months, beginning on the date at which samples were
obtained and ending at the time of the fourth dose of the
vaccine, at 12 months of age. For those children from whom
serum samples were obtained at 13 months of age, the followup period was also 5 months, beginning on the date at which
samples were obtained and ending at 18 months of age. Compliance was excellent in the FinOM Vaccine Trial: 96% of the
enrolled children completed the follow-up as specified in the
protocol [2, 6]. The few children who discontinued the study
before the scheduled date for collection of samples or from
whom a blood sample could not be obtained were not included
in this analysis.
Diagnosis and Definition of AOM
Predefined criteria for diagnosis of AOM, on the basis of symptoms and pneumatic otoscopy findings, were used [2, 6]. When
AOM was diagnosed, myringotomy with aspiration of middleear fluid (MEF) was performed. MEF samples were plated immediately, and culture and bacteriological identification were
performed as described elsewhere [2, 6]. If the MEF culture
tested positive for pneumococci, the AOM event was considered
to be pneumococcal.
According to the findings, the children could be either (1)
children who experienced ⭓1 culture-confirmed AOM event
caused by the serotype of interest (6A, 6B, 19F, or 23F) during
follow-up (hereafter referred to as “children with AOM”) or
(2) children who did not experience an AOM event caused by
the serotype of interest (6A, 6B, 19F, or 23F) during follow-up
(hereafter referred to as “children without AOM”).
Antibody Measurements
The concentrations of IgG antibodies to the capsular polysaccharides of the 3 serotypes (anti-6B, -19F, and -23F) were determined by use of EIA after absorption with C polysaccharide, as described elsewhere [7, 8]. The lowest reliably measured
concentrations—that is, the limits of quantification (LOQ)—for
anti-6B, -19F, and -23F were 0.09, 0.16, and 0.08 mg/mL, re-
spectively. The corresponding percentages below the LOQ were
39%, 28%, and 34% in the HBV group and 3%, 0.3%, and 0.8%
in the immunized group (PncCRM and PncOMPC combined).
Statistical Methods
Groups compared. Hitherto, no evidence has emerged that
pneumococcal conjugate vaccines can produce antibody levels
that are quantitatively the same but qualitatively very different.
Furthermore, if the quality of the antibodies produced by different vaccines was to drastically differ, all attempts to relate
antibodies with protection would be useless because of the lack
of a commensurable surrogate for protection. Therefore, no distinction was made between the 2 pneumococcal conjugate vaccine arms of the FinOM Vaccine Trial, and children in either of
these groups were treated as a single immunized group (1576
children). For determination of VE, this immunized group was
compared with the nonimmunized (i.e., the HBV vaccine) group
(772 children).
GMCs. The average antibody concentrations for children
with AOM and children without AOM were expressed as GMCs
with 95% CIs, for both groups. GMCs for children from whom
samples were obtained at 7 and at 13 months of age were combined. Because of a notably large percentage of values below the
LOQ in the nonimmunized group, we assumed that the antibody concentrations were log-normally distributed, with censoring below the LOQ. GMCs and the corresponding CIs were
calculated by use of the expectation-maximization algorithm [9].
Association model. To investigate the relationship between
antibody concentration and the risk of AOM, the outcome
variable for the immunized children was defined as 1 if the
child suffered a serotype-specific AOM event during follow-up
and as 0 otherwise. The antibody concentration, on a log scale,
at the beginning of the interval was the explanatory variable.
The LOQ value was assigned for the few immunized children
with antibody concentrations below quantification. Bayesian
generalized linear models (GLMs) for binomial data, using reference proper priors [10], were fitted for each serotype, to assess
the effect that antibody concentration has on the incidence of
AOM events during follow-up. It was assumed that antibodies have a continuous effect on the incidence of disease [11].
None of the common link functions were superior to the others,
so a complementary log-log link was used, which produces a
pertinent fit (i.e., incidence of disease) to our objectives. Fitted
values of incidence versus the corresponding antibody concentrations were depicted for the immunized group (figure 1).
Similar models were fitted to estimate the risk of contracting
AOM for the HBV group. For these children, the effect of
antibody concentration was negligible and was left out of the
final model as an explanatory variable.
Prediction of VE. To predict VE in a population with a
predefined postimmunization antibody distribution, a generic
Figure 1. Effect of antibody concentration on the yearly incidence of
serotype-specific acute otitis media (AOM). Fitted values are from the
generalized linear model for each serotype. The risk of AOM caused by
serotype 6A is associated with the concentration of cross-reactive 6B
antibodies.
population (n p 10,000) was created, with the assumption that
their antibody concentrations are log-normally distributed, with
a predefined GMC and an SD estimated from the present trial
(average of the SD of the 2 vaccine groups and the 2 time points).
VE was calculated as the following:
VE p 1 ⫺
p1⫺
average risk of immunized children
baseline risk of nonimmunized children
1
n
冘h
n
ip1
(a vac + bvac ⫻ log [antibody])
i
⫺1
h⫺1(actrl)
,
where avac, bvac, and actrl are the posterior parameters of the
above GLM models (avac and actrl represent the intercepts for
the immunized and nonimmunized groups, respectively, and
bvac is the effect of log[antibody] for the immunized group).
The inverse function of the complementary log-log link, used
to obtain incidence of disease, is denoted by h⫺1. The predictive
distribution of VE was obtained by integrating over the posterior distributions of the GLM parameters, to account for the
sample uncertainty in the estimation of the association model.
Finally, a series of generic populations with different GMCs
were generated by use of a range of plausible values (0.5–10
mg/mL), to assess the shape of the efficacy curve. The predictive
efficacy distribution was plotted as a line through the average,
with pointwise 95% credible intervals (figure 2).
RESULTS
GMCs of antibody for children with and without AOM.
Among the recipients of pneumococcal conjugate vaccines, the
GMCs of the homologous serotype antibody were higher in
children without AOM (2.42–5.35 mg/mL) than in children with
Serological Correlate of Protection • JID 2004:190 (1 August) • 547
Figure 2. Predicted vaccine efficacy (VE) for generic populations with various antibody distributions. Data are presented as average population
efficacy (solid line) with 95% credible intervals (dotted lines), by geometric mean concentration (GMC). The VE against acute otitis media caused by
serotype 6A is associated with the concentration of cross-reactive 6B antibodies.
AOM (0.58–3.86 mg/mL) (table 1), suggesting that children with
lower levels of antibody are at an increased risk of acquiring
pneumococcal AOM. The difference was ∼3-fold for serotypes
23F, 6B, and 6A (the latter 2 assessed on the basis of anti-6B)
but was less clear (1.4-fold) for serotype 19F. In the HBV vaccine group, there were no differences in the serotype-specific
GMCs between children with AOM and children without AOM
(range, 0.1–0.3 mg/mL) (table 1), suggesting that, at this very
low level, the antipolysaccharide antibodies did not contribute
to protection against pneumococcal AOM.
Association of antibody concentration with the risk of AOM.
The association between antibody concentration and the risk
of AOM, among the pneumococcal conjugate vaccine recipients, was then examined by use of generalized linear modeling.
The relationship between increasing antibody concentration
and the risk of AOM is demonstrated in figure 1. In general,
the higher the antibody concentration, the lower the risk of
serotype-specific AOM. The association was strongest for 23F
and weakest for 19F (figure 1). Increasing anti-6B concentration
was moderately associated with decreased risk of AOM caused
by both serotype 6B and serotype 6A; the risk ratios confirmed
this finding (table 2). A 10-fold increase in antibody concentration decreased the risk of serotype-specific AOM; this was
significant for serotypes 6A and 23F and was less clear for
548 • JID 2004:190 (1 August) • Jokinen et al.
serotype 19F (table 2). For 6B, high VE and consequently low
numbers of breakthrough cases (7 in the present study; table
1) made the CIs wide. Practically all of the antibody levels in
the control group were below those in the immunized group.
Therefore, the antibody concentration in these children was
found to have no effect on the risk of AOM caused by serotypes
6A, 6B, 19F, and 23F.
Prediction of VE. Association of antibody concentration
with the risk of AOM does not directly measure the magnitude
of VE. Antigens with moderate associations might still have
high efficacy, as did 6B in the present study. Therefore, comparison with a nonimmunized population is needed. On the
basis of the above association model and the baseline risk of
AOM obtained for the control group, predictions of VE were
obtained for generic populations (figure 2). The graphs for the
4 serotypes are very different, indicating different correlations
of antibody concentrations to protection: even low concentrations of anti-6B (GMC, ⭓0.5 mg/mL) achieved after vaccination
with pneumococcal conjugate vaccine provide ⭓65% protection against AOM caused by serotype 6B. This protection is
not further increased with an increase in antibody concentration. For cross-reactive serotype 6A, protection increases with
an increase in GMC (from 0.5 to 2 mg/mL, for the 6B antigen)
but remains at a lower level than that for 6B. A similar pattern
Table 1. No. of children with and without 6A, 6B, 19F, or 23F acute otitis media
(AOM) events and their geometric mean concentrations (GMCs) of respective IgG
antibodies, in groups of children immunized with hepatitis B virus (HBV) vaccine
or pneumococcal conjugate vaccine.
GMC (95% CI), mg/mL
No. of children
Vaccine group, serotype
Pneumococcal conjugate
6A
6B
19F
23F
HBV
6A
6B
19F
23F
Without
AOMa
With
AOMb
Children
without AOM
Children
with AOM
1559
1569
1555
1558
17
7
21
18
2.43
2.42
5.35
2.42
(2.24–2.65)
(2.23–2.63)
(5.04–5.68)
(2.27–2.58)
0.81
0.58
3.86
0.81
(0.41–1.63)
(0.10–3.39)
(2.31–6.46)
(0.38–1.71)
758
754
753
753
14
18
19
19
0.11
0.11
0.30
0.12
(0.11–0.12)
(0.11–0.12)
(0.28–0.33)
(0.11–0.13)
0.10
0.11
0.29
0.14
(0.06–0.15)
(0.07–0.19)
(0.20–0.44)
(0.11–0.18)
NOTE. For 6A, the GMCs are calculated by use of cross-reactive anti-6B concentrations. CI,
confidence interval.
a
b
Children without AOM caused by the indicated serotype during follow-up.
Children with AOM caused by the indicated serotype during follow-up.
applies to 23F. Anti-19F concentrations have a negligible effect
on the risk of AOM caused by serotype 19F, even up to a GMC
of 10 mg/mL.
DISCUSSION
There is no generally approved methodology for defining serological correlates of protection or protective levels of antibodies.
Aggregate-level methods, such as comparison of VE and GMC
or percentage above a threshold, have been used to define protective levels of antibodies. However, the association detected on
aggregate level may not reflect the individual-level relationship.
Comparison of GMCs for children with AOM and those for
children without AOM, as shown in table 1, gives valuable information for an exploratory analysis, but, because of the huge
differences in the numbers of individuals representing these 2
groups, it lacks confirmatory power and does not provide a
measure for correlates of protection. For the definition of a protective level, an alternative possibility is to determine a level of
incidence of disease and a threshold and to use graphs, such as
that in figure 1, to obtain a qualitative protective level of antibody
concentration from the cut point of the curve and the predefined
incidence. However, a single-point incidence level as a cutoff for
protection is rather arbitrary and, because of the continuous
nature of the effect of antibodies on the risk of disease, lacks
biological basis. Our approach was to quantify the risk associated
with a range of antibody concentrations on an individual level
and to compare the risk of disease in a nonimmunized population with that in an immunized population with predefined
GMCs of antibody. This provided us with information about the
shape and magnitude of VE in populations with different postimmunization antibody distributions.
Because of the immunization schedule, the follow-up in the
present study was relatively short: 5 months. However, because
of the T cell–dependent nature of conjugate vaccines and the
generation of immunological memory, elevated antibody concentrations persist for years [12], and, thus, neither the length
of follow-up nor the age of measurement of the antibody concentration is expected to have a major effect on the shape of
the curves in figure 2. The breakthrough cases in the present
study were spread evenly across the follow-up period, and results from the FinOM Vaccine Trial [2, 6] showed that the
relative risk for all vaccine serotypes combined was constant
from 6 to 24 months.
Our results demonstrate an individual-level association between antibody concentration and protection against AOM.
Table 2. Effect of a 10-fold increase in antibody
concentration on the risk of serotype-specific
acute otitis media (AOM) for immunized children:
estimates from the generalized linear model for
each serotype.
Serotype
6A
6B
19F
23F
RR (95% CI)
0.41
0.41
0.59
0.26
(0.21–0.80)
(0.15–1.17)
(0.25–1.42)
(0.12–0.58)
NOTE. The risk of AOM caused by serotype 6A is associated with the concentration of cross-reactive 6B antibodies. CI, credible interval; RR, relative risk.
Serological Correlate of Protection • JID 2004:190 (1 August) • 549
Furthermore, the association varied from serotype to serotype,
clearly speaking against the setting of a unique serological correlate of protection for all serotypes. The methods for comparing future pneumococcal conjugate vaccine candidates with
the licensed ones should take into account these differences
between serotypes. For example, on the basis of our results, for
6B, a GMC of ⭓0.5 mg/mL predicted a virtually constant efficacy. Therefore, requiring the demonstration of noninferiority
to a vaccine with a GMC of 2 mg/mL may be too harsh a
requirement and may actually underestimate the usefulness of
an efficacious vaccine with a lower GMC.
The robustness of our approach to various model assumptions—such as choice of the link function, priors, assignment of
the values below the LOQ, and the SD of the antibody distribution—was investigated. These assumptions had negligible effects on the results, and, thus, when drawing conclusions from
the analysis, the model uncertainty is inconsequential, compared
with sample uncertainty. Admittedly, because of the small numbers of children with AOM, the CIs in figure 2 are wide. However,
if the precision of our results raises concern, this concern serves
as a caution that inferences from invasive disease data derived
from studies in which there are even fewer cases can be far more
questionable. In addition, regardless of the wide CIs, the difference between serotypes is apparent.
Although it is clear that the antibody concentrations needed
for prevention of invasive infections are lower than those needed
for prevention of AOM, the basis of the antibody-dependent
protection is the same. Therefore, we suggest that the shapes of
the efficacy curves in figure 2 also apply to invasive infections
and that the case of correlates of protection against AOM can
be taken as a model for correlates of protection against invasive
disease and, furthermore, against pneumococcal pneumonia, the
most important pneumococcal disease globally.
550 • JID 2004:190 (1 August) • Jokinen et al.
Acknowledgment
We would like to thank Robert Kohberger (Wyeth Vaccine Research) for
ideas and several useful discussions about this topic.
References
1. Black S, Shinefield H, Fireman B, et al. Efficacy, safety and immunogenicity of heptavalent pneumococcal conjugate vaccine in children.
Pediatr Infect Dis J 2000; 19:187–95.
2. Eskola J, Kilpi T, Palmu A, et al. Efficacy of a pneumococcal conjugate
vaccine against acute otitis media. N Engl J Med 2001; 344:403–9.
3. Gardner S, Anderson D, Webb B, et al. Evaluation of Streptococcus
pneuminiae type XIV opsonins by phagocytosis-associated chemiluminescence and bactericidal assay. Infect Immun 1982; 35:800–8.
4. Brown E, Hosea S, Frank M. The role of antibody and complement
in the reticuloendothelial clearance of pneumococci from the bloodstream. Rev Infect Dis 1983; 5(Suppl 4):S797–805.
5. Gillespie S. Aspects of pneumococcal infection including bacterial virulence, host response and vaccination. J Med Microbiol 1989; 28:237–48.
6. Kilpi T, Åhman H, Jokinen J, et al. Protective efficacy of a second
pneumococcal conjugate vaccine against pneumococcal acute otitis
media in infants and children. Clin Infect Dis 2003; 37:1155–64.
7. Käyhty H, Åhman H, Rönnberg P-R, Tillikainen R, Eskola J. Pneumococcal polysaccharide—meningococcal outer membrane protein complex conjugate vaccine is immunogenic in infants and children. J Infect
Dis 1995; 172:1273–8.
8. Åhman H, Käyhty H, Tamminen P, Vuorela A, Malinoski F, Eskola J.
Pentavalent pneumococcal conjugate vaccine PncCRM is well-tolerated
and able to induce an antibody response in infants. Pediatr Infect Dis
J 1996; 15:134–9.
9. Dempster AP, Laird NM, Rubin DB. Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc Ser B 1977; 39:1–38.
10. Raftery AE. Approximate Bayes factors and accounting for model uncertainty in generalised linear models. Biometrika 1996; 83:251–66.
11. Jódar L, Butler J, Carlone G, et al. Serological criteria for evaluation
and licensure of new pneumococcal conjugate vaccine formulations
for use in infants. Vaccine 2003; 21:3265–72.
12. Mäkelä PH, Käyhty H, Leino T, et al. Long-term persistence of immunity after immunization with Haemophilus influenzae type b conjugate vaccine. Vaccine 2003; 22:287–92.