S

C

Ma

b

c

d

e

f

g

a

ARRAA

KVIH

1

edficipvwffuccvct

D

t

0d

Vaccine 30 (2011) 1– 4

Contents lists available at SciVerse ScienceDirect

Vaccine

jou rn al h om epa ge: www.elsev ier .com/ locate /vacc ine

hort communication

omparison of multiple estimates of efficacy for influenza vaccine�,��

ark Loeba,b,c,d,∗, Margaret L. Russell e, Kevin Fonseca f, Richard Webbyg, Stephen D. Walterc

Department of Pathology and Molecular Medicine, McMaster University, Hamilton, Ontario, CanadaDepartment of Medicine, McMaster University, Hamilton, Ontario, CanadaDepartment of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario, CanadaMichael G. DeGroote Institute for Infectious Disease Research, McMaster University, Hamilton, Ontario, CanadaDepartment of Community Health Sciences, University of Calgary, Calgary, Alberta, CanadaProvincial Laboratory for Public Health and Department of Microbiology & Infectious Diseases, University of Calgary, Calgary, Alberta, CanadaSt. Jude Children’s Research Hospital and WHO Collaborating Center, Memphis, TN, United States

r t i c l e i n f o

rticle history:eceived 9 March 2011eceived in revised form 20 October 2011

a b s t r a c t

Influenza vaccine trials typically report vaccine efficacy for infection-confirmed symptomatic illness. Dataon indirect vaccine efficacy for susceptibility, the degree of vaccine protection to susceptibles, or indirectvaccine efficacy for illness given infection, are sparse. Using inactivated influenza vaccine randomized

ccepted 26 October 2011vailable online 7 November 2011

eywords:accine

nfluenza

trial data, we calculated indirect vaccine efficacy for susceptibility of 20% [95% CI 9–30] and indirectvaccine efficacy for illness among infected persons 12% [95% CI 2–22], values inferior to a direct vaccineefficacy for infection-confirmed symptomatic illness of 55% [95% CI −21 to 84] and an indirect effect of61% [95% CI 8–83]. Such data reveal variance in protective efficacy of the vaccine for multi-dimensionaldirect and indirect efficacy measures.

erd effects

. Introduction

The most common method for estimating influenza vaccinefficacy in clinical trials is to identify symptomatic participantsuring the influenza season and then to obtain laboratory con-rmation either using RT-PCR or viral culture [1]. Halloran andolleagues previously labeled this measure vaccine efficacy fornfection-confirmed symptomatic illness [2]. In addition they pro-osed vaccine efficacy parameters to evaluate the ability of theaccine to reduce infection, vaccine efficacy for susceptibility, asell as vaccine efficacy for illness among infected persons [i.e.

or progression from infection to illness] [2,3]. Vaccine efficacyor infection confirmed symptomatic illness is typically measuredsing RT-PCR to detect influenza in symptomatic individuals. Inontrast, vaccine efficacy for susceptibility can be defined by effi-acy of the vaccine to prevent serological diagnosis of infection oriral shedding. Vaccine efficacy for illness among infected persons

an be defined by vaccine efficacy against symptomatic illness inhose who met criteria for serological infection or viral shedding.

� The study was presented, in part, at the 9th Canadian Immunization Conference,ecember 7, 2011, in Quebec City.

�� Trial Registration: clinicaltrials.gov identifier NCT00877396.∗ Corresponding author at: McMaster University, 1200 Main Street West, Hamil-

on, Ontario, Canada L8N 3Z5. Tel.: +1 905 525 9140x26066; fax: +1 905 389 5822.E-mail address: loebm@mcmaster.ca (M. Loeb).

264-410X/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.oi:10.1016/j.vaccine.2011.10.069

© 2011 Elsevier Ltd. All rights reserved.

Although immunizing children and adolescents againstinfluenza provides substantial indirect benefit to non-vaccinatedindividuals [4–10], there is sparse data on indirect measures ofvaccine efficacy for susceptibility or vaccine efficacy for illnessamong infected persons with such interventions. In this report,we estimate indirect vaccine efficacy for susceptibility and vaccineefficacy for illness among infected persons as a secondary dataanalysis from a cluster randomized trial, using data from this trial,and compare them to direct and indirect estimates of vaccineefficacy for infection-confirmed symptomatic illness [11].

2. Methods

2.1. Cluster randomized trial

Residents of Hutterite colonies in the provinces of Alberta,Saskatchewan, and Manitoba, were enrolled in the trial fromSeptember 22 to December 23, 2008 [11]. Healthy children aged36 months to 15 years were eligible to be vaccinated, while thosewith contraindications to either influenza or hepatitis A vaccinewere excluded. Other colony members were enrolled to assess theindirect effect of vaccinating the children. Entire Hutterite colonieswere randomized to either vaccination of children with inactivated

influenza vaccine recommended for the 2008–2009 influenza sea-son [A/Brisbane/59/2007 [H1N1]-like virus, A/Brisbane/10/2007[H3N2]-like virus, B/Florida/4/2006-like virus] [VaxigripTM, SanofiPasteur] or to vaccination of children with hepatitis A vaccine

2 M. Loeb et al. / Vaccine 30 (2011) 1– 4

Table 1Direct and indirect vaccine efficacy for infection-confirmed symptomatic illness for inactivated influenza vaccine compared to hepatitis A vaccine.

Colonies where children wereallocated to Influenza vaccine

Colonies where children wereallocated to Hepatitis A vaccine

Efficacy (95% CI)

RT-PCR positive for influenzan (%)

RT-PCR positive for influenzan (%)

Vaccine recipients* N = 502 N = 445VESP direct† 41 (8.2) 79 (17.8) 55 (−21 to 84)A Brisbane H1N1‡ 0 (0) 17 (3.8) 96 (64–100)A Brisbane H3N2 8 (1.6) 20 (4.5) 66 (−89 to 94)B Brisbane 33 (6.6) 42 (9.4) 32 (−193, 84)

Vaccine non-recipients† N = 1271 N = 1055VESP indirect§ 39 (3.1)¶ 80 (7.6) 61 (8 to 83)A Brisbane H1N1 2 (0.16) 24 (2.3) 94 (−78 to 93)A Brisbane H3N2 17 (1.3) 36 (3.4) 64 (−78 to 93)B Brisbane 16 (1.3) 20 (1.9) 37 (−158 to 84)

* Children vaccinated with study vaccine (either influenza vaccine or hepatitis A vaccine) in the trial.† Previously published [11].

in SA

[nfDmw

RIhnfsi[

2

2e

to[–sviwdwhdipBsi

pcmpin

‡ The confidence intervals were derived from the hazard ratio using PROC PHREG§ Refers to trial participants who did not receive study vaccine.¶ Four specimens could not be typed.

Avaxim-PediatricTM, Sanofi Pasteur] as the control. Study vacci-ated children and non-recipients of study vaccine were assessed

or signs and symptoms of influenza over the follow up period, fromecember 28, 2008 until June 23, 2009 and nasopharyngeal speci-ens were obtained if two or more signs or symptoms of influenzaere present [11].

Influenza was confirmed on the basis of detection of viralNA in respiratory samples using Real-Time RT-PCR [12].

nfluenza titres to the seasonal subtypes determined by theemagglutination inhibition assay were obtained in only theon-recipients of study vaccine. Infection was defined by a ≥4-

old increase in titre between baseline and post season serumamples using turkey erythrocytes and the antigens circulat-ng [A/Brisbane/59/2007 [H1N1]-like virus, A/Brisbane/10/2007H3N2]-like virus, B/Brisbane/60/2008-like virus] [13].

.2. Estimation of vaccine efficacy measures

.2.1. Cox proportional hazards regression model for vaccinefficacy against infection-confirmed symptomatic illness

The estimate of vaccine efficacy for infection-confirmed symp-omatic illness was previously reported as the primary outcomef the trial since symptomatic illness was confirmed by RT-PCR11]. Because symptomatic illness as detected by RT-PCR is a time

to – event variable, we used a Cox proportional hazards regres-ion model for the analysis, allowing us to assess the effect ofaccination on time between participant entry into the study andnfluenza infection. Therefore, the dependent variable in the model

as symptomatic infection confirmed by RT-PCR with the indepen-ent variable being the vaccination status of the colony, i.e. a colonyhere colony children had been immunized to either influenza orepatitis A. Models were developed for vaccine recipients (chil-ren vaccinated with study vaccine) to show an overall effect (all

nfluenza A subtypes and influenza B) as well as separate models forarticipants infected with each influenza A subtype and influenza. The number of participants in each group is shown in Table 1. Aet of models was also developed for vaccine non-recipients, thats trial participants who did not receive study vaccine.

In order to account for the likely possibility that study partici-ants who were clustered in the same colonies would have had aorrelated risk of influenza, we used robust sandwich variance esti-

ates to account statistically for the effect of this clustering in the

roportional hazards model [14]. In this paper, we report direct andndirect vaccine efficacy for infection-confirmed symptomatic ill-ess by subtype. The protective effectiveness of the vaccine for the

S.

subtypes was estimated using the formula [1 − hazard ratio] × 100both for overall protection and separately for influenza B and forinfluenza A subtypes.

2.2.2. Logistic regression analyses for vaccine efficacy againstsusceptibility to infection

To calculate indirect vaccine efficacy for susceptibility, we usedthe results of HAI serology in non-recipients of study vaccine. Thatis, an individual with a ≥4-fold increase in titre between base-line and post season was considered a laboratory-confirmed case.Therefore, the dependent variable in the model was binary, i.e.whether participants had a ≥4-fold increase in titres or not. A modelfor an overall effect (including all influenza A subtypes and B) wasbuilt as were models for each influenza A subtype and influenza Bseparately. The number of participants considered for each model isshown in Table 2. The only independent variable in each model wasthe vaccination status of the colony. Since PCR was not done sys-tematically to assess shedding but only in response to symptoms,the evaluation was limited to serology. This way vaccine efficacyfor susceptibility would be distinct from vaccine efficacy for symp-tomatic illness. In contrast to the RT-PCR, where onset of symptomsany time during follow up may have led to testing, serological dataare not time – to – event and therefore use of proportional hazardsmodel would not be appropriate. We therefore used generalizedestimating equations to account for membership in the random-ized clusters with the logit-link function for dichotomous variables[15]. This technique is a variant of logistic regression where thegeneralized estimating equation accounts for the clustered natureof the data. Protective effectiveness was specified by the formula[1 − odds ratio] × 100.

2.2.3. Logistic regression analyses for vaccine efficacy for illnessamong infected persons

Indirect vaccine efficacy for illness among infected persons wasestimated by determining whether influenza-like illness [two ormore signs and symptoms] occurred within a 10 day period giventhe presence of a ≥4-fold increase in titre between baseline andpost season in non-recipients of study vaccine. Similar to the modelfor vaccine efficacy for susceptibility, the dependent variable in themodel was binary, i.e. whether infected persons were symptomaticor not, with the subgroup in the analysis being those participants

with a ≥4-fold increase in titres. Again a model for an overall effect(including all influenza A subtypes and B) was built as were mod-els for each influenza A subtype and influenza B separately withthe independent variable in each model being vaccination status

M. Loeb et al. / Vaccine 30 (2011) 1– 4 3

Table 2Indirect vaccine efficacy against susceptibility to infection and vaccine efficacy for illness among infected persons for inactivated influenza vaccine compared to hepatitis Avaccine.

Vaccine to which children incolonies were allocated to

Efficacy (95% CI)

Influenza Hepatitis A

Vaccine non-recipients in whom serology was performed N = 714 N = 603VES indirect 290 (40.6) 309 (51.2) 20 (9 to 30)A Brisbane H1N1 67 (9.4) 80 (13.3) 41 (−1 to 67)A Brisbane H3N2 256 (35.9) 258 (42.8) 26 (−10 to 53)B Brisbane 24 (3.4) 38 (6.3) 46 (0–71)

Vaccine non-recipients who had ≥four-fold rise in antibody titres N = 290 N = 309VEP indirect 95 (32.8) 136 (44.0) 12 (2–22)

olidcueaw[

3

3

a1cvcAc5c

ihtAyva

13b

3

tiwicpt3

A Brisbane H1N1A Brisbane H3N2

B Brisbane

f the colony. We used generalized estimating equations with theogit-link function in this analysis in order to account for correlationnduced by cluster membership. We did not include PCR confirmedisease to ensure that this effect estimate was distinct from vac-ine efficacy for symptomatic illness. Similar to the approach wesed for vaccine efficacy for susceptibility, this is consistent withstimates done in challenge studies where methods that could bepplied systematically to all subjects (serology or viral shedding)ere used [3]. Protective effectiveness was specified by the formula

1 − odds ratio] × 100.

. Results

.1. Study population

As previously reported, 46 colonies, 22 in the influenza groupnd 24 in the hepatitis A group that were followed [11]. There were271 non-recipients of study vaccine in the influenza vaccinatedolonies (94 or 7% were children) and 1055 non-recipients of studyaccine in the hepatitis vaccinated colonies. Of study vaccinatedhildren, 502 received influenza vaccine and 445 received hepatitis

vaccine. The mean vaccine coverage among healthy children oflusters assigned to influenza vaccine was 83%, range by cluster3% to 100%. This was similar to the mean vaccine coverage amongolonies assigned to hepatitis A vaccine, 79%, range 50–100%.

Serology was obtained from 714 non-vaccine recipients innfluenza vaccinated colonies and 603 non-vaccine recipients inepatitis A colonies. The age distribution was similar between thewo groups in all age categories. Comparing influenza to hepatitis

vaccinated colonies 10% versus 7% were aged between 3 and 15ears; 75% versus 76% were aged between 16 and 49 years; 11%ersus 12% were between 50 and 64 years; and 4% versus 5% wereged 64 years and over.

RT-PCR detected 124 influenza A cases [43 H1 and 81 H3], and11 cases of influenza B. Including the results of serology, 912 of272 [27.9%] of all participants had either influenza A or B detectedy PCR or a four-fold rise in HI titres to influenza A or B.

.2. Vaccine efficacy for infection-confirmed symptomatic illness

The results of vaccine efficacy for infection-confirmed symp-omatic illness are shown in Table 1. Of 502 children who receivednfluenza vaccine there were 41 [8.2%] versus 79 of 445 [17.8%]

ho received hepatitis A vaccine who were confirmed as havingnfluenza by RT-PCR. Thus, the direct vaccine efficacy for infection-

onfirmed symptomatic illness was 55% [95% CI −21% to 84%] asreviously published [11]. As shown in Table 1, the highest protec-ion was for H1N1 [96%; 95% CI 64–99%]. For vaccine non-recipients,9/1271 [3.1%] were confirmed as having influenza in colonies

85 (33.2) 111 (41.2) 7 (−11 to 22)22 (32.8) 33 (41.2) 7 (−11 to 22)8 (33.3) 16 (42.1) 4 (−29 to 29)

allocated to child vaccination with inactivated influenza vaccinecompared to 80/1055 [7.6%] in colonies allocated to child vaccina-tion with hepatitis A vaccine. Thus, the indirect vaccine efficacy forinfection-confirmed symptomatic illness was 61% [95% CI 8–83%].

3.3. Indirect vaccine efficacy for susceptibility and vaccineefficacy for illness among infected persons

The estimates of indirect vaccine efficacy for susceptibility andvaccine efficacy for illness among infected persons are shown inTable 2. Of 714 participants in colonies where children were allo-cated to influenza vaccination, 290 [40.6%] developed a four-foldrise in antibody titres against influenza of any type in contrastto 309/603 [51.2%] of those in colonies where children were vac-cinated against hepatitis A. Thus, the estimated indirect vaccineefficacy for susceptibility was 20% [95% CI 9–30%]. The highest indi-rect vaccine efficacy for susceptibility was for B Brisbane. Whenonly those who mounted a four-fold rise in titres were considered,95/290 [32.8%] in influenza vaccinated colonies versus 136/309[44%] in hepatitis A vaccinated colonies developed acute respi-ratory illness, for an indirect vaccine efficacy for illness amonginfected persons of 12% [95% CI 2, 22]. The highest effect was for ABrisbane H1N1 [Table 2].

4. Discussion

Few vaccine trials report measures of vaccine efficacy other thanlaboratory confirmed symptomatic infection [1]. Moreover, to thebest of our knowledge, indirect estimates of vaccine efficacy for vac-cine efficacy for illness among infected persons not been reported.

We found that the point estimate of the indirect vaccine effi-cacy for infection-confirmed symptomatic illness was 61%, similarto that of the direct vaccine efficacy for infection-confirmedsymptomatic illness [55%]. That the latter was not statisticallysignificant was not unexpected given that it was derived fromthe children who received study vaccine, a subgroup for whichour cluster randomized trial was not powered to detect an effect.These results suggest a strong effect of herd immunity in reducingsymptomatic illness due to influenza. A summary of vaccineefficacy for infection-confirmed symptomatic illness estimatesfrom challenge studies for inactivated vaccine reveals an effectsize of similar magnitude, 63% [3]. A review of our vaccine efficacyfor infection-confirmed symptomatic illness estimates for thesubtypes suggests that the vaccine had the highest efficacy againstA Brisbane H1N1 for both direct and indirect estimates, with >95%

efficacy in both. There was a match between circulating antigensand vaccine antigens for both A Brisbane H1N1 and A BrisbaneH3N2 in the 2008–2009 influenza season [13]. It is unclear whythe efficacy of the A Brisbane H1N1 is higher than the H3N2, but

4 accine

pBwtFai

wchtsbswoptioifev

vpvfpseised

A

R

[

[

[

[

[14] Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models.Biometrika 1986;73:13–22.

[15] Wei LJ, Lin DY, Weissfeld L. Regression-analysis of multivariate incom-plete failure time data by modeling marginal distributions. J Am Stat Assoc1989;84:1065–73.

M. Loeb et al. / V

erhaps there was better immunogenicity provided with the Arisbane H1N1 antigen than the A Brisbane H3N2 antigen. Thereas a drift noted from the B antigen in the vaccine, such that

here was a mismatch between B Brisbane in the vaccine and Blorida that circulated [13]. This explains the much lower directnd indirect estimates of 32% and 37% for vaccine efficacy fornfection-confirmed symptomatic illness for influenza B.

The estimate of indirect vaccine efficacy for susceptibility, 20%,as substantially lower than indirect vaccine efficacy for infection-

onfirmed symptomatic illness. This suggests that the herd effectas a greater benefit at preventing influenza confirmed symp-omatic illness than serological infection. Data from challengetudies also show a lower efficacy for vaccine efficacy for suscepti-ility [43%] compared to vaccine efficacy for infection-confirmedymptomatic illness 63% [3]. However, confidence intervals areide and no firm conclusions can be made. The indirect effect

f the inactivated vaccine on preventing illness among infectedersons (or progression to disease) was 12%, which is lowerhan vaccine efficacy for susceptibility and vaccine efficacy fornfection-confirmed symptomatic illness. The relative effectivenessf vaccine efficacy for illness among infected persons using our datas in keeping with that in challenge studies, where vaccine efficacyor illness among infected persons was 29% compared to vaccinefficacy for infection-confirmed symptomatic illness of 63% andaccine efficacy for susceptibility of 43% [3].

Study limitations include the fact that the serology to estimateaccine efficacy for susceptibility was incomplete, because we hadaired specimens from only 65% of participants. We did not includeiral shedding data for our estimates of indirect vaccine efficacyor susceptibility and vaccine efficacy for illness among infectedersons. Because we did not collect end of season specimens forerology from the children, we could not calculate direct vaccinefficacy for susceptibility and vaccine efficacy for illness amongnfected persons among the children and adolescents who receivedtudy vaccine. In summary, our data reveal variance in protectivefficacy of the inactivated influenza vaccine for multi-dimensionalirect and indirect efficacy measures.

cknowledgements

The study was funded by the Canadian Institutes for Healthesearch and NIAID.

30 (2011) 1– 4

References

[1] Demicheli V, Di Pietrantonj C, Jefferson T, Rivetti A, Rivetti D. Vaccines forpreventing influenza in healthy adults. Cochrane Database Syst Rev 2007;2,doi:10.1002/14651858.

[2] Halloran ME, Longini Jr IM, Struchiner CJ. Design and interpretation of vaccinefiled studies. Epidemiol Rev 1999;21:73–88.

[3] Basta NE, Halloran ME, Matrajt L, Longini Jr I. Estimating influenza vaccineefficacy from challenge and community-based study data. Am J Epidemiol2008;168:1343–52.

[4] Monto A, Davenport FM, Napier JA, Francis T. Modification of an outbreak ofinfluenza in Tecumseh, Michigan by vaccination of school children. J Infect Dis1970;122:16–25.

[5] Esposito S, Marchisio P, Cavagna R, et al. Effectiveness of influenza vacci-nation of children with recurrent respiratory tract infections in reducingrespiratory-related morbidity within the households. Vaccine 2003;21:3162–8.

[6] Piedra PA, Gaglani MJ, Kozinetz CA, et al. Herd immunity in adults againstinfluenza-like illnesses with use of the trivalent-live attenuated influenza vac-cine [CAIV-T] in children. Vaccine 2005:1540–8.

[7] Reichert TA, Sugaya N, Fedson DS, et al. The Japanese experience withvaccinating school children against influenza. N Engl J Med 2001;344:889–96.

[8] Hurwitz ES, Haber M, Chang A, et al. Effectiveness of influenza vaccination ofday care children in reducing influenza-related morbidity among householdcontacts. JAMA 2000;284:1677–82.

[9] Rudenko LG, Slepushkin AN, Monto AS, et al. Efficacy of live attenuated and inac-tivated influenza vaccines in schoolchildren and their unvaccinated contacts inNovgorod, Russia. J Infect Dis 1993;168:881–7.

10] King Jr JC, Stoddard JJ, Gaglani MJ, et al. Effectiveness of school-based influenzavaccination. N Engl J Med 2006;355:2523–32.

11] Loeb M, Russell ML, Moss L, et al. Effect of influenza vaccination of childrenon infection rates in hutterite communities: a randomized control trial. JAMA2010;303(10):943–50.

12] Dawood FS, Jain S, Finelli L, et al. Novel swine-origin influenza A [H1N1] virusin vaccine efficacy for susceptibility litigation team. Emergence of a novelswine-origin influenza A [H1N1] virus in humans. N Engl J Med 2009;360:2605–15.

13] National Advisory Committee on Immunization. Statement on seasonaltrivalent inactivated influenza vaccine [TIV] for 2009–2010. CanadaCommunicable Disease Report 2009;35(ACS-6):1–41, http://www.phac-aspc.gc.ca/publicat/ccdr-rmtc/09vol35/acs-dcc-6/index-eng.php [accessed5.01.2011].