Embed Size (px)
Citation preview
Cy
Ja
1b
a
ARR1AA
KSZCUI
1
dt3iacacct
o
(n
0d
Vaccine 30 (2012) 675– 684
Contents lists available at SciVerse ScienceDirect
Vaccine
jou rn al h om epa ge: www.elsev ier .com/ locate /vacc ine
ost-effectiveness of vaccination against herpes zoster in adults aged over 60ears in Belgium
oke Bilckea,∗, Christiaan Maraisa, Benson Ogunjimia, Lander Willema, Niel Hensa,b, Philippe Beutelsa
Center for Health Economics Research and Modeling Infectious Diseases (CHERMID), Vaccine and Infectious Disease Institute (Vaxinfectio), University of Antwerp, Universiteitsplein, 2610 Wilrijk, BelgiumInteruniversity Institute for Biostatistics and statistical Bioinformatics (I-BioStat), Hasselt University, Agoralaan, 3590 Diepenbeek, Belgium
r t i c l e i n f o
rticle history:eceived 14 July 2011eceived in revised form6 September 2011ccepted 16 October 2011vailable online 24 November 2011
eywords:hinglesonaost-utilityncertainty
llness
a b s t r a c t
Aim: To assess the cost-effectiveness of vaccinating all or subgroups of adults aged 60 to 85 years againstherpes zoster.Methods: A deterministic compartmental static model was developed (in freeware R), in which cohorts canacquire herpes zoster according to their age in years. Surveys and database analyses were conducted toobtain as much as possible Belgian age-specific estimates for input parameters. Direct costs and Quality-Adjusted Life-Year (QALY) losses were estimated as a function of standardised Severity Of Illness (SOI)scores (i.e. as a function of the duration and severity of herpes zoster disease).Results: Uncertainty about the average SOI score for a person with herpes zoster, the duration of protectionfrom the vaccine, and the population that can benefit from the vaccine, exerts a major impact on theresults: under assumptions least in favour of vaccination, vaccination is not cost-effective (i.e. incrementalcost per QALY gained >D 48,000 for all ages considered) at the expected vaccine price of D 90 per dose.At the same price, but under assumptions most in favour of vaccination, vaccination is found to be cost-
effective (i.e. incremental cost per QALY gained <D 5500 for all ages considered). Vaccination of age cohort60 seems more cost-effective than vaccination of any older age cohort in Belgium.Discussion: If the vaccine price per dose drops to D 45, HZ vaccination of adults aged 60–64 years is likelyto be cost-effective in Belgium, even under assumptions least in favour of vaccination. Unlike previousstudies, our analysis acknowledged major methodological and model uncertainties simultaneously and6 dif
presented outcomes for 2. Introduction
Herpes zoster (HZ), also known as shingles or zona, is a viralisease characterized by a painful dermatomal skin rash. The life-ime probability of getting HZ has been estimated at approximately0% [1], but the probability to get HZ increases with age, and
s larger for immunocompromised persons [2]. Also the durationnd severity of the pain increases with age [3,4]. Recently, a vac-ine has been shown to be efficacious in preventing HZ for peopleged 60 and older (Zostavax®) [5]. Several studies evaluated the
ost-effectiveness of HZ vaccination programs [6–15], but theironclusions differed with respect to whether vaccination is likelyo be cost-effective, and at which age vaccination is likely to beAbbreviations: HZ, herpes zoster; QALY, Quality-Adjusted Life-Year; SOI, severityf illness.∗ Corresponding author. Tel.: +32 3 265 28 95; fax: +32 3 265 27 52.
E-mail addresses: [email protected] (J. Bilcke), [email protected]. Marais), [email protected] (B. Ogunjimi), [email protected],[email protected] (N. Hens), [email protected] (P. Beutels).
264-410X/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.oi:10.1016/j.vaccine.2011.10.036
ferent target ages at which vaccination can be considered (ages 60–85).© 2011 Elsevier Ltd. All rights reserved.
(most) cost-effective. Some of these differences are due to country-specific differences in HZ-related health care use and costs. Hence,it is crucial that for each country a separate cost-effectiveness anal-ysis is performed, based on as much country-specific informationas possible.
The only study that has assessed the cost-effectiveness of HZvaccination in Belgium, derived almost all evidence on HZ burdenfrom other countries [6]. In the meantime, data on the burden ofHZ in Belgium have been extracted from representative Belgiandatabases and specific surveys have been conducted and analyzed[7], These newly obtained data provide the opportunity to conducta more relevant analysis of the cost-effectiveness of HZ vaccinationin Belgium.
Furthermore, previous studies have shown that the cost-effectiveness of HZ vaccination is sensitive to the age at whichvaccination is administered [8–11,14,15]. Our study aims toimprove on these studies by estimating a more precise age (in
years) at which vaccination is most cost-effective. Unlike previouspublished analyses our study aims to account for major sources ofmethodological and model uncertainties simultaneously, includ-ing uncertainty about the subgroups which can benefit from the
6 cine 3
vfeod
2
2
[evbcm[sgaedyta
2
fsynit
H2syttgatertwerHdudattHe
sct
76 J. Bilcke et al. / Vac
accine, uncertainty about the interpretation of efficacy outcomesrom the pivotal trial and uncertainty about the choice of vaccinefficacy model, which expresses in our analysis the joint influencef age at vaccination and time since vaccination on the level anduration of vaccine efficacy.
. Methods
.1. Model and methodological choices
A static cohort model was developed (and implemented in R16]) to obtain the age-specific annual number of herpes zosterpisodes and deaths in Belgium with and without vaccination. Indi-iduals belonging to a single age cohort are modelled to transitionetween 3 states (healthy; herpes zoster; and death) in yearlyycles, until everybody is absorbed in the ‘death state’. The life tableethod is used to estimate state membership of the cohort model
17]. Twenty-six age cohorts (60 up to 85) are followed over time,o that the impact of vaccination at different ages can be investi-ated (e.g. only vaccinating at 60, only vaccinating at 61, vaccinatingll people aged 60–65, and so on). The number of herpes zosterpisodes with and without vaccination is compared based on theirect costs (health care payer perspective) and consequences (life-ears lost and Quality-Adjusted Life-Years (QALYs) lost) related tohe episodes. These costs and consequences are discounted at 3%nd 1.5% respectively, according to Belgian guidelines [18].
.2. Estimation of the values of the parameters within the model
Demography – Rather than using Belgian demographic datarom a single or several years, a Gompertz curve is fitted onuch data reflecting average Belgian demography (data fromears 1990 to 2007 obtained from Statbel: http://statbel.fgov.be/l/statistieken/cijfers/bevolking/structuur/leeftijdgeslacht/belgie/
ndex.jsp, accessed 12/07/2011). Life expectancy is extracted fromhe same database, for the latest available year (2007).
HZ incidence and health care use – The age-specific number ofZ hospitalizations in Belgium is available for years 2000 up to007 from a national database; the number of visits for HZ to aentinel system of general practitioners in Belgium is available forears 2006–2008 [7]. By fitting a generalized additive model [19]o these data, estimates are obtained for the average hospitaliza-ion rate for HZ and the average rate at which people consult aeneral practitioner at least once for HZ. A representative surveymong HZ hospitalized patients in Belgium showed that 10.5% ofhese patients did not visit a general practitioner for HZ, but wereither (i) admitted directly through the emergency department, (ii)eferred to the hospital through a specialist doctor, or (iii) hospi-alized for another reason and got HZ in the hospital [7]. Hence,e estimate the age-specific HZ incidence by adding 10.5% of the
stimated average HZ hospitalization rate to the estimated averageate at which people consult a general practitioner at least once forZ. The average age specific HZ mortality rate is based on Belgianeath certificates. The opinion of five experts independently eval-ating death certificates containing HZ as a possible cause of deathiffered substantially. Therefore the cost-utility will be assessed for
scenario least in favour of vaccination (i.e. assuming no deaths dueo HZ), and a scenario most in favour of vaccination (i.e. assuminghe number of HZ-related deaths equals all registered deaths withZ as a possible cause, without the ones for which at least fourxperts agreed they were not due to HZ).
Vaccine coverage and costs – The marginal intervention costs con-ist of the purchasing costs as well as the marginal administrationosts of the vaccine. The Centres for Disease Control and Preven-ion report a price per dose of Zostavax® of $105.9 (D 85.5, 10-pack,
0 (2012) 675– 684
1 dose vial) and $116.7 (D 94.2, 1-pack, single dose 0.65 mL vials);the private sector price per dose of Zostavax® is $153.9 (D 124.3, 10-pack, 1 dose vial) and $161.5 (D 130.4, 1-pack, single dose 0.65 mLvials) (http://www.cdc.gov/vaccines/programs/vfc/cdc-vac-price-list.htm#adult, accessed 21/06/2010). A price per dose of D 90 isused, as this is likely to reflect the price for bulk purchase by apublicly funded program. Because vaccines in this age group aredelivered by general practitioners in Belgium, the administrationcosts were set to D 21.53, based on the cost of one consultation.Based on the experience with influenza and pneumococcal vaccinesfor this target group, vaccine uptake (or coverage) is assumed to be30%, and independent of age and gender. It is varied in scenarioanalyses between 10% and 70%.
Vaccine efficacy – A single trial (the Shingles Prevention Study)measured efficacy of the Zostavax® vaccine in preventing HZ [5].However, three major sources of uncertainty remain regarding theprotective efficacy of the vaccine:
1. The Shingles Prevention Study excluded immunocompromisedpersons and persons with other conditions (e.g. diabetes melli-tus). Hence, vaccine efficacy in such persons is unclear, becauseit was not measured.
2. The Shingles Prevention Study measured efficacy of theZostavax® vaccine against the burden of illness due to HZ (pri-mary endpoint), against the number of HZ cases and againstthe number of post herpetic neuralgia cases [5]. A choice hasto be made about which endpoint to use in a cost-effectivenessanalysis.
3. Vaccine efficacy data from the Shingles Prevention Study for afollow-up period of 5 years are published [5] (and have beenpresented in a poster for a follow-up period of 7 years [20]).Also, vaccine efficacy data for an average follow-up period of 3years are available separately for 5-year age groups (i.e. personsvaccinated at age 59–64, 65–69, and so on) [21]. Bilcke et al.[22] showed that a range of different models fit the data equallywell, but result in very different estimates for vaccine efficacyagainst the number of HZ cases as a joint function of age atvaccination and time since vaccination. For instance, dependingon the chosen vaccine efficacy model, vaccination is expectedto protect persons against HZ for 7 years up to lifelong, andis expected to reduce HZ incidence in people vaccinated forinstance at age 80 by between 35% and 41% for the first year aftervaccination.
We account for these 3 major sources of uncertainty simultane-ously by presenting results of the cost-effectiveness analysis for ascenario least and most in favour of vaccination.
The following assumptions are made on vaccine efficacy for thescenario least in favour of vaccination: We assume only peoplecompliant with the Shingles Prevention Study (i.e. mainly immuno-competent persons) can benefit from vaccination, and we assumevaccination only decreases the number of HZ cases. Choosing thevaccine efficacy model which results in estimates least in favourof vaccination is less straightforward: for instance the vaccine effi-cacy model which predicts lowest efficacy for vaccination at age60, is different to the vaccine efficacy model which predicts lowestefficacy for vaccination at age 70 (for details see ref. [22]). Hence,the following approach was used to determine the vaccine effi-cacy model and associated vaccine efficacy estimates which resultin cost-effectiveness values least in favour of vaccination, for eachage at vaccination separately:
• Step 1: Age-specific cost-effectiveness values are obtained usingthe vaccine efficacy estimates from one vaccine efficacy modelas input, and assuming the choices least in favour of vaccinationfor the other uncertainties (listed in Table 1). This results in 26
J. Bilcke et al. / Vaccine 30 (2012) 675– 684 677
Table 1Uncertainties and the choices most and least in favour of vaccination against herpes zoster (HZ) for each of these uncertainties.
Uncertainties Choice most in favour of vaccination Choice least in favour of vaccination
Population that can benefit from the vaccine Everybody According to inclusion criteria ShinglesPrevention Study [5]
Mortality rate Highest estimate No deathsed on
den o Fig. 1
•
•
sbaiwitaavmvmvlHt
Ft
Severity-of-illness (SOI) score ambulatory HZ episode BasEndpoint for vaccine efficacy BurVaccine efficacy by age at vaccination and time since vaccination See
cost-effectiveness values, for the 26 ages at which vaccination isconsidered (age 60 up to 85).Step 2: Step 1 is repeated, but using the vaccine efficacy estimatesfrom another vaccine efficacy model as input. This is repeated forall other vaccine efficacy models (6 × 7 = 42 models in total (seeref. [22])), so that 42 cost-effectiveness values are obtained foreach of the 26 ages at which vaccination is considered.Step 3: For each of the 26 ages at which vaccination is considered,the greatest of the 42 cost-effectiveness values is determined(i.e. the value least in favour of vaccination). The vaccine effi-cacy estimates (and the associated vaccine efficacy model) whichresulted in this highest cost-effectiveness value, are presentedin Fig. 1 (and Appendix Table A7), for each age at vaccinationconsidered.
The following assumptions are made on vaccine efficacy for thecenario most in favour of vaccination: We assume everybody canenefit from vaccination and we assume vaccination reduces theverage duration and severity of HZ in addition to reducing thencidence of HZ. Vaccine efficacy against the HZ burden of illness
as not found to differ significantly between age groups [23], hencet was only obtained as a function of time since vaccination. Datao estimate vaccine efficacy against HZ burden of illness by timere currently lacking (i.e. severity scores per subject are not avail-ble) and therefore we assume the same waning as estimated foraccine efficacy against the number of HZ cases. For the scenarioost in favour of vaccine no waning is assumed. Fig. 1 shows the
accine efficacy estimates used as input for the cost-effectivenessodel for the scenario most in favour of vaccination. Note that the
accine efficacy against the HZ burden of illness should never beower than the estimated vaccine efficacy against the number ofZ cases, as it accounts for the decrease in SOI score, in addition
o the decrease in the number of HZ cases. Therefore, when the
ig. 1. Estimated vaccine efficacy over time (X-axis) and by age at vaccination (differenthe scenario most in favour of vaccination (right panel).
Drolet et al. data Based on Scott et al. dataf illness due to herpes zoster Number of herpes zoster cases
and Appendix Table A7 See Fig. 1 and Appendix Table A7
no waning models estimate vaccine efficacy against number of HZcases for a particular age at vaccination higher than the vaccineefficacy against burden-of-illness presented in Oxman et al. [5] forthe same age, the incidence estimate was used for that particularage at vaccination to approximate the burden-of-illness estimate(Fig. 1).
As post herpetic neuralgia is defined in the Shingles PreventionStudy in terms of duration (at least 3 months) and severity (painof at least score 3) of HZ, it is implicitly taken into account in themeasure for efficacy against burden of illness due to HZ (see below).Therefore we obtain results for scenarios considering vaccine effi-cacy against number of HZ cases and against HZ burden of illnessonly.
Burden of illness – To assess the impact of vaccine efficacy againstthe burden of illness due to HZ, the HZ-related burden of illness inBelgium, and consequently the average SOI score of a HZ episodein Belgium, need to be estimated. The HZ-related burden of illnessin a population is defined as the average severity-of-illness scorefor HZ (SOI score) among all subjects in that population [5,24]. TheSOI score is defined as the area under the curve of HZ pain (scoredfrom 0 (no pain) to 10 (worst imaginable pain) for the ‘worst pain’question in the Zoster Brief Pain Inventory [25]) plotted against thenumber of days with HZ pain starting from rash onset until 182days [5,24,25].
SOI scores are obtained for a sample of hospitalized HZ patientsin Belgium who reported retrospectively a score for the worst painthey experienced during their HZ episode, and who reported thenumber of days they had HZ [7]. SOI scores are calculated by mul-tiplying the reported pain score by the number of days with HZ
for each respondent. The SOI score for 32% of the respondents isright-censored as they were still ill at the moment they filled in thequestionnaire. On the calculated SOI scores a model is fitted to esti-mate the average SOI score for a hospitalized patient as a functionlines) used as input for the scenario least in favour of vaccination (left panel), and
6 cine 3
oa
iHpnnrECa4sdSttfA
b(aaaoSHp
traa(pfitala
SpiaalgaalQSi
SfiapwmHbh
78 J. Bilcke et al. / Vac
f age, taking the right-censoring into account (see Appendices A1nd A2).
No reliable estimates for SOI score for an ambulatory HZ patientn Belgium are available, and therefore the SOI score for an averageZ case is obtained from the to our knowledge only two availablerospective studies. One study was conducted in a general commu-ity in East London [26,27], the other in Canada [3,28,29]. As it isot clear which of the two studies is most appropriate, two sepa-ate scenarios are presented: one based on data from the study fromast London (‘Scott et al. data’), and another based on data from theanadian study (‘Drolet et al. data’). Drolet et al. [3,28] publishedverage SOI scores for several age groups (n = 261): 248.0 (ages9–60), 250.1 (ages 61–70), 331.3 (ages 71–96), which are used asuch. Scott et al. provided individual data (n = 65) on pain scores anduration of HZ (personal communication), from which individualOI scores are calculated in the same way as in the Shingles Preven-ion Study [5]. Similar as for hospitalized patients, a model is fittedo estimate the average SOI score for an average HZ episode as aunction of age, taking right-censoring into account (see Appendix3).
The average SOI score for ambulatory HZ patients by age cane derived from the estimated SOI score for an average HZ casebased on data from either Drolet et al. or Scott et al., as indicatedbove), the estimated SOI score for a hospitalized patient and thege-specific proportion of HZ patients hospitalized versus treatedmbulatory [7]. The HZ-related burden of illness by age is the sumf the number of hospitalized HZ cases multiplied by the averageOI score for hospitalized patients and the number of ambulatoryZ cases multiplied by the average SOI score for ambulatory HZatients, divided by the age-specific Belgian population.
HZ-related costs – Data on the direct costs related to HZ for hospi-alized and ambulatory patients in Belgium are available from twoecent surveys [7]. A model was fitted to these data to estimate theverage direct cost of HZ for hospitalized and ambulatory patientss a function of SOI score, taking right-censoring into accountsee Appendices A4 and A5). Costs and SOI scores for ambulatoryatients of the survey are likely biased (see ref. [7]), hence in thetted model for cost by SOI score, the age-specific SOI scores fromhe survey are replaced by the estimated age-specific SOI scores formbulatory patients based on observational data from either Dro-et et al. or Scott et al. and the resulting age-specific costs are useds input for the cost-utility analysis.
HZ-related QALY loss – HZ-related QALY loss is obtained from thecott et al. data, as no such estimates are available for the Belgianopulation. The data provided by Scott et al. contained an util-
ty score (EQ5D) measured at baseline, four weeks, three monthsnd six months from study initiation. The total number of qualitydjusted life weeks (QALWs) lived by each patient was estimated byinear interpolation between the observed utility values and inte-rating the area under the curve. The maximum value of this area in
six-month period is 26 weeks and so the difference between therea under the utility curve and 26 results in the number of QALWsost by each patient due to HZ. In order to estimate the averageALW (and thence QALYs) lost due to a HZ episode as a function ofOI score, a model was fitted to these data, taking right-censoringnto account (see Appendix A6).
Parameter values for persons who would have been included in thehingles Prevention Study – A clinician investigated the diagnosiseld of a national database containing all hospitalized HZ patientsnd the questionnaires completed by ambulatory HZ patients asart of a survey conducted in the Belgian population, to assess whoould have been included in the Shingles Prevention Study (i.e.
ainly immunocompetent persons) [7]. Estimation of the averageZ-related SOI score by age and the average cost and QALYs losty SOI score was redone selecting only the persons who wouldave been included in the Shingles Prevention Study. No original0 (2012) 675– 684
diagnoses were available from Scott et al., but additional analy-ses were done on only those persons who were specified to beimmunocompetent.
2.3. Presenting uncertainty in the results
Results are presented for a scenario least and most in favour ofvaccination [30], with respect to the uncertainties involved in thiscost-effectiveness analysis (Table 1).
2.4. Assessing the impact of each uncertainty on the results
Scenarios are presented which differ from the scenarios mostand least in favour of vaccination, in the choice that is made fora particular uncertainty (all uncertainties considered are summa-rized in Table 1). More specifically, for a particular uncertainty, wemade the least favourable choice for the scenario most in favour ofvaccination, and we did the reverse for the scenario least in favourof vaccination (univariate scenario analysis) [30]. Next, we imple-mented multivariate scenario analysis by changing the choices forseveral uncertainties simultaneously [30]. In separate analyses, theimpact of the discount rate, vaccine price and vaccine coverage onthe results are explored. Under the static cohort model approach ofthe economic evaluation of HZ vaccination, and given that there areno costs that scale nonlinearly to vaccination coverage, the incre-mental cost per QALY gained remains constant under any variationof vaccination coverage [31]. Therefore, the budget-impact is ofgreater interest to estimate under various levels of vaccine uptake.Note that there is a direct relation between coverage and the netcosts, implying a doubling of uptake implies a doubling of the netcosts.
3. Results
3.1. Estimated input parameters
Table 2 gives an overview of all estimated input parameters, andassociated models. Fig. 1 shows the vaccine efficacy estimates usedas input for the scenarios most and least in favour of vaccination,the vaccine efficacy models on which these estimates are based areshown in Appendix A7.
3.2. Cost-effectiveness of HZ vaccination
Clinical and economic impact of HZ vaccination is shown inTable 3 for the scenario most and least in favour of HZ vaccina-tion. The scenario which is least in favour of vaccination, results inan incremental cost per QALY gained above D 48,000 for every ageat vaccination considered (Table 3 and Fig. 2). The scenario whichis most in favour of vaccination, results in an incremental cost perQALY gained below D 5500 for every age at vaccination (Table 3 andFig. 3). For the scenario least in favour of vaccination, the incre-mental cost per QALY gained increases substantially with age atvaccination (Table 3 and Fig. 2). This is much less evident for thescenario most in favour of vaccination (Table 3 and Fig. 3).
Most influential uncertain choices:
• The incremental cost per QALY gained is substantially lower whenassuming SOI based on Drolet et al. data instead of on Scott et al.data (Figs. 3 and 4) and when assuming no waning instead ofassuming the lowest and shortest estimated vaccine protection
(Figs. 2–4). It is also lower when assuming everybody can benefitfrom the vaccine instead of only the persons who comply withthe inclusion criteria of the Shingles Prevention Study (mainlyimmunocompetent persons) (Figs. 2 and 4).
J. Bilcke et al. / Vaccine 30 (2012) 675– 684 679
Table 2Estimated mean values of each input parameter as a function of age or severity-of-illness score (continuous), and models used to estimate these mean values.
Input parameter Estimated values by age (yrs) Distributiona Fitted model for the meanb
Age
60 65 70 75 80 85
HZ hospitalization rate per 100,000person-years
17 24 31 44 63 85 Poissonc log( nr of HZ hospitalisationsj
Belgian populationj
)= ˇ0 + f (agej)
HZ rate at least one GP visit per100,000 person-years
548 642 744 884 1075 1287 Poissonc log( nr of visits for HZj
catchment populationj
)= ˇ0 + f
(agej
)
HZ mortality rate per 100,000person-years
0 or 0.06 0 or 0.63 NA NA
SOI hospitalized persond 430 465 499 533 567 601 Exponential E(SOIhospj) = ˇ0 + ˇ1agej
SOI average episoded 30 33 37 42 46 52 Lognormal E(SOIaveragej) = eˇ0+ˇ1agej+�2/2
% HZ hospitalized SPS 41% 60% 56% NA NA% HZ ambulatory SPS 29% 33% 38% NA NASOI hospitalized SPS d 430 465 499 533 567 601 Exponential E(SOIhospj) = ˇ0 + ˇ1agej
SOI average episode SPSd 22 24 27 90 33 37 Lognormal E(SOIaveragej) = eˇ0+ˇ1agej+�2/2
Input parameter Estimated values by severity-of-illness (SOI) score Distributiona Fitted model for the meanb
SOI
430 465 499 533 567 601
Costs hospitalized patientsd (D ) 7990 8242 8493 8745 8996 9247 Log logistic E(Costhospj) = (ˇ0 + ˇ1SOIs) × ϕ�sin(ϕ�)
Input parameter Estimated values by severity-of-illness (SOI) score Distributiona Fitted model for the meanb
SOI
30 42 52 250 333
Costs ambulatory patientsd (D ) 108 121 134 370 429 Log logistic E(Costambs) = eˇ0+ˇ1[log (SOI]s+0.1 × ϕ�sin(ϕ�)
QALYsd 0.12 0.15 0.17 0.44 0.52 Exponential E(QALYs) = ˇ0 + ˇ1SOIˇ2s
Vaccine efficacy Estimated values by age at vaccination and time since vaccination, see Fig. 1 See Appendix Table A1
HZ = herpes zoster; GP = general practitioner; SOI = severity of illness; QALY = Quality-Adjusted Life-Year; SPS = Shingles Prevention Study, denotes input parameters estimatedfor persons who would have been included in this study (mainly immunocompetent persons).
a Best-fitting distribution on (uncensored) observed data (@Risk).b Choice of model based on exploratory data analysis. Model parameters ˇ0, ˇ1, �2, ϕ and � estimated using maximum likelihood (for details see Appendix). E( ) refers to
the expected value of the input parameter that is estimated. Index j refers to the age at vaccination, index s to the SOI score.he Be
s
•
Fause
c Number of hospitalizations or visits is assumed to be Poisson distributed, and tpline, which is a special function used for smoothing.
d Right-censoring taken into account (see Appendix A2).
The older the age at vaccination considered, the more importantthe choice of the vaccine efficacy endpoint becomes: the incre-
mental cost per QALY gained is lower when taking into accountthe additional effect of the vaccine in decreasing the duration andig. 2. Incremental cost per Quality-Adjusted Life-Year (QALY) gained of vaccinatinggainst HZ compared to not vaccinating, for age cohorts 60 up to 85. Results of anivariate scenario analysis based on the scenario least in favour of vaccination:cenarios are shown in which a single estimate or assumption is changed into thestimate or assumption most in favour of vaccination.
lgian or catchment population is used as offset. The function f(agej) is fitted with a
severity of HZ disease (i.e. assuming the vaccine protects againstthe burden of illness due to HZ, Figs. 2–4).
• For a given age at vaccination, different combinations of func-tions used to estimate vaccine efficacy by age at vaccination and
Fig. 3. Incremental cost per Quality-Adjusted Life-Year (QALY) gained of vaccinatingagainst HZ compared to not vaccinating, for age cohorts 60 up to 85. Results of aunivariate scenario analysis based on the scenario most in favour of vaccination:scenarios are shown in which a single estimate or assumption is changed into theestimate or assumption least in favour of vaccination.
680 J. Bilcke et al. / Vaccine 30 (2012) 675– 684
Table 3Avoided disease and economic burden through HZ vaccination of different age cohorts over their lifetime in Belgium, for a scenario most and least in favour of vaccination.The scenario most respectively least in favour of vaccination, is a scenario for which for five uncertain input assumptions, the estimates most in favour, respectively least infavour of HZ vaccination are chosen. Measures involving discounted costs (3%) and effects (1.5%) are marked by a ‘(d)’.
Scenario cohort 60 cohort 70 cohort 80 cohort 85
Avoided HZ cases Most 3515 1478 374 142Least 701 315 82 19
Avoided HZ hospitalizations Most 178 87 25 10Least 21 9 4 1
Avoided PHN casesa Most 678 373 135 64Least 74 51 22 7
Avoided HZ deaths Most 1.22 0.79 0.35 0.19Least 0 0 0 0
Avoided HZ life-years Most 9 5 1 1Least 0 0 0 0
Avoided disease-related costs (D ) Most 2,931,474 1,475,147 521,302 225,881Least 221,760 106,148 38,070 9449
Total vaccine cost (D ) Most 3,697,485 2,791,266 1,467,991 818,296Least 3,697,485 2,791,266 1,467,991 818,296
Incremental costs (D ) Most 766,011 1,316,119 946,690 592,415Least 3,475,725 2,685,117 1,429,921 808,847
Incremental costs (D ) (d) Most 1,759,838 1,669,711 1,026,773 619,514Least 3,505,428 2,695,159 1,432,193 809,239
Incremental QALYs Most 1725 833 280 120Least 77 39 11 3
Incremental QALYs (d) Most 1406 728 257 113Least 72 37 11 3
Cost (D ) per prevented HZ case (d) Most 609 1292 2990 4642Least 5368 8972 18,096 43,969
Cost (D ) per avoided HZ hospitalization (d) Most 12,372 22,198 44,534 67,883Least 176,251 300,287 407,529 935,817
Cost (D ) per life-year gained (d) Most 260,071 402,474 711,205 1,069,356Least NA NA NA NA
Cost (D ) per QALY gained Most 444 1580 3380 4927Least 45,160 69,689 128,003 297,141
Cost (D ) per QALY gained (d) Most 1251 2294 3988 549848,97
H NA: nJan va
•
wgf
TFm
Least
Z: herpes zoster, PHN: post-herpetic neuralgia, QALY: Quality-Adjusted Life-Year,a Based on age-specific proportions of HZ cases being PHN obtained from Albert
time since vaccination, result in (slightly) different incrementalcosts per QALY gained (Fig. 5 for the scenarios least in favour ofvaccination). For the scenarios most in favour of vaccination, thedifference is very small and therefore not shown.The uncertainty about the annual number of deaths due to HZ(Figs. 2 and 3) and the assumed discount rate (Fig. 6) have verylittle impact on the incremental cost per QALY gained.
Impact of vaccine price – The vaccine price needs to decreaseith 50% (i.e. D 45 per dose) for the incremental cost per QALY
ained to be below D 30,000 for vaccinating age cohort 60 up to 64,or the scenario least in favour of vaccination (Fig. 7). Even when
able A7or each age at which vaccination is considered, the models for vaccine efficacy by age aost and least in favour of vaccination against herpes zoster. For details see ref. [22].
Vaccine efficacy Most in favour of vaccination
Endpoint Vaccine efficacy against burden of illness due to herpeAge at vaccination Model
60–63 Independent of age + no waning64–65
66–73
74–76
77–79
80–85
8 73,513 132,220 303,705
ot applicable (because no avoided life-years).n Hoek (personal communication).
the vaccine price decreases by 80% (i.e. D 18 per dose), the incre-mental cost per QALY gained of vaccinating age cohorts 76 up to 85stays above D 30,000 for the scenario least in favour of vaccination(Fig. 7). There is no official threshold ratio in Belgium under whichinterventions are regarded as being cost-effectiveness, the thresh-old of D 30,000 per QALY gained presented here is for illustrativepurposes only.
Impact of assumed vaccine coverage – Given 30% uptake of the
vaccine (which would be similar to the uptake of other vaccinesin this age group in Belgium), roughly between D 0.6 million andD 1.8 million per targeted age cohort is incurred in net direct coststo the health system (Fig. 8). So targeting everyone between 60 andt vaccination and time since vaccination which result in cost-effectiveness values
Least in favour of vaccination
s zoster Vaccine efficacy against the number of herpes zoster casesModel1-exponential age + linear waningpower age + linear waningpower age + 1-exponential waninglogarithmic age + 1-exponential waning1-power age + 1-exponential waning1-exponential age + 1-exponential waning
J. Bilcke et al. / Vaccine 30 (2012) 675– 684 681
Fig. 4. Incremental cost per Quality-Adjusted Life-Year (QALY) gained of vaccinatingagainst HZ compared to not vaccinating, for age cohorts 60 up to 85. Results of amsi
7am
4
4
ucotQmc
iampmDtv
FaRaa
Fig. 6. Incremental cost per Quality-Adjusted Life-year (QALY) gained of vaccinating
can benefit to the same extent from vaccination), and (ii) theydid not do a full sensitivity analysis, but only explored uncertaintythrough univariate sensitivity analysis of the ‘base case’ scenario.When varying each of these base case assumptions one by one, the
ultivariate scenario analysis based on the scenario least in favour of vaccination:cenarios are shown in which more than one estimate or assumption is changednto the estimates or assumptions most in favour of vaccination.
0 years of age and achieving 30% uptake in this age group wouldmount to net costs of about D 20 million the first year, and D 1.8illion per year thereafter.
. Discussion
.1. Cost-effectiveness of HZ vaccination in Belgium
Several key characteristics of HZ disease and the vaccine are stillncertain. This uncertainty impacts on how cost-effective HZ vac-ination will be at various ages. Under assumptions least in favourf vaccination, vaccination is unlikely to be judged cost-effective athe expected vaccine price of D 90 per dose (incremental cost perALY gained >D 48,000). At the same price, but under assumptionsost in favour of vaccination, vaccination is likely to be considered
ost-effective for all age cohorts (60–85).A previous industry-sponsored study for Belgium [6] found
ncremental costs per QALY gained for vaccination of individualsged 60 years and older similar to what we found for the scenarioost in favour of vaccination, despite assuming a higher vaccine
rice (D 114.18 per dose). In sensitivity analysis, they found the
ajority of the incremental costs per QALY gained to remain below30,000. The reason for this difference is likely that (i) almost allheir base case assumptions on uncertain choices and values areery much in favour of vaccination (e.g. they accounted for the
ig. 5. Incremental cost per Quality-Adjusted Life-Year (QALY) gained of vaccinatinggainst herpes zoster compared to not vaccinating, for age cohorts 60 up to 85.esults are presented for the scenario least in favour of vaccination, for six differentssumptions on which functions to use to model vaccine efficacy as a function ofge at vaccination (‘age’) and time since vaccination (‘waning’).
against HZ compared to not vaccinating, for age cohorts 60 up to 85. Results of aunivariate scenario analysis based on the scenario least in favour of vaccination:scenarios are shown for different prices of the vaccine.
additional protection of the vaccine against post herpetic neuralgia,and they assumed everybody (including immunocompromised)
Fig. 7. Incremental cost per Quality-Adjusted Life-Year (QALY) gained of vaccinatingagainst HZ compared to not vaccinating, for age cohorts 60 up to 85. Results of aunivariate scenario analysis based on the scenario least in favour of vaccination:scenarios are shown for different discount rates of costs and benefits.
Fig. 8. Incremental cost of vaccinating against HZ compared to not vaccinating,for age cohorts 60 up to 85. Results are shown for the scenario most in favour ofvaccination, but with the vaccine coverage altered.
6 cine 3
rwffwcq
tCbocar
4
vnttao
•
•
•
4wt
pst
82 J. Bilcke et al. / Vac
esults remained cost-effective. This is similar to what we foundhen doing univariate sensitivity analysis on the scenario most in
avour of vaccination. However, by presenting a scenario least inavour of vaccination, and by doing multivariate scenario analysis,e show a more complete picture for Belgium i.e. that under some
ombinations of uncertain choices, the assumed vaccine price isuite high for its benefits.
It has been hypothesized that widespread childhood vaccina-ion against varicella may induce an increase in HZ incidence [1,32].urrently there is no such varicella vaccination program in Belgium,ut if it would be introduced in the future, the cost-effectivenessf a HZ vaccination program may need to be reconsidered. Theost-effectiveness of introducing a combined program (vaccinationgainst varicella and against HZ) in Belgium has been evaluatedecently [33].
.2. Age at which vaccination is most cost-effective in Belgium
Under the scenarios most and least in favour of vaccination,accination of age cohort 60 seems more cost-effective than vacci-ation of any older age cohort in Belgium. This result is in line withhe results of Pellissier et al. [12], but most other economic evalua-ions of HZ vaccination found vaccination to be most cost-effectivet around age 70 [11,13–15]. This is probably due to one or moref the following reasons:
Most of these previous studies assumed the vaccine to decreasethe average duration and severity of a HZ episode, additional todecreasing the number of HZ cases [11,13–15]. This decrease inaverage duration and severity of a HZ episode was translatedinto a smaller average QALY loss for HZ episodes still occurringafter vaccination (compared to the QALY loss estimated pre-vaccination), but not into a lower average cost for HZ episodes stilloccurring after vaccination (compared to the average cost esti-mated pre-vaccination). However a decrease in disease durationand severity may influence both incremental costs and effects.For example, costs are likely strongly related with the durationof a HZ episode.Vaccine efficacy was not estimated correctly by age at vaccina-tion. As some evidence exists that vaccine efficacy decreases withincreasing age at vaccination, using an average vaccine efficacyfor ages 60–69 results in an overestimation of the efficacy of thevaccine for age cohorts 65–69, and an underestimation for agecohorts 60–65. Indeed, Annemans et al. [6] found vaccination tobe more cost-effective at age 65–69 compared to at age 60–64,assuming the same vaccine efficacy for these age cohorts. Wedealt with this by fitting flexible statistical models to estimatevaccine efficacy as a decreasing function of age at vaccination (inyears).QALY loss was estimated to increase faster with increasing age(for instance by van Hoek et al. [14]). It is however problematicto assess which of the age-specific HZ-related QALY loss estimatesare most reliable (if any), as very different data and methods wereused. Several studies used utility weights measured in personswith post herpetic neuralgia [34], or in eligible people from aclinical trial [35]. We preferred to estimate QALY loss as a functionof SOI score, with SOI scores measured in the general community.
.3. Uncertain characteristics of HZ disease and the vaccine,hich influence the cost-effectiveness of vaccination in Belgium
he most
We found, in line with previous studies [6–15], the duration ofrotection of the vaccine, the vaccine cost, and the duration andeverity of pain and QALYs lost due to HZ to influence the resultshe most. Furthermore, we found the population that can benefit
0 (2012) 675– 684
from the vaccine, and the additional efficacy of the vaccine againstthe duration and severity of disease influential. The two studiesthat explored these uncertainties, also found them to be influential[12,14]. In contrast to our study, most other studies found the ageat vaccination, and the uncertainty about post herpetic neuralgiarelated characteristics (duration, incidence, costs, and vaccine effi-cacy) to be influential. We did not consider post herpetic neuralgiaseparately, but accounted for it by estimating all costs and QALYslost as a function of the duration and severity of HZ disease (SOIscore). And indeed, we found the estimate used for the SOI scoreto be influential on the results. One study [13] found gender to beinfluential on the cost-effectiveness of HZ vaccination. This wasbased on a higher HZ attack rate measured in women comparedto men [36]. In Belgium, the rate at which people visit a GP at leastonce for HZ is slightly higher for women aged 40–90 years oldcompared to men [33], but the HZ hospitalization rate is similarfor men and women [33]. Life expectancy is higher for womenthan men (82 compared to 77 in 2009, http://statbel.fgov.be/nl/statistieken/cijfers/bevolking/sterfte leven/tafels/, accessed30/06/2011). Therefore, also in Belgium the cost-effectiveness ofvaccination against HZ may differ when assessed for women andmen separately.
For the scenario least in favour of vaccination, even when thevaccine costs only D 1, vaccination is found to be quite expensivefor limited benefits (especially for the oldest age groups). This isbecause (i) the administration cost is D 21.53 (i.e. one visit to a gen-eral practitioner), hence the total cost of buying and administeringone vaccine is D 22.53; (ii) the scenario least in favour of vacci-nation assumes that immunocompromised people are vaccinated,but not protected from the vaccine (i.e. vaccines are purchasedand administered, but without yielding benefits in immunocom-promised persons); and (iii) for instance at age 85, the vaccineefficacy is estimated to be about 35% the first year after vaccinationin an immunocompetent person, and to decrease rapidly there-after. Hence, although the scenario least in favour of vaccinationassuming a vaccine price of D 1 per dose is a pessimistic scenario,we believe it is plausible given our current state of knowledge.
5. Limitations
We opted for estimating QALY loss as a function of SOI score,with SOI score based on prospective studies conducted in the gen-eral community. The major downside of this approach is that theonly two studies that measured HZ-related QALY loss and SOI scorein the general community by age, measured quite different averageSOI scores, and that we are not able to explain this difference. Possi-bly, the Scott et al. study failed to capture more severe HZ cases dueto its small sample size. However, we did not have data on QALYloss from the Drolet et al. study, so that the relationship betweenHZ-related QALY loss and SOI score was only based on the Scottet al. data. This likely resulted in an overestimation of age-specificQALY loss when estimating SOI score obtained from the Drolet et al.study. Still, the age-specific QALY loss for an average HZ episodebased on the Scott et al. study is of the same order of magnitudeas the estimates from other studies [9,12,14]. A large prospectivestudy measuring HZ-related (SOI score and) QALY loss in the gen-eral community would be very useful to improve analyses on thissubject.
We ran scenarios assuming only people compliant with theSPS trial inclusion criteria (mainly immunocompetent) can ben-efit from the vaccine. Although such scenarios are realistic because
the vaccine has only been tested in these immunocompetent per-sons, some difficulties arise with this approach. Our scenarios didnot account for differences in life expectancy and average quality-of-life between immunocompetent and immunocompromised
cine 3
peampettepsvvopipp(aa
euiahtfuemttdt[msa
6
lgnvwtfecvoatt
A
UMH
J. Bilcke et al. / Vac
eople. In practice, it remains problematic to estimate lifexpectancy and quality-of-life separately for immunocompetentnd immunocompromised people, because (i) it is difficult to deter-ine who is immunocompromised, and (ii) immunocompetent
ersons can become immunocompromised and vice versa. How-ver, when considering vaccination of elderly (or any interventionargeting immunocompromised and frail elderly), such informa-ion would be very valuable. Since observed life-expectancy is atvery age a weighted average of life-expectancy in immunocom-etent and immunocompromised persons, our approach leads to alight underestimation of the life-years gained for the scenarios ofaccinating immunocompetent persons only. For the scenarios ofaccinating immunocompromised persons, the reverse situationccurs. The influence of this aspect increases with the increasingroportion of immunocompromised persons (i.e. also with increas-
ng age at vaccination). Also, average age-specific quality-of-life (foreople without HZ) was not assessed separately for immunocom-etent and immunocompromised persons and was assumed to be 1as no age-specific utilities exist in Belgium), which may have led ton overestimation of the QALYs gained, especially for the scenariosssuming everybody can benefit from vaccination.
Due to the use of flexible statistical models to estimate param-ters as a function of age and SOI score continuously, parameterncertainty is not straightforward to take into account correctly
n our analysis. Indeed, difficulties in estimating the uncertaintyround flexible functions have been discussed [37]. Therefore, weave restricted the analysis to estimating accurately the mean func-ions and did not account for the uncertainty surrounding theseunctions. As such, the uncertainty presented in the results is annderestimation of all uncertainty involved. However, for this cost-ffectiveness analysis, we believe obtaining accurate age-specificean estimates to be more important than being able to quan-
ify parameter uncertainty. This is because it seems highly likelyhat in agreement with observations for other vaccine preventableiseases, data source and model uncertainty outweigh parame-er uncertainty for cost-effectiveness analysis of HZ vaccination30,38], and we did explore the impact of model, data source and
ethodology uncertainty on the results. However, future researchhould focus on improving techniques to estimate uncertaintyround estimated flexible functions.
. Conclusion
At a vaccine price of about D 45 per dose, vaccination wouldikely be considered cost-effective (i.e. incremental cost per QALYained <D 30,000) in Belgium for age cohorts 60–64, under a sce-ario least in favour of vaccination. That is, assuming that theaccine only decreases the number of HZ cases, that vaccine efficacyanes fast (<9 years of protection), that only people complying with
he inclusion criteria of the Shingles Prevention Study trial benefitrom vaccination (i.e. mainly immunocompetent), and that the low-st estimate for average QALY loss and SOI score available from twoommunity studies applies. Under a wide range of assumptions,accination of age cohort 60 is more cost-effective than vaccinationf any older age cohort in Belgium. Unlike previous studies, ournalysis acknowledged major methodological and model uncer-ainties simultaneously and presented outcomes for 26 differentarget ages at which vaccination can be considered (ages 60–85).
cknowledgments
We thank Judith Breuer (Department of infection and Immunity,niversity College London, UK) and John Edmunds (Centre for theathematical Modelling of Infectious Diseases, London School ofygiene and Tropical Medicine, UK) for providing data from the
0 (2012) 675– 684 683
Scott et al. study. We also thank two anonymous referees for theirthoughtful comments on a previous version of this manuscript.
This study was co-funded by the Belgian Health Care KnowledgeCenter, the Institute for the Promotion of Innovation by Science andTechnology in Flanders (strategic basic research project, Simulationmodels of infectious disease transmission and control processes(SIMID)), and the IAP research network nr P6/03 of the BelgianGovernment (Belgian Science Policy). NH acknowledges supportfrom the University of Antwerp scientific chair in Evidence-BasedVaccinology, financed in 2009–2011 by a gift from Pfizer.
Appendix A. Appendix
A.1. Modeling SOI for a hospitalized person as a function of age
The SOI data for hospitalized persons were fitted best by anexponential distribution (according to the �2 goodness-of-fit statis-tic), of which the expected value (the mean) is 1/l, with l being therate parameter. The expected value was then replaced by a func-tion of age. Different functions were fitted, and the one with thebest fit (smallest squared error) was retained. This was the functionˇ0 + ˇ1 Age, implying a linear function between SOI and age. Theparameters ˇ0 And ˇ1 are estimated using maximum likelihood,taken right-censoring into account (see below).
A.2. Accounting for right-censoring
Right-censoring is taken into account as follows: If x follows adistribution f(x ; �) where � is a list of parameters and there is nocensoring, the maximum likelihood estimates of � are found bymaximizing:
L =n∏
i=1
fx(xi; �)
where fx(xi ; �) is the density function of x evaluated at the pointxi given �. If an observed value xi is right censored, we know thatthe true value of the observation is at least xi. This information isadded to the likelihood function as 1 − Fx(xi ; �), where Fx(xi ; �) isthe distribution function of x at the point xi given �. The likelihoodfunction that should be estimated then becomes:
L =∏
uncensored
fx(xi; �)∏
censored
1 − fx(xi; �)
A.3. Modeling SOI for an average HZ episode as a function of age
The SOI data for an average HZ episode were fitted best by a log-normal distribution (according to the �2 goodness-of-fit statistic),of which the expected value (the mean) is e�+(�z/2), with � beingthe mean and � the standard deviation of SOI’s natural logarithm.If a random variable x follows a lognormal distribution, we knowthat ln(x) follows a normal distribution with parameters � and �such that the expected value of ln(x) is �. � was then replaced by afunction of age. Different functions were fitted, and the one with thebest fit (smallest squared error) was retained. This was the functionˇ0 + ˇ1 Age+ˇ2 Gender. When related back to the lognormal distri-bution, this results in an exponential relationship between SOI and
age and a different SOI differs for men and women (Gender is 1 forwomen, and 0 for men): SOI = eˇ0+ˇ1 Age+ˇ2 Gender+(�z/2) The param-eters ˇ0, ˇ1 , ˇ2 and � are estimated using maximum likelihood,taken right-censoring into account (see A2).
6 cine 3
AS
gwt3kaoefblfwThi
AS
astflbeep
A
nTfQu(
R
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
[
84 J. Bilcke et al. / Vac
.4. Modeling cost for a HZ hospitalized patient as a function ofOI
The cost data for hospitalized patients were fitted best by a loglo-istic distribution (according to the �2 goodness-of-fit statistic), ofhich the expected value (the mean) is aϕ�/sin(ϕ�), with a being
he scale parameter, 1/ϕ the shape parameter and � the constant.14. If a random variable x follows a log-logistic distribution, wenow that ln(x) follows a logistic distribution with parameters �nd s such that the expected value of ln(x) is �. The parametersf the log-logistic distribution are then a = e� and (1/ϕ) = (1/s). Thexpected value of the logistic distribution, �, was then replaced by aunction of age. Different functions were fitted, and the one with theest fit (smallest squared error) was retained. This was the function
n(ˇ0 + ˇ1 SOI). When related back to the log-logistic distributionunction, this results in a linear relationship between costs and SOIith the expected cost being: E(Cost) = (ˇ0 + ˇ1 SOI)(ϕ�/sin(ϕ�)).
he parameters ˇ0, ˇ1 and � are estimated using maximum likeli-ood with the statistical software package R, taken right-censoring
nto account (see A2).
.5. Modeling cost for a HZ ambulatory patient as a function ofOI
The cost data for hospitalized patients were fitted best by loglogistic distribution (according to the �2 goodness-of-fittatistic). The same approach as described in A4 was usedo model estimate cost as a function of SOI. The best-fittingunction was ln(ˇ0 + ˇ1SOI). When related back to the log-ogistic distribution function, this results in a power relationshipetween costs and SOI with the expected cost being: E(Cost) =(ˇ0+ˇ1 log(SOI+0.1))(ϕ�/ sin(ϕ�)). The parameters ˇ0, ˇ1 and � arestimated using maximum likelihood with the statistical softwareackage R, taken right-censoring into account (see A2).
.6. Modeling QALW loss as a function of SOI
The Quality of life weeks lost were fitted best by an expo-ential distribution (according to the �2 goodness-of-fit statistic).he same approach was used as described in A1. This best fittingunction was ˇ0 + ˇ1 SOIˇz , implying a power function betweenALWs lost and SOI. The parameters ˇ0, ˇ1 and ˇ2 are estimatedsing maximum likelihood, taken right-censoring into accountsee A2).
eferences
[1] Hope-Simpson RE. The nature of herpes zoster: a long-term study and a newhypothesis. Proc R Soc Med 1965;58:9–20.
[2] Wareham DW, Breuer J. Herpes zoster. BMJ 2007;334(7605):1211–5.[3] Drolet M, Brisson M, Levin MJ, Schmader KE, Oxman MN, Johnson RW,
et al. A prospective study of the herpes zoster severity of illness. Clin J Pain2010;26(8):656–66.
[4] Helgason S, Petursson G, Gudmundsson S, Sigurdsson JA. Prevalence of pos-therpetic neuralgia after a first episode of herpes zoster: prospective studywith long term follow up. BMJ 2000;321(7264):794–6.
[5] Oxman MN, Levin MJ, Johnson GR, Schmader KE, Straus SE, Gelb LD, et al. Avaccine to prevent herpes zoster and postherpetic neuralgia in older adults. NEngl J Med 2005;352(22):2271–84.
[6] Annemans L, Bresse X, Gobbo C, Papageorgiou M. Health economic evaluationof a vaccine for the prevention of herpes zoster (shingles) and post-herpeticneuralgia in adults in Belgium. J Med Econ 2010;13(3):537–51.
[7] Bilcke J, Ogunjimi B, Marais C, De Smet F, Callens M, Callaert K, et al. The health
and economic burden of chickenpox and herpes zoster in Belgium. EpidemiolInfect, in press.[8] Brisson M, Pellissier JM, Camden S, Quach C, De Wals P. The potential cost-effectiveness of vaccination against herpes zoster and post-herpetic neuralgia.Hum Vaccines 2008;4(3):238–45.
[
[
0 (2012) 675– 684
[9] Edmunds WJ, Brisson M, Rose JD. The epidemiology of herpes zoster andpotential cost-effectiveness of vaccination in England and Wales. Vaccine2001;19(23–24):3076–90.
10] Hornberger J, Robertus K. Cost-effectiveness of a vaccine to prevent her-pes zoster and postherpetic neuralgia in older adults. Ann Intern Med2006;145(5):317–25.
11] Najafzadeh M, Marra CA, Galanis E, Patrick DM. Cost effectiveness of herpeszoster vaccine in Canada. Pharmacoeconomics 2009;27(12):991–1004.
12] Pellissier JM, Brisson M, Levin MJ. Evaluation of the cost-effectiveness in theUnited States of a vaccine to prevent herpes zoster and postherpetic neuralgiain older adults. Vaccine 2007;25(49):8326–37.
13] Rothberg MB, Virapongse A, Smith KJ. Cost-effectiveness of a vaccine to pre-vent herpes zoster and postherpetic neuralgia in older adults. Clin Infect Dis2007;44(10):1280–8.
14] van Hoek AJ, Gay N, Melegaro A, Opstelten W, Edmunds WJ. Estimating thecost-effectiveness of vaccination against herpes zoster in England and Wales.Vaccine 2009;27(9):1454–67.
15] van Lier A, van Hoek AJ, Opstelten W, Boot HJ, de Melker HE. Assessing thepotential effects and cost-effectiveness of programmatic herpes zoster vacci-nation of elderly in the Netherlands. BMC Health Serv Res 2010;10:237.
16] Ihaka R, Gentleman RR. A language for data analysis and graphics. J ComputGraphical Stat 1996;5(3):299–314.
17] Barendregt JJ. The half-cycle correction: banish rather than explain it. Med DecisMaking 2009;29(4):500–2.
18] Cleemput I, Van Wilder P, Vrijens F, Huybrechts M, Ramaekers D. Guide-lines for pharmacoeconomic evaluations in Belgium. KCE reports 78C; 2008.http://kce.fgov.be/index en.aspx?SGREF=5223&CREF=11069.
19] Wood SN. Generalized additive models: a introduction with R. CRC/Chapman& Hall; 2006.
20] Schmader K, Oxman M, Levin M, Betts RF, Morrison VA, Gelb LD, et al. Persis-tence of zoster vaccine efficacy. Durham: ICAAC/IDSA; 2008.
21] Ahnn S. FDA statistical review and evaluation: zostavax. Document for theVaccines and Related Biological Products Advisory Committee; 2005.
22] Bilcke J, Ogunjimi B, Hulstaert F, Van Damme P, Hens N, Beutels P. Estimatingthe age-specific duration of herpes zoster vaccine protection: a matter of modelchoice? Vaccine 2011, doi:10.1016/j.vaccine.2011.09.079.
23] Oxman MN, Levin MJ. Vaccination against herpes zoster and postherpetic neu-ralgia. J Infect Dis 2008;197(Suppl. 2):S228–36.
24] Chang MN, Guess HA, Heyse JF. Reduction in burden of illness: a new efficacymeasure for prevention trials. Stat Med 1994;13(18):1807–14.
25] Coplan PM, Schmader K, Nikas A, Chan IS, Choo P, Levin MJ, et al. Develop-ment of a measure of the burden of pain due to herpes zoster and postherpeticneuralgia for prevention trials: adaptation of the brief pain inventory. J Pain2004;5(6):344–56.
26] Scott FT, Johnson RW, Leedham-Green M, Davies E, Edmunds WJ, Breuer J.The burden of herpes zoster: a prospective population based study. Vaccine2006;24(9):1308–14.
27] Scott FT, Leedham-Green ME, Barrett-Muir WY, Hawrami K, Gallagher WJ, John-son R, et al. A study of shingles and the development of postherpetic neuralgiain East London. J Med Virol 2003;70(Suppl. 1):S24–30.
28] Drolet M, Brisson M, Schmader K, Levin M, Johnson R, Oxman M, et al. Predictorsof postherpetic neuralgia among patients with herpes zoster: a prospectivestudy. J Pain 2010;11(11):1211–21.
29] Drolet M, Brisson M, Schmader KE, Levin MJ, Johnson R, Oxman MN, et al. Theimpact of herpes zoster and postherpetic neuralgia on health-related qualityof life: a prospective study. CMAJ 2010;182(16):1731–6.
30] Bilcke J, Beutels P, Brisson M, Jit M. Accounting for methodological, structural,and parameter uncertainty in decision-analytic models: a practical guide. MedDecis Making 2011;31(4):675–92.
31] Beutels P. Economic evaluations of hepatitis B immunization: a global reviewof recent studies (1994–2000). Health Econ 2001;10(8):751–74.
32] Brisson M, Gay NJ, Edmunds WJ, Andrews NJ. Exposure to varicella boostsimmunity to herpes-zoster: implications for mass vaccination against chick-enpox. Vaccine 2002;20(19–20):2500–7.
33] Bilcke J, Marais C, Ogunjimi B, van Hoek AJ, Lejeune O, Callens M, et al.Cost-utility of vaccination against chickenpox in children and against herpeszoster in elderly in Belgium. KCE reports 151; 2011. http://www.kce.fgov.be/index en.aspx?SGREF=5211&CREF=18897.
34] Oster G, Harding G, Dukes E, Edelsberg J, Cleary PD. Pain, medication use,and health-related quality of life in older persons with postherpetic neuralgia:results from a population-based survey. J Pain 2005;6(6):356–63.
35] Bala MV, Wood LL, Zarkin GA, Norton EC, Gafni A, O’Brien B. Valuing outcomes inhealth care: a comparison of willingness to pay and quality-adjusted life-years.J Clin Epidemiol 1998;51(8):667–76.
36] Insinga RP, Itzler RF, Pellissier JM, Saddier P, Nikas AA. The incidence ofherpes zoster in a United States administrative database. J Gen Intern Med2005;20(8):748–53.
37] Briggs A, Claxton K, Sculpher M. Decision modelling for health economic eval-uation. New York: Oxford University Press; 2006.
38] Bilcke J, Beutels P. Reviewing the cost effectiveness of rotavirus vaccination: theimportance of uncertainty in the choice of data sources. Pharmacoeconomics2009;27(4):281–97.
