Cost–effectiveness models of pneumococcal conjugate vaccines: variability and impact of modeling assumptions

Abstract

Currently, 13-valent pneumococcal conjugate vaccine (PCV); and ten-valent PCV vaccine are marketed. Neither vaccine obtained regulatory approval based on efficacy trials, but instead were approved based on a surrogate end point: immunogenicity data measuring effective antibody levels. Therefore, direct measures of efficacy were unavailable at the time economic analyses were conducted. The authors systematically reviewed cost–effectiveness studies of ten-valent PCV and 13-valent PCV from the literature to analyze the methodologies and compare the assumptions made about vaccine effectiveness. The following three inputs were found the most variant across analyses: efficacy against acute otitis media; inclusion of indirect effects; and cross protection. These assumptions are discussed with regard to the validity of supporting data and implications on decision-making.

Keywords::

• acute otitis media
• cost–effectiveness analysis
• cross protection
• herd effects
• invasive pneumococcal disease
• nontypable Haemophilus influenzae
• PCV7
• PCV10
• PCV13
• PHiD-CV
• pneumococcal conjugate vaccine
• pneumonia

Cost–effectiveness analyses are useful for predicting potential costs and benefits, and can assist stakeholders in making informed decisions about the value of proposed interventions [101]. Mathematical models are often employed for this purpose in order to extrapolate evidence from efficacy studies of a new intervention to populations where epidemiology and medical practice may differ, and to project the impact of the intervention over longer periods of time. However, models are inherently uncertain, either because the modeling methods are insufficient to perfectly replicate reality or because the input parameters used to inform the models are themselves uncertain. In order to compensate for these issues, outcomes researchers define a base case that represents the assumptions believed to be most strongly supported and then employ thorough sensitivity analyses, and tests to verify the impact of varying the parameters that have the greatest uncertainty and/or impact. Nonetheless, the base case analysis should always be supported by the most competent and reliable scientific evidence available at the time of the analysis.

Streptococcus pneumoniae is a significant cause of infectious disease in adults and children resulting in significant healthcare costs; however, vaccination with a pneumococcal conjugate vaccine (PCV) has been proven to be a very effective preventive strategy [1,2]. The seven-valent PCV (PCV7), which includes serotypes 4, 6B, 9V, 14, 18C, 19F and 23F conjugated to nontoxic diphtheria toxoid CRM₁₉₇, was introduced for use in young children and was shown to prevent invasive pneumococcal disease (IPD), pneumonia and acute otitis media (AOM) in this age group [3,4]. Recently, two higher-valent PCVs have been developed and are currently marketed: a ten-valent PCV (PCV10) and a 13-valent PCV (PCV13).

PCV10 is based on an 11-valent PCV (PCV11) in which all 11 serotypes (1, 3, 4, 5, 6B, 7F, 9V, 14, 18C, 19F and 23F) were conjugated to Haemophilus influenzae protein D. Following the POET study assessing the impact of PCV11 on AOM [5], the formulation of the vaccine was modified – the saccharide for serotype 3 was removed, the saccharide for serotype 18C was conjugated to tetanus toxoid, and the saccharide for serotype 19F was conjugated to diphtheria toxoid [102]. PCV13 is based on PCV7, and like PCV7, the serotypes in PCV13 are individually conjugated to nontoxic diphtheria toxoid CRM₁₉₇. In addition to the serotypes in PCV7, PCV13 also includes serotypes 1, 3, 5, 6A, 7F and 19A [103].

Neither PCV10 nor PCV13 obtained regulatory approval on the merits of efficacy trials, rather approvals were granted based on immunogenicity data measuring effective antibody levels, which are used as a surrogate marker to inform potential efficacy. Therefore, the efficacies of the newer vaccines are still unknown although emerging effectiveness data suggest that PCVs continue to impact disease caused by S. pneumoniae serotypes included in the vaccines [2,6,104,106].

Decision makers in many countries are deciding whether to select PCV10 or PCV13 as part of their national immunization program (NIP). To inform these decisions, several cost–effectiveness analyses have been conducted to evaluate the potential value of PCV10 and/or PCV13. In many countries, these analyses play an influential role in the selection of the vaccine for use in an NIP.

PCV cost–effectiveness analyses estimate cases of disease (IPD, pneumonia and AOM) that are averted by immunization based on assumptions of vaccine effectiveness. Because the effectiveness of new vaccines in a specific country is unknown owing to several factors, including lack of efficacy studies, variable local distribution of serotypes, levels of vaccination uptake, serotype replacement and others, researchers must make choices regarding assumptions about the local effectiveness of vaccines. Because there is much variability in the results of recent cost–effectiveness models of higher-valent PCVs, the authors endeavored to examine the factors that lead to this variability. In the authors’ review, cost–effectiveness studies of PCV10 and/or PCV13 were examined in order to compare the variability of assumptions in vaccine evidence. Key assumptions are further discussed with regard to the validity of supporting data and implications on decision making.

Methods

Literature search

The Medline database was searched using PubMed, with the keywords ‘cost’ AND ‘pneumococcal conjugate vaccine’; ‘Prevenar 13’ AND ‘cost’; and ‘Synflorix’ AND ‘cost’. Results were limited to articles published within the past 5 years. Two of the authors (RA Farkouh and RM Klok) assessed the list of articles for inclusion in the review. Articles assessing cost–effectiveness of PCVs in a pediatric population, and which specifically examined PCV10 and/or PCV13 use in an NIP, were included.

Data analysis

Information regarding the effectiveness parameters for each vaccine was extracted from the methods sections of each article including: country of analysis; funding source; model design; vaccines evaluated; vaccination schedule (number and timing of doses); disease states included; perspective of analysis; year of epidemiologic data; vaccine effectiveness estimates for direct protection against IPD, pneumonia and AOM; herd protection against IPD, pneumonia and AOM; inclusion of cross protection to noncovered serotypes; the impact of vaccination on antibiotic resistance; and inclusion of nontypeable H. influenzae (NTHi) impact. Input data were compared to identify points of general agreement and areas where assumptions varied significantly. Source data referenced in support of assumptions were reviewed, and the robustness and applicability of that data were described. Our analysis focused on base case assumptions; however, variation in the data considered in sensitivity analysis was examined and discussed when appropriate. Base case cost–effectiveness results from each analysis were gathered to explain the impact of assumptions on results. Input data specific to local circumstances, such as disease epidemiology, serotype coverage and costs were not included as variability is explained by localization. Quality-of-life (utility) measures for specific health states, both acute disease and sequelae, were considered out of scope for this analysis.

Results

Literature search results

The literature search was completed on 8 December 2011. The terms ‘cost’ AND ‘pneumococcal conjugate vaccine’ yielded 113 publications. Of these, 33 abstracts indicated that a cost–effectiveness estimate was performed. Of these 33, 18 were excluded; one because it was an estimate in an adult population; one because it looked at a catch-up program; 15 because they estimated the cost of PCV7 only; and one because it duplicated information in a separate publication. The terms ‘Prevenar 13’ AND ‘cost’ and ‘Synflorix’ AND ‘cost’ yielded one article each. This article was the same for both searches and was included in the review [7]. In total, 16 articles were included in the final analysis (Table 1) [7–22].

Model comparisons

Most studies compared both PCV10 and PCV13 with PCV7. One study included nine-valent PCV (an unmarketed vaccine [20]), two compared PCV10 with PCV7 [8,13], and one compared PCV13 with PCV7 [11]. Seven were cohort-based models [15–21], nine studies used some variation of a population-based model [7–14,22], and most were from the healthcare perspective (Table 1).

The epidemiologic data used in the analyses, specifically serotype coverage, and disease incidence, were variable across studies (Table 2) [3,5,7–26]. Serotype coverage was either taken from a time period prior to widespread PCV7 use or the most current serotype coverage data were utilized. Serotype coverage for IPD was obtained from national surveillance programs, where the distribution of serotypes causing invasive disease is assumed to represent the circulating serotypes in the population. The etiology of pneumonia and AOM is rarely known and can be due to many different pathogens. In all analyses, the distribution of serotypes causing pneumococcal pneumonia was assumed to be the same as those serotypes observed to cause IPD. For AOM, four different methods were used to derive the serotype coverage data. Seven studies used IPD serotype coverage as a proxy for AOM [6,9–11,14,19,21]. One analysis [22] used Israeli nasopharyngeal carriage (NPC) data as a proxy for AOM etiology. Four analyses [7,13,17,18], used the distribution of serotypes taken from a meta-analysis of pneumococcal isolates collected from 1994 to 2000 [27] as a proxy for local representation of current AOM etiology. And finally, two studies obtained data from a local study where the etiology of AOM cases had been examined [8,15].

All models used a consistent approach to extrapolate the effectiveness of PCV10 and PCV13 against IPD, by basing the effectiveness of the newer vaccines on the serotype distribution in the respective country for which the analysis was performed (Table 2). In general, the assumption of IPD effectiveness against covered serotypes for PCV7 was obtained from either the pivotal Phase III trial [3] (94% intent to treat and 97% per protocol) or from the CDC surveillance (96% [28]). All models including PCV10 and PCV13 apply the selected value to the extent of serotype coverage in the newer vaccines to account for the impact of broader serotype coverage. Overall, 15 of 16 papers only included invasive disease caused by S. pneumoniae [7–17,19–22], whereas Knerer et al. also included invasive disease caused by NTHi [18]. In this analysis, the efficacy of PCV10 against NTHi invasive disease was assumed to be equivalent to the efficacy against AOM caused by NTHi, estimated from the POET study using PCV11 [5,18].

All models generally took a similar approach to estimation of PCV impact on pneumonia. PCV10 and PCV13 effectiveness was assumed to apply to a measure of all-cause pneumonia. Pneumonia effectiveness was applied based on that observed with PCV7 use (4.3–39.0% [4,29]). This was then extrapolated to PCV10 and/or PCV13 in proportion to the serotype coverage of the higher-valent vaccines compared with the serotype coverage of PCV7.

Assumptions of vaccine efficacy against all-cause AOM varied most widely across analyses and are presented in Table 2. One study did not evaluate AOM [20], another obtained efficacy values directly from clinical trials of similar vaccines without serotype adjustment [16], and 14 studies adjusted data from clinical trials based on the relative serotype coverage of the higher-valent vaccine relative to PCV7 [7–15,17–19,21,22]. Nine studies attributed effectiveness against NTHi for AOM or other diseases to PCV10 [7–10,12,13,16–18], based on the results of the POET trial [5]. In the 14 studies that calculated the efficacy for PCV10 and PCV13, two approaches were consistently used to estimate AOM effectiveness. The first was a serotype coverage extrapolation consistent with the methods used to derive estimatesfor all-cause pneumonia. Generally, the formula is as follows:

Where, VE is vaccine efficacy and ST is serotype coverage. Some of these analyses [8,12] have appended an additional effect for NTHi (3–4%) to the serotype extrapolated value for PCV10, based on a calculation of the effectiveness of PCV11 against NTHi AOM multiplied by the proportion of cases that are caused by NTHi.

The second method used an etiology-weighted approach, which includes effectiveness against NTHi and serotypereplacement of noncovered pneumococcal serotypes. The equation is as follows:

Where, VE_VT is vaccine efficacy against vaccine serotypes; P_VT is the percentage of AOM cases due to vaccine serotypes; VE_NVT is the vaccine efficacy against nonvaccine serotypes; P_NVT is the percentage of AOM cases due to nonvaccine serotypes; VE_NTHi is vaccine efficacy against disease caused by NTHi; and P_NTHi is the percentage of AOM cases due to NTHi.

The proportion of AOM cases attributable to specific pathogens is typically assumed to be evenly distributed across S. pneumoniae, NTHi and other [30], where ‘other’ is omitted from the equation because no efficacy of PCV10 or PCV13 is assumed against other organisms. Vaccine efficacy against vaccine serotypes is either obtained from the POET study of PCV11 (57.6% [5]) or the FinOM study of PCV7 (57% [23]), and efficacy against NTHi (35.6%) is always obtained from the POET study of PCV11 [5]. Some models have included a negative impact of vaccination against nonvaccine serotypes of -33% to offset potential serotype replacement, and others included a -11% effect against NTHi to account for a potential serotype replacement effect of NTHi as the causative agent in all-cause AOM [23].

Indirect effects are defined here as the net effect of two opposing forces: disease reduction in unvaccinated individuals through ‘herd effects’, and increased disease caused by nonvaccine serotypes through ‘serotype replacement’. Herd effects were considered in 11 base cases (Table 2) [8,9,11–14,17–19,21,22], three analyses considered herd effects in sensitivity analysis [10,16,20], and two analyses did not consider herd effects [7,15]. Some models restricted herd effects to occur against IPD only, whereas some included pneumonia and/or AOM along with IPD. Herd effects were serotype extrapolated for PCV10 and PCV13 based on either observed herd effects for PCV7 or a fixed herd effect was applied across all vaccines equally. Two analyses included herd effects for PCV7 and PCV13, but omitted herd effects for PCV10 [12,14]. Further details of the values used in the analyses are available in the (Supplementary Table 1). Indirect effects were most often extrapolated from: US data, specifically from the CDC Active Bacterial Core Surveillance program as PCV7 had been in use for the longest duration in the USA [8,9,11,13,14,17,18,20]; more recent European data [12,19]; country-specific data [10]; or as a fixed percentage reduction in incidence of disease in nonvaccinated age groups [16,21,22]. Serotype replacement for IPD was considered in four of 16 papers [7,8,11,22]; however, serotype replacement for pneumonia is not explicitly modeled in any of the papers. In five of 16 analyses [7,8,13,17,18], serotype replacement was entered into the calculations of AOM effectiveness as discussed above.

Cross protection, the ability for vaccine serotypes 6B and 19F to protect against related serotypes 6A and 19A, respectively, which are not included in PCV7 or PCV10, was also considered in five analyses [7,8,13,17,18] in the base case, and by one study in a sensitivity analysis (Table 2) [10]. No models quantitatively included costs or outcomes related to antibiotic resistant infections.

Model results are highly variable, ranging from cost-saving for PCV10 to cost-saving for PCV13 (Table 2). All analyses, including no vaccination or PCV7 as the basis of comparison, found higher-valent vaccines (PCV9, PCV10 and PCV13) to be at a minimum cost effective and cost saving in several instances.

Expert commentary

Our review of cost–effectiveness models finds that the models used in PCV cost–effectiveness estimation are generally very similar in structure. Models often take a similar approach to estimates of IPD and pneumonia effectiveness of the higher valent vaccines. However, they differ significantly in their assumptions surrounding estimated vaccine efficacy against AOM, incorporation of indirect effects and cross protection. Although effectiveness data are beginning to emerge for PCV13 and PCV10, only immunogenicity and safety data were available for the higher valent PCVs when these studies were conducted. Therefore, it was necessary to assume a certain level of effectiveness for PCV10 and/or PCV13, in order to estimate the value of the vaccines used in a NIP.

Although all data on disease states will impact the results of a model, assumptions of effectiveness against AOM have been identified as a key economic driver in PCV cost–effectiveness analyses [31]. Although the mean cost per episode is relatively low, the incidence rate of AOM is high, resulting in significant overall healthcare costs, making assumptions around efficacy against AOM a very influential parameter in cost–effectiveness models. The assumed impact of PCV10 effectiveness on NTHi is a crucial component to the cost–effectiveness of pneumococcal vaccines and largely responsible for any estimated cost and outcome advantages for PCV10 compared with PCV7 and PCV13.

The assumed effectiveness for PCV10 and PCV13 to prevent AOM is frequently derived from the POET and FinOM trials conducted with PCV11 and PCV7, respectively [5,23]. Although both the POET and FinOM studies reported similar efficacy of PCVs versus AOM caused by vaccine-covered serotypes, the reported efficacies versus all AOM cases differ. The reported efficacy of PCV11 from the POET trial applies to a sample of more severe AOM cases because of the study’s requirement for a referral to an otolaryngologist to confirm the diagnosis of AOM. In the words of the study’s primary investigator, “Our study was not designed to capture every AOM episode, but only the most ‘disturbing’ cases that resulted in referral to ENT specialists” [5]. In contrast, the FinOM study was designed to capture all cases of AOM. As most pharmacoeconomic models consider the vaccine’s impact on all-cause AOM, using the results of the FinOM trial as a basis for extrapolation might be most appropriate to match the epidemiologic input data. However, one might also argue that the more severe cases of AOM are those with the greatest pharmacoeconomic impact. Therefore, application of the epidemiology of severe otitis media with corresponding effectiveness data may allow for more complete estimation of the value of PCVs, where such data are available. Models building on the most comprehensive information on mild and severe AOM should give the most precise estimates.

Several studies assumed effectiveness against NTHi in the estimation of PCV10 AOM effectiveness, in addition to protection against S. pneumoniae. The H. influenzae protein D carrier protein in PCV10 is suggested to confer protective efficacy against disease caused by NTHi [5], which is a significant pathogen implicated in AOM. Evidence in support of this assumption is derived from the POET study, which demonstrated efficacy of PCV11 against NTHi AOM in a subset of the most disturbing AOM cases. However, studies with PCV10 have shown a lower immune response to protein D than that observed with PCV11 [32,105]. The clinical significance of protein D in PCV10 is still unknown, as no efficacy or effectiveness data of PCV10 against AOM are available. Upon review of the evidence, the EMA and the Pharmaceutical Benefits Advisory Committee of Australia both denied PCV10 an indication against NTHi AOM. The EMA only accepts the effectiveness of PCV10 and PCV13 against AOM caused by vaccine-serotype pneumococcus. In some countries in South America (Argentina, Chile, Ecuador, Mexico and Peru) and Asia (Philippines, Thailand, Pakistan and India), PCV10 has received an indication for prevention of AOM caused by NTHi. The Clinical Otitis Media and Pneumonia Study (COMPAS [104]), a randomized controlled trial of PCV10, is expected to report AOM end points in the near future, which should provide the required evidence of the potential effectiveness of PCV10 against both S. pneumoniae- and NTHi-causing AOM.

Initially, PCV7 was not considered cost effective because the indirect effects caused by it were still unknown, and early analyses only included the direct effects of the vaccine on vaccinated children [33]. As data showing disease reduction in unvaccinated cohorts began to emerge, the earlier models were revised to include herd effects. This led to PCV7 becoming cost effective and eventually to cost-savings in markets where PCV7 penetration was high over a longer duration and led to a substantial herd effect [19,34–38].

Pneumococcal herd protection was first reported in an analysis by the CDC in the USA, which reported a decline in pneumococcal disease in unvaccinated cohorts after licensure of PCV7 [39]. Because studies showing herd effects in populations other than those in the USA are limited, several of the models limited analyses of the impact of herd effects to scenario or sensitivity analyses. Herd effects are believed to be mediated by reducing NPC of vaccine serotypes in vaccinated children. PCV7 has consistently demonstrated a statistically significant reduction of NPC vaccine serotypes [40–44]. In three different settings, the impact of PCV10 or PCV11 has shown a trend towards but never consistently shown a statistically significant reduction in NPC in vaccinated individuals [32,45,46]. Therefore, evidence on herd effects with PCV10 is needed to validly insert these into cost–effectiveness models. Early models that assumed PCV13 would exhibit herd effects similar to those observed following PCV7 were based on the assumption that both PCV7 and PCV13 are conjugated vaccines with the same carrier protein and have comparable immunogenicity; recent data validate earlier assumptions by showing that PCV13 causes a statistically significant reduction in NPC [47–49].

Counteracting the benefits of herd protection is the phenomenon of serotype replacement, where the incidence of disease caused by nonvaccine serotypes increased after the vaccination of young children with PCV7. Serotype replacement occurs at different rates in different regions, and the emergence of various nonvaccine serotypes is also qualitatively different. Few models included a quantitative solution to serotype replacement because it is a highly uncertain occurrence. A challenge faced by all the analyses included in this review is the inability to predict the serotype distribution following the introduction of a higher-valent PCV. PCVs will alter the fundamental distribution of circulating serotypes across all age groups. Indirect effects for pneumonia and AOM in the presence of PCV7 are based on ecologic data; therefore, serotype replacement is inherently accounted for in these estimates. Notably, in a recent cost–effectiveness analysis, Rozenbaum et al assumed that herd protection due to PCV7 may be largely offset by serotype replacement, resulting in relatively small net indirect effects [19]. This was in line with data from a few countries, such as the UK and The Netherlands. However, recent data from the UK suggest that after accounting for changes in the surveillance system over time that resulted in overestimation of serotype replacement, the net effect is a significant reduction in disease [50]. Theoretically, enhanced herd protection provides better opportunities for serotype replacement. Therefore, reduced effects of vaccination on NPC, as suggested for PCV10, might result in relatively lower serotype replacement as a trade-off to lower herd protection effects.

Another common claim in cost–effectiveness analyses is the assumption that serotypes 6B and 19F, contained in both PCV7 and PCV10, are cross protective against serotypes 6A and 19A, respectively. However, data from several years of PCV7 use have demonstrated a difference between cross-reactivity and cross protection. Infants receiving PCV7 achieved a pneumococcal immunoglobulin G antibody protection ≥0.35 µg/ml after the primary series against both non-PCV7 vaccine serotypes 6A and 19A; therefore, it was hypothesized that PCV7 would show an impact on both serotypes 6A and 19A. However, after several years of use the incidence of serotype 6A declined whereas serotype 19A increased [51]. Although early results on immunogenicity indicated that PCV7 would have cross-reactivity on serotypes 6A and 19A, early models of PCV7 did not include any assumptions that the vaccine would prevent disease caused by nonvaccine-serotypes. The unexpected rise in serotype 19A shows that it is important to be conservative regarding assumptions of effect against disease caused by nonvaccine serotypes.

Some models assume cross protection for PCV10 against infections caused by serotype 6A, and cite CDC PCV7 data [28] as supporting evidence. PCV10 demonstrates functional activity against serotype 6A; however, this activity is a half to a third lower than that seen with PCV7 [52]. Therefore, the extent to which PCV10 will elicit cross protection for serotype 6A is unclear, and further evidence is needed to validly insert these into cost–effectiveness models.

Serotype 19A is the most important emerging serotype globally; its prevalence is currently increasing in countries that routinely use PCV7 [53–55]. Where PCV7 is routinely given, serotype 19A has frequently become a leading cause of post-PCV7 residual IPD cases. Although PCV10 induces higher opsonophagocytosis activity assay geometric mean titers to serotype 19A than PCV7 [52], the levels of this cross-reactivity elicited by both PCV10 and PCV7 are low compared with that seen with PCV13 after both the three-dose infant series and the toddler booster [56]. As serotype 19A disease has not decreased with immunization with PCV7, data suggest that it might also not decrease with PCV10 but further evidence is needed to validly insert these assumptions into cost–effectiveness models. Data from Quebec, Canada was inconclusive regarding this matter [57] due to the short duration over which PCV10, was used prior to the Canadian health authority’s decision to transition to PCV13. Surveillance data from The Netherlands, Finland, Kenya and Brazil, where PCV10 is used in NIPs, should be available soon and the uncertainty around PCV10’s impact on serotype 19A may be clarified.

The epidemiology data used in the different studies differ in timing. Some studies use the most recent epidemiology data available, whereas others utilize epidemiology that does not reflect changes in the underlying distribution of serotypes observed following PCV7 use. It is crucial that the models use epidemiology and serotype-derived effectiveness data from a consistent time period. It is incorrect to use epidemiology data from one time period and serotype prevalence from another. This may over- or under-state the impact that the vaccines are expected to have on the population, which in turn would skew the cost–effectiveness estimates. Prior to PCV7 use, the seven serotypes in PCV7 covered >70% of circulating serotypes in Europe and North America [58], and the marginal benefit of PCV10 and PCV13 was very small. With the near elimination of disease caused by the seven serotypes in PCV7 coupled with the emergence of non-PCV7 serotypes, the marginal benefit of PCV10 and PCV13 has increased after PCV7 introduction. Studies may utilize older incidence and serotype coverage data due to data availability limitations or other reasons; however, utilizing older data does not account for the current burden of disease, and biases model results toward lower-valent vaccines. This is due to replacement of serotypes, specifically 19A and 7F. By considering older serotype data, when the emergent serotypes were less prevalent, the benefit of broader serotype coverage is marginalized; that is, the relative serotype coverage of PCV7, PCV10 and PCV13 is similar. When considering current epidemiology, which takes into account serotype replacement with serotypes not contained in PCV7 and in some cases PCV10 (i.e., serotype 19A), the vaccines are more differentiated. Studies using serotype coverage and disease incidence measured prior to the widespread use of PCV7 are less relevant to current decision-making; therefore, local current epidemiological data on serotype prevalence should be considered.

NTHi-caused invasive disease is considered in a single analysis [18]. This study assumed PCV10 provided protective efficacy against NTHi-caused invasive disease equivalent to the efficacy against severe otitis media. Given the lack of evidence in this matter, results are speculative and, therefore, should not be used in a base case analysis. However, such sensitivity analyses do provide relevant information on potential future situations should such evidence accumulate.

Pneumococcal vaccination programs have had an impact on the overall presence of resistant pneumococcal strains. However, this analysis found that antimicrobial resistance is not taken into account in any of the cost–effectiveness model calculations. The authors hypothesize that this is due to a lack of data or lack of perceived impact on the model results. For example, Rozenbaum et al. state in their paper: “The impact of this inclusion is expected to be small given that penicillin resistance is less than 0.4% in The Netherlands” [19].

Results presented from the analyses are highly variable. Although results are presented comparing no vaccination and PCV7 as a comparator, at the moment the most relevant measure of cost–effectiveness for policy makers is the comparison of PCV10 with PCV13, as these are the two vaccines currently being considered for inclusion into NIPs. Of the 11 models that directly compared PCV10 with PCV13 [7,9,10,12,14,16–21], five found PCV13 cost-saving compared with PCV10 [10,12,14,20,21], three found PCV10 cost saving compared with PCV13 [7,17,18], two found PCV13 cost effective compared with PCV10 (US$28,147 per disability-adjusted life year [16] and €38,880 per quality-adjusted life year) [19]. Chuck et al. found that the results depended on the inclusion of the effectiveness of PCV10 against NTHi AOM [9]; that is, when NTHi effects were assumed, PCV10 was cost-saving compared with PCV13. However, when NTHi effects were not included, PCV13 was found to be cost-saving compared with PCV10. The three models that favored PCV10 to PCV13 all included an NTHi impact for PCV10.

Because efficacy data for the newer vaccines are just starting to emerge, the assumptions used to determine vaccine effectiveness should be competent and reliable, and based on the best available data. PCV10 and PCV13 are based on different conjugation chemistries, have different clinical trial evidence, and possess observational data in different contexts. Therefore, it should not per se be assumed that these vaccines will produce identical results when used in NIPs. As such, any assumptions of a PCV’s impact beyond pneumococcal serotypes contained in the respective vaccine cannot be credibly supported in a base case until sufficient evidence is generated. Data on the impact of the newer pneumococcal vaccines on the incidence of IPD, pneumonia and AOM after their widespread use will improve cost–effectiveness estimates.

Five-year view

Currently, all analyses discussed in this review utilize vaccine efficacy estimates based on assumptions. With the growing body of knowledge on the real-world effectiveness, the uncertainty on the different input parameters will diminish. These data should be incorporated into models in order to provide a more precise estimate of vaccine impact and cost–effectiveness.

Future cost–effectiveness analyses may employ more complicated modeling techniques, such as transmission dynamics, in order to incorporate dynamic issues related to pneumococcal vaccination such as NPC reduction and serotype replacement.

Table 1. Studies included in the review.

Study (year)	Country	Funding	Type of model	Vaccines/schedule	Diseases	Perspective	Ref.
De Wals et al. (2009)	UK	GSK^†‡	Population-based, 1-year decision tree	PCV7 versus PCV10/3 + 1	IPD, AOM, CAP	Not applicable	[8]
Chuck et al. (2010)	Alberta, Canada	Wyeth Canada^†	Population-based, 1-year decision tree	PCV7, PCV10, PCV13/3 + 1	IPD: PNE-H, MEN, BAC-H, BAC-O, non-IPD, OM, PNE-O, NTHi remainder, and not infected	HC	[9]
Newell et al. (2011)	Australia	GSK^†	Population-based, 100-year Markov model	PCV7, PCV10, PCV13/3 + 0 or 3 + 1	IPD (MEN or BAC), OM (H or O), PNE-H and PNE-O	HC	[10]
Morano et al. (2011)	Spain	GSK^§	Population-based, 1-year decision tree	PCV7, PCV10, PCV13/3 + 1	MEN, PNE and AOM	HC	[7]
Rubin et al. (2010)	USA	Pfizer/Pfizer authors	Population-based, 10-year Markov model	PCV7, PCV13/3 + 1	IPD (MEN, BAC), PNE and AOM	S	[11]
Strutton et al. (2012)	Germany, Greece and The Netherlands	Pfizer/Pfizer authors	Population-based, 1-year decision tree	No vaccine, PCV7, PCV10, PCV13/3 + 1	IPD, PNE and AOM	HC	[12]
Talbird et al. (2010)	Germany, Mexico, Canada and Norway	GSK/GSK authors	Population-based, 1-year decision tree	PCV7, PCV10/3 + 1 in Canada and Germany; 2+1 in Mexico and Norway	AOM, MEN, PNE and BAC	HC	[13]
Muciño-Ortega et al. (2011)	Mexico	Pfizer/Pfizer authors	Population-based, 1-year decision tree	PCV7, PCV13/ both 2+1; PCV10/3 + 1	BAC, bacterial MEN, PNE-H, PNE-O and OM	HC	[14]
Sartori et al. (2012)	Brazil	Country^¶	TRIVAC# deterministic state-transition model, cohorts from 0–5 years	PCV7 (high risk only), PCV10/3 + 1	IPD, PNE and AOM	HC and S	[15]
Ureña et al. (2011)	Argentina	NR	TRIVAC# deterministic state-transition model, cohorts from 0–5 years	No vaccine, PCV10, PCV13/both 3 + 1	AOM, PNE, pneumococcal BAC/SEP and pneumococcal MEN	HC	[16]
Robberstad et al. (2011)	Norway	GSK^†/GSK authors	Birth cohort, 95-year Markov model	No vaccine, PCV7, PCV10, PCV13/2 + 1	MEN, BAC, PNE and AOM	S	[17]
Knerer et al. (2012)	Canada and the UK	GSK/GSK authors	Birth cohort, 95-year Markov model	PCV10, PCV13/3 + 1	MEN, BAC, PNE and AOM	HC	[18]
Rozenbaum et al. (2010)	The Netherlands	Wyeth^†, GSK^†	Birth cohort, 5-year Markov model	PCV7, PCV10, PCV13/2 + 1	MEN, PNE (noninvasive and invasive), BAC (with/without focus) and AOM	HC	[19]
Kim et al. (2010)	The Gambia	Private^††	Birth cohort, 5-year Markov model	PCV7, PCV9, PCV10, PCV13/3 + 0	PNE, pneumococcal MEN and pneumococcal SEP	S	[20]
Tyo et al. (2011)	Singapore	Country^‡‡	0–5-year-olds, 5-year Markov model	PCV7, PCV10, PCV13/2 + 1	IPD (MEN or BAC), AOM and CAP	H	[21]
Díez-Domingo et al. (2011)	Valencia, Spain	Pfizer^†	Population-based, 10-year decision tree	No vaccine, PCV13/2 + 1	IPD (MEN, SEP, bacteremic PNE) and AOM	HC	[22]

^†Unrestricted grant.

^‡GSK funded manuscript development.

^§The present study was funded by GSK. The analysis was performed by an independent consulting firm contracted by GSK for this purpose.

^¶Ministry of Health of Brazil and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq).

^#Vaccine cost–effectiveness model from Pan American Health Organization’s ProVac Initiative.

^††Bill and Melinda Gates Foundation.

^‡‡Ministry of Health, Singapore; Heller School at Brandeis.

AOM: Acute otitis media; BAC: Bacteremia; CAP: Community-acquired pneumonia; GSK: GlaxoSmithKline; H: Hospitalized; HC: Healthcare system; IPD: Invasive pneumococcal disease; MEN: Meningitis; NR: Not reported; NTHi: Nontypable Haemophilus influenzae; O: Outpatient; OM: Otitis media; PCV: Pneumococcal conjugate vaccine; PCV7: Seven-valent PCV; PCV9: Nine-valent PCV; PCV10: Ten-valent PCV; PCV13: 13-valent PCV; PNE: Pneumonia; S: Societal; SEP: Septicemia.

Table 2. Variables and assumptions used in each model.

Country	Epidemiologic year^†	AOM effectiveness	NTHi considered	Herd effects considered	Cross protection to nonvaccine serotypes	Cost–effectiveness results	Ref.
UK	2001–2006 (UK)	PCV7: 6% FinOM [23]	Yes	Yes, IPD only	Yes, 6A and 19A	PCV10 3+1 more effective than PCV7 in all scenarios^§	[8]
	1999–2002 (Scotland)	PCV10: 20.6%^‡ POET [5]
Alberta, Canada	Pre-PCV7; 2000–2002 Post-PCV7; 2003–2008	Not specified; assumed to be NCKP [3]	Yes PCV10: 5% additional reduction above ST extrapolated values [5]	Yes; both with and without herd impacts evaluated	No	PCV10 CS versus PCV13 when NTHi included. PCV13 CS versus PCV10 when NTHi not included	[9]
Australia	2009	PCV7: 6.6% [24]	Yes	Base case: no; test in SA	Base case: no; tested in SA for 6A and 19A	PCV13 versus PCV7: A$55,300/QALY. PCV10 versus PCV7: A$55,200/QALY	[10]
		PCV10: 17.3% ST extrapolated + POET [5]
		PCV13: 8.1% ST extrapolated
Spain	2008	PCV10: 22.9% POET [5]	Yes	No	Base case: 6A	PCV10 CS versus PCV13	[7]
		PCV13: 11.2% ST extrapolated
Mexico	2000–2005 [25]	PCV7: mild, 7.0% NCKP; moderate to severe, 15% [24]	No	Base case: PCV7 and PCV13 included, PCV10 excluded; Tested in SA	No	PCV7, PCV10, and PCV13 CS versus no vacc. PCV13 CS versus PCV7 and PCV10	[14]
		PCV10: mild, 8%; moderate to severe, 17–21% ST extrapolated + NTHi (country specific)
		PCV13: mild, 9%; moderate to severe, 14–19% ST extrapolated (country specific)
Brazil	2000–2005 (Mexico)	PCV7: NR	No	No	No	PCV10 versus PCV7 high risk: R$22,066/DALY	[15]
		PCV10: 4.6%^‡ POET [5] and FinOM [23]
Norway	2004–2005	PCV7: 9.2% (6.5%^‡ estimated) FinOM [23] and NCKP [3]	Yes	Yes, IPD only	Yes, 6A and 19A	PCV10 CS versus PCV13	[17]
		PCV10: 23.8% POET [5] + NTHi
		PCV13: 11.3% POET [5]
Canada and UK	‘Pre-PCV7’, appendix data not available	PCV10: 22.9%^‡ POET [5] + NTHi + 6A cross-protection	Yes, also considered impact of NTHi on invasive disease in base case	Yes, IPD only	Yes, 6A for AOM, 6A and 19A for IPD	PCV10 CS versus PCV13	[18]
		PCV13: 14.9%^‡ POET [5], SP only
Netherlands	2004–2006	PCV7: 7% [24]	No	Yes, IPD only in children	No	Compared with no vacc.: PCV7: €113,891/QALY; PCV10: €52,947/QALY; PCV13: €50,042/QALY	[19]
		PCV10: 8% ST extrapolated
		PCV13: 9% ST extrapolated
Gambia	2009	AOM not included in analysis		Base case: no Tested in SA	No	Compared with no vacc.: PCV7: $910/DALY PCV9: $670/DALY; PCV10: $670/DALY PCV13: $570/DALY^¶	[20]
USA	2007	PCV7: 0% assumed that all benefit of PCV7 had been realized	PCV10 not included in analysis	Yes	PCV10 not included in analysis	PCV13 CS versus PCV7	[11]
		PCV13: mild, 6.7% (age specific, ST extrapolated from NCKP [3]; moderate to severe, 10.3% (age specific, ST extrapolated from NCKP [3]
Argentina	2006–2010	PCV10: 33.6% POET [5]	Yes, also considered impact of NTHi on pneumonia in SA	Base case: no Tested in SA	No	Compared with no vacc.: PCV10: $8973/DALY PCV13: $10,948/DALY; PCV13 versus PCV10: $28,147/DALY	[16]
		PCV13: 7% NCKP [3]
Singapore	2009	PCV7: 5.9% NCKP [3]	Base case: No; Tested in SA	Yes	No	Compared with no vacc.: PCV7: $43,275/QALY; PCV10: $45,100/QALY; PCV13: $37,644/QALY	[21]
		PCV10: 6.0% ST extrapolated
		PCV13: 6.9% ST extrapolated
Germany, Norway, Canada, Mexico	Various years, pre-PCV7	PCV7: 0–10.1%^‡ FinOM [23]	Yes	Yes, IPD only	Yes, 6A and 19A	PCV10 CS versus PCV7	[13]
		PCV10: 19.4–28.1%^‡ POET [5] + NTHi
Germany, Greece, Netherlands	Germany, 2006	PCV7: mild, 7.0% NCKP; moderate to severe, 15% [24]	Yes PCV10: 4% additional reduction above ST extrapolated values [5]	PCV10: no in base case PCV7, PCV13: yes in base case. Both tested in sensitivity analysis	No	PCV13 CS versus no vacc. in Germany and Greece; €38/QALY in The Netherlands. PCV13 CS versus PCV10	[12]
	Greece, 2008	PCV10: mild, 8%; moderate to severe, 17–21% ST extrapolated + NTHi (country specific)
	The Netherlands, 2006	PCV13: mild, 9%; moderate to severe, 14–19% ST extrapolated (country specific)
Valencia, Spain	2007–2008 (AVE database)	PCV13: 9% [26]	No	Yes; Tested in SA	No	PCV13 versus PCV7: €10,407/QALY	[22]

^†Year of the epidemiological data used to estimate serotype coverage.

^‡Calculated by authors based on available data in the publication.

^§Costs not calculated, effectiveness defined as cases of disease avoided.

^¶Price parity assumed for all vaccines.

AOM: Acute otitis media; AVE: Análisis de Vigilancia Epidemiológica; CS: Cost saving; DALY: Disability-adjusted life year; FinOM: Finnish Otitis Media study; IPD: Invasive pneumococcal disease; NCKP: Northern California Kaiser Permanente; no vacc.: No vaccination comparator; NR, Not reported; NTHi: Nontypable Haemophilus influenzae; PCV: Pneumococcal conjugate vaccine; PCV7: Seven-valent PCV; PCV9: Nine-valent PCV; PCV10: Ten-valent PCV; PCV13: 13-valent PCV; POET: Pneumococcal Otitis Media Efficacy Trial; QALY: Quality adjusted life years; SA: Sensitivity analyses; SP, Streptococcus pneumoniae; ST: Serotype.

Key issues

• • Two higher-valent pneumococcal conjugate vaccines (PCVs), a 13-valent PCV (PCV13) and ten-valent PCV (PCV10) have been evaluated in numerous cost–effectiveness analyses.

• • Decision-makers in many countries are deciding whether to select PCV10 or PCV13 as part of their national immunization program (NIP) and several cost–effectiveness analyses have been conducted to evaluate the potential value of PCV10 and/or PCV13.

• • Vaccine cost–effectiveness analyses estimate cases of disease (invasive pneumococcal disease [IPD], pneumococcal pneumonia and acute otitis media [AOM]) that are averted by immunization based on assumptions of vaccine effectiveness. Because the effectiveness of newer vaccines in a specific country is unknown, due to the lack of efficacy studies, variable local distribution of serotypes, levels of vaccination uptake, serotype replacement and other reasons, researchers must make choices regarding assumptions about the local effectiveness of vaccines.

• • All models used a consistent approach to extrapolating the effectiveness of PCV10 and PCV13 against IPD and all-cause pneumonia by basing the effectiveness of the newer vaccines on the serotype distribution in the respective country for which the analysis was performed.

• • Assumptions of vaccine efficacy against all-cause AOM varied most-widely across analyses, and AOM is identified as a key economic driver in PCV cost–effectiveness analysis. Inclusion of indirect effects and cross protection were the second and third most variant parameters.

• • Assumptions about PCV10 effectiveness against AOM caused by nontypable Haemophilus influenzae are based on an efficacy study carried out in a patient population with more severe AOM using a different vaccine, 11-valent PCV, in which all 11 serotypes were conjugated to H. influenzae protein D. Data for the reformulated PCV10, with less protein D, is currently not available, which leads to large differences in assumptions on effectiveness of PCV10 against AOM.

• • Assumptions used in cost–effectiveness models should be conservative in the absence of data, until evidence in the public domain is sufficient to make claims about how a vaccine will perform when used in a NIP.

• • Data on the impact of the newer pneumococcal vaccines on the incidence of IPD, pneumonia and AOM after sustained use in NIPs would improve cost–effectiveness estimates.

Financial & competing interests disclosure

This study was sponsored by Pfizer Inc. RA Farkouh, RM Klok, CS Roberts and DR Strutton are employees of Pfizer Inc. MJ Postma has received funding from Pfizer Inc. and other pharmaceutical companies for cost–effectiveness model development. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Medical writing support was provided by Nancy Price at Excerpta Medica and was funded by Pfizer Inc.