A Critical Literature Review of Health Economic Evaluations in Pertussis Booster Vaccination

Aurelie Millier; Samuel Aballea; Lieven Annemans; Mondher Toumi; Sibilia Quilici


Expert Rev Pharmacoeconomics Outcomes Res. 2012;12(1):71-94. 

In This Article

Critical Analysis


There are substantial differences among models used by authors, both in methodology and in structure. Four major types of models are available: decision trees, Markov models, discrete events simulation (DES) and dynamic compartmental models, as described in Figure 1.

Figure 1.

Four types of models used in reviewed economic evaluations. (A) Generic decision tree used for calculation of the cost–effectiveness of potential pertussis immunization strategies. (B) Structure of the Markov model. (C) Schematic representation of the possible pathways within the pertussis discrete event simulation model.
(D) Diagram of the immunological and infectious states and transitions between states in the age-specific pertussis model.
V: Vaccine status; VC: Vaccination coverage; WN: Waning of natural immunity; WV: Vaccine waning.
(A) Adapted with permission from [21]; (B) Adapted with permission from [22]; (C) Adapted with permission from [13]; (D) Adapted with permission from [16].

Model Structures

Out of the 13 articles presenting results from a model, three did not describe the model structure[14,23,24] and there were six Markov models,[15,17,18,20,22,25] one decision tree,[21] one dynamic DES model,[13] one dynamic compartmental model,[24] and one only indirectly used results from a dynamic model.[19]

Westra et al. used a decision tree with probabilities for individuals to be infected or not, for the infection to be symptomatic or not, and for the case to be reported or not.[21] As time is not explicitly defined, duration of infection is not modeled, and individuals could not be infected more than once. Also, the authors designed separate models for adults and infants; therefore, indirect protection was not included in the model.

Lee et al. used a Markov model (static cohort model) with 1-year cycles.[15,20,25] In such a model, patients may go from one health state to another according to transition probabilities that depend on the current state. Health outcomes and costs associated with these states were simulated for a cohort of people over their lifetime. Advantages of the Markov model versus decision tree is that patients are followed-up over time, taking into account variations of immunity level. Adolescents and adults with pertussis were classified as having mild or severe cough illness, or pneumonia. Infants who had pertussis developed either respiratory or neurologic complications or died as a result of the disease.

Scuffham et al. also used a Markov model, arguing that the differential timing of events could be modeled explicitly.[22] They used a time horizon of only 6 months, with 26 1-week cycles, and seven health states including susceptible, infected, immune, vaccinated (× three doses) and death.

De Vries et al. designed a stochastic DES model that represents a chronological sequence of events, where individuals are modeled rather than cohorts.[13] Stochastic models are generally recommended when there is concern about first-order uncertainty; for example, when a small number of infected cases in a population could lead to a large outbreak by chance. This is a situation encountered with pertussis. The model distinguished between three types of infections (primary, recurrent and asymptomatic infections), considered two scenarios of duration of immunity acquired by natural infection (8 and 15 years) and distinguished vaccine-induced immunity for infection (2 years) and for disease (an additional 6 and 13 years, respectively). The main disadvantage of this stochastic model is the running time, as authors could only run 20 iterations per scenario.

Coudeville et al. developed a transmission dynamic compartmental model.[16] This takes into account some specific features of pertussis transmission such as the progressive switch from naturally acquired immunity to vaccine-acquired immunity, and can capture herd immunity by modeling transmission mechanisms via computer simulations. Individuals are compartmentalized into three main pertussis states: fully susceptible, fully or partially immune, and infectious. They may be transferred from one to another according to a 'who acquires infection from whom' matrix.

Time Horizon

The time horizon should be long enough for the analysis to capture the entire difference in costs and outcomes between alternative strategies. Out of the 13 articles, the time horizon was 6 months for one study,[22] 8 years for one study,[21] 10 years for four studies,[17,18,23,24] 25 years for one study,[13] four presented a lifetime analysis[14,15,20,25] and two dynamic models presented results at steady state, which corresponds to a period over which the full impact of the vaccination strategy has been reached.[19,21] The model used by Coudeville et al. reached steady state at approximately 80 years.[16] Edmunds et al. did not explicitly state the time horizon.[19] Scuffham et al. used a time horizon of 6 months, assuming it was adequate to capture the main effects of infection in infants.[22] The authors considered it to be a conservative approach, as potential additional benefits would not be taken into account. Only Iskedjian et al. conducted a sensitivity analysis on the time horizon and found that it can have an impact: decreasing the 10-year horizon to a 5-year horizon increased incremental costs per child per year by approximately twofold.[17]

Epidemiological Inputs

Incidence rates, under-reporting correction factors, other epidemiological inputs and their sources are presented in Table 3 .


When reported, incidence rates differed largely between studies (some did not present explicit incidence rates and some just mentioned that they were age specific). For several reasons pertussis disease is generally under-reported: most of the time only culture-positive cases or cases with highly typical symptoms are reported; and it is typically considered to be a childhood disease and passes unnoticed in other populations. Most studies used a correcting factor to estimate incidence, from 2.5 to 660 depending on the country. Data on hospitalizations and deaths were particularly influenced by the characteristics of the surveillance system in place, reflected by the large variation in the estimates observed. After correction, incidence rates of pertussis varied from 22 to 58.5 per 100,000 for infants, from 95 to 511 per 100,000 for adolescents and from 11 to 507 per 100,000 for adults. This parameter can strongly influence the results, and may inverse the sense of conclusions. Also, regarding the nonreported cases (although still symptomatic), some could argue that they are in general milder and hence cheaper; however, several studies showed that the burden associated with these cases was significant. Input values for healthcare resource use and mortality also varied considerably. In addition, the incidence has increased over recent years in several countries. For example, in 2007 Lee et al. used revised incidence in the second study performed in the USA, contributing to more favorable results towards vaccination.[25]

Asymptomatic Cases

These are unrecognized infections with mild or subclinical symptoms, which are not associated with significant costs or effects on quality of life. There is no need to account for such cases in static models, whereas their consideration is crucial in dynamic models as they may participate in pertussis infection transmission. The majority of infections in adults and adolescents are actually asymptomatic, ranging from 78 to 95%, according to recent studies.[26,27] It is also suggested that 16% of infections in infants are the result of transmission via asymptomatic cases.[26] Out of the three dynamic models, only two clearly considered asymptomatic cases.[13,16] For both analyses the duration of infectiousness was estimated to be 1 week with no productivity losses.

Herd Immunity

Herd immunity refers to the indirect protection conferred by the vaccine: vaccination of some age groups, in particular adolescents, may considerably reduce the transmission of the disease in the nonvaccinated population, which will impact the cost–effectiveness of the vaccination strategy under consideration. Table 4 presents information relative to herd immunity considered in the included economic analyses.

Despite the different methodologies adopted, most authors tried to incorporate herd immunity in their model (or clearly stated that they did not incorporate its benefits as a conservative approach).

Sensitivity analyses on the percentage of herd immunity were often performed to examine the impact on model outcomes. Lee et al. found that herd immunity had a greater impact on cost–effectiveness when disease incidence was low, suggesting that a certain threshold of cases prevented is needed for the intervention to appear cost effective.[25] It seems then more appropriate to use a dynamic approach, especially to evaluate booster vaccination of adolescents. Indeed, herd immunity effects depend on the age groups considered, and potential positive and negative indirect effects can be associated with it. For instance, the age shift observed could be perceived as a positive effect since disease is less likely to be severe in adults. Nevertheless, it increases the risk of transmission from parents to infants. Unfortunately, no publication reported the size of indirect effects relative to direct effects.

Vaccine-related Data

Efficacy Several estimates of efficacy were used in the models: Edmunds et al. used an initial rate of 95%, with no information about waning values.[19] Lee et al. assumed that immunity waned each year for 15 years, from 100 to 0%, based on the basis of expert panel input and published data.[15,20,25] Depending on the authors assumptions, protective efficacy varied from 85%[17,18] to 89%.[13] Only Coudeville et al.[16] and Westra et al.[21] used dose-dependent efficacy data. Westra et al. assumed that the maximum level of efficacy was reached at the third dose, whereas Coudeville et al. considered an increase of the vaccine efficacy up to the fourth dose.[16] The impact of inputs related to protective efficacy was investigated with low and high vaccine efficacy scenarios, which were found to substantially change the results. For example, Coudeville et al. found that in the low vaccine efficacy scenario implementing routine decennial vaccination appeared to be a more cost-effective solution than the combination of a cocoon vaccination with a single booster dose at 40 years of age.[16] The situation was reversed for a high vaccine efficacy scenario: implementing only the adolescent or adolescent and cocoon seems to be sufficient and the cost–effectiveness ratio is higher than US$1,000,000 for the booster dose at 40 years of age. The impact of vaccine efficacy on ICERs was also established in Caro et al., where a duration of protection of 3 instead of 10 years considerably deteriorated the cost–effectiveness.[14]

Adverse Events In most of the studies reviewed, side effects of vaccination were not considered. Five studies included side effects as part of a conservative approach: Lee et al. included local reaction, systemic reaction and anaphylaxis,[15,20,25] Purdy et al. included local or systemic reactions[23] and Caro et al. considered adverse reactions of sufficient severity to warrant a physician visit.[14] In all studies, the impact was negligible, except for adult vaccination in Lee et al., which resulted in fewer QALYs compared with no vaccination, mostly due to adverse events.[15]


Perspectives Eight studies reported costs from both a TPP and societal perspective, one considered only TPP perspective and four only considered societal perspective. All studies adopted the perspective recommended in local health economic guidelines.

Table 5 lists all costs included in the economic analyses.

Vaccination Cost Vaccination cost is often presented as an incremental cost of Tdap booster versus Td. Values largely differ according to the geographical area: from an extra cost of GBP£5 in the UK to a vaccine price above €18 in The Netherlands, where the Td vaccine was not available. Administration costs were sometimes added to the vaccine cost when authors thought an additional medical visit would be needed. Westra et al found a moderate impact of administration cost assumption on ICERs in The Netherlands; conclusions were not affected.[21]

Disease-related Costs All studies clearly described what types of costs were included in terms of direct medical, direct nonmedical and indirect costs.

Direct Costs Direct medical costs included clinical resources, such as hospitalizations, emergency room and physician visits, laboratory tests for diagnosis and medications. Direct nonmedical costs included additional childcare provision and travel expenses incurred for consultations. Both medical and nonmedical costs were mostly age specific. Direct costs are considered in the TPP perspective.

Indirect Costs Indirect costs included those associated with time diverted from normal activities and reduced work productivity. These costs were added to the direct costs when a societal perspective was considered.

Ratio of Direct/Indirect Costs Ratio of direct/indirect costs varied a lot between studies, countries and age groups: for Purdy et al., the indirect costs due to productivity losses comprise by far the largest part (~80–90%) of the costs associated with pertussis diseases,[23] whereas it accounts for less than half of total costs for Iskedjian et al.[17,18] and Coudeville et al.[16]

Health-state Utilities

The QALYs combine survival and health-related quality of life (i.e., utility values) into a single health outcome. This is the preferred health outcome measure for several health technology assessments using economic evaluations to support decision making. There is only one available data source for utility values for pertussis disease.[28] In this study the authors conducted a survey using time trade-off (TTO) and contingent valuation methods to determine the utility of either short- or long-term health scenarios for adolescents and adult patients, referring to vaccination health states (local reaction or systemic reaction) and disease health states (mild cough, severe cough or pneumonia). For infants, they considered respiratory and neurologic complications.

Due to the specificity of the models used in economic evaluations, the authors had to make several assumptions. For example, Lee et al. assigned for moderate cases the value reported for mild cough, and assumed a higher utility for mild cases.[15,20,25] They also assumed a utility for anaphylaxis and for outpatient respiratory complications. De Vries et al. considered the reported value for mild cough as input value for unreported cases, and moderate cough as input for reported cases.[13]

Utility input values had limited impact on results in Lee et al.[15,20,25] and Westra et al.,[21] but significantly affected results in De Vries et al.[13]

Scuffham et al. used DALY, which is an index combining morbidity and mortality according to weights for each health state multiplied by the duration in that state.[22] DALYs are useful for broad cross-country comparisons, more than for comparing specific diseases and priority setting within a country. The authors used the DALY weights for pertussis infection and complications reported from the global burden of disease study,[22,29–31] computed with the TTO method. Changes in the DALY weights for illness and hospitalization had insignificant impact on the ICERs.

Discount Rate

Outcomes and costs were discounted at a rate of 3% in all studies, except for The Netherlands, where guidelines suggest using 4 and 1.5%. Costs were not discounted in Scuffham et al. since all costs incur within a short time horizon (less than a year).[22] The impact of using discount rates was investigated by sensitivity analysis, showing generally a limited impact except for Edmunds et al.[19] and Caro et al.[14] It is notable that for these two studies the outcome expressed as life-year gained was very sensitive to discount rate.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.