1 Introduction Meningococcal disease is a leading infectious cause of death in young children in the UK  and remains an important cause of morbidity and mortality worldwide, despite improvements in critical care and the availability of vaccines against some capsular groups. Globally five capsular groups cause most disease (A, B, C, W, Y though X is increasing) and B and C are dominant outside Africa and Asia . The key to reducing incidence is prevention through vaccination, because early signs of the disease can be non-distinct, the infection can progress rapidly, and can be fatal in 5–10% of cases even if treatment is initiated early . Effective vaccines are available against capsular groups A, C, W and Y. The meningococcal serogroup C conjugate (MCC) vaccine was first introduced in the UK in 1999  and subsequently by several other European countries, Australia and Canada . MCC vaccination achieved high uptake rates, and has led to a considerable reduction in group C disease  due both to high vaccine effectiveness and protection against carriage, interrupting transmission and generating herd immunity . Until recently there was no broadly effective vaccine against capsular group B (MenB) the most common cause of meningococcal disease in the UK and Europe  (MenB disease accounted for 89% of cases in England and Wales in 2009/10 ). Progress towards a MenB vaccine has been hindered because the serogroup B capsule shares homologous structures with human neural tissue, resulting in the polysaccharide being poorly immunogenic in people and concerns about a MenB capsular-based vaccine inducing auto-immunity . New vaccines with the capacity to protect against MenB, based on protein antigens, are in advanced stages of development [10,11] and one, Bexsero, was granted an EU license in January 2013. Policy makers are now faced with decisions about if, and how, to introduce the vaccine. To help inform policy decisions we developed mathematical and economic models to predict the potential impact of introducing a new vaccine in England, with the capacity to protect against MenB disease (henceforth referred to as a ‘MenB’ vaccine). 2 Methods 2.1 Model structures It is unknown whether the new meningococcal vaccines will reduce carriage. Consequently, we developed two models (using Berkley Madonna software ) to assess the potential impact of these vaccines: a cohort model that assumes the vaccine prevents disease only, and a transmission dynamic model that also allows the vaccine to prevent carriage [13,14]. 2.1.1 Details common to both models The model populations are stratified into 100 single year of age classes. Incidence rates include all capsular groups of meningococcal disease because the new vaccines are not group specific. Following disease, individuals may survive with or without sequelae, or die. Survivors with sequelae are assumed to have a reduced quality of life and fatal cases lose the average life expectancy for the age at which they die. Individuals may die due to causes other than meningococcal disease; published mortality rates were adjusted to remove deaths due to meningococcal disease as these are explicitly modelled. Vaccinated individuals have a reduced risk of disease. Immunity from vaccination wanes over time, and individuals then have the same risks of infection as unvaccinated individuals. For each vaccination scenario the model results were compared to the situation without vaccination. Models were run for 100 years (time horizon) to capture the full benefits of vaccination and effects of invasive disease over the lifetimes of individuals. 2.1.2 Cohort model specific details The cohort model was constructed using a Markov model, with monthly time steps. Individuals are born into a susceptible non-vaccinated state (Fig. 1). Meningococcal disease cases arise by multiplying the age-specific probability of disease (in a given interval) by the population. We assumed individuals only have disease once and are removed from the susceptible pool (instances of repeat invasive disease are rare and are associated with individuals with immune deficiencies and anatomical defects ). Years of life are weighted by the age-specific quality of life. The cohort sizes were based upon population figures for 2008. 2.1.3 Transmission dynamic model specific details Individuals can have multiple episodes of asymptomatic carriage of meningococci in their lifetimes [16,17], therefore we used a Susceptible-Infected-Susceptible (SIS) model, with a daily time step, to represent the transmission dynamics of carriage in the population (Fig. 1). Individuals are born susceptible. They may then become carriers of a meningococcal strain (vaccine preventable or non-vaccine preventable), from which they recover and return to the susceptible state. We did not consider co-infection in the model because current evidence suggest carriage of multiple meningococcal strains is rare [18,19]. Cases of invasive disease are not explicitly included, but are generated from the number of new carriers arising over time (see Supplementary Material) using an age-specific case: carrier ratio. This ratio captures changes in disease risk given carriage acquisition across ages, which could be due to a number of factors including maturation of the immune system, physical changes in the pharynx, exposure to other pathogens and immunity following meningococcal carriage. Vaccinated individuals with vaccine induced immunity can have a reduced risk of becoming a carrier in addition to a reduced risk of disease. 2.2 Model parameters Data sources used to estimate the parameters in the models, are summarised below and in Table 1 with further details provided in the Supplementary Material. We used carriage prevalence estimates from a recent systematic review , with contact patterns estimated using a simple preferential mixing structure and recently published survey data on self-reported contacts . Disease incidence naturally fluctuates over time; incidence peaked in the late 1990s and has declined since then. We therefore based disease incidence and case fatality upon hospital admissions from 2004/05–2005/06 to represent current low incidence. Data from 1997/98–2005/06 (adjusting for the decline in incidence due to MCC), which includes peak incidence years, were used to generate a ‘higher’ incidence comparator. We assumed all meningococcal disease cases were hospitalised and estimated those requiring augmented care from hospital admissions (1998/99–2005/06). We included published costs for time in hospital including augmented care , and all survivors of disease were assumed to have a hearing test and a follow-up review in line with recent NICE guidelines . The proportion of survivors with minor and major sequelae following disease was estimated from a recent systematic review of sequelae following bacterial meningitis . Those with sequelae were assumed to have a reduced quality of life (0.2 utility reduction [25–27]) compared to susceptible individuals, and survivors of disease without sequelae . Long term costs of supporting those with mild and severe sequelae were estimated at £500 and £10,000 per year per individual respectively. For public health management we included costs of chemoprophylaxis (rifampicin for 3 adults and 2 children ) and staff time associated with contact tracing. Costs of outbreak control were not included. Several vaccination strategies were considered (Table 2). Vaccination uptake for routine vaccination was assumed to equate to MCC in infants, and for catch-up cohorts, match the MCC catch-up programme . Vaccine administration costs [31–34] were included separately from the cost of the vaccine itself, and were greater if given outside of current schedules. The full characteristics of the new meningococcal protein vaccines are not yet known; assumptions regarding vaccine effectiveness and duration of protection were based on data from trials, other meningococcal vaccines, such as the MCC or Outer Membrane Vesicle vaccines, and expert opinion. Data from trials of Bexsero have indicated, however, that the vaccine is immunogenic in infants , and adolescents , that responses are evident after two doses of the vaccine in infancy  and that it is possible to boost an individual's response . Early genotypic estimates of strain coverage suggested 100% strain coverage was possible  however recent phenotypic approaches suggest strain coverage in England may be 73% (95% CI 57–87%), though these results are based on a method which may underestimate coverage . In the base case model the vaccine was assumed to protect against all meningococcal strains. We included costs, but not quality of life losses, for adverse vaccine events. We assumed the vaccine cost £40 per vaccine dose in the base case, but varied this widely in the sensitivity analysis. 2.3 Scenario and sensitivity analysis The cohort model was probabilistic, with distributions around the parameters reflecting uncertainty (Table 1). Where probabilistic analysis was not possible or appropriate (e.g. vaccine price will be fixed, but at a level currently unknown) we ran scenario analyses. Cost-effectiveness ratios from probabilistic results were calculated using the ratio of the means . Cost-effectiveness acceptability curves were generated using a net benefit approach. 2.4 Costs and discounting Three health outcomes were considered: cases averted; deaths averted; and quality adjusted life years (QALYs) gained. The cost-effectiveness (utility) analysis was undertaken from the perspective of the NHS and personal and social services according to NICE guidance ; the primary outcome was cost per QALY gained. Costs were measured in pounds sterling at 2008 prices, with previous years inflated to 2008 levels. All costs and benefits were assumed to occur at the start of the year with future costs and benefits discounted according to HM Treasury recommendations (discount rate of 3.5% for the first 30 years, 3.0% in years 31–75 and 2.5% in years 76–99) . 3 Results 3.1 Impact of vaccination assuming direct effects only The model estimates 1799 cases of meningococcal disease (all capsular groups) could be expected in England over the lifetime of the 2008 birth cohort, resulting in 18,215 hospital bed days and 91 deaths. An estimated 484 (27%) cases (3756 bed days and 11 deaths) could be prevented by introducing routine early infant vaccination (strategy A Tables 2 and 3). Protection begins at 4 months, following two vaccine doses, and most cases are averted between the ages of one and two years (Supplementary Fig. 1). Due to waning vaccine immunity, averted cases decline rapidly after this. Infants immunised later in childhood experience longer-lasting immunity, therefore, routine late infant vaccination could result in more averted cases despite protection starting later in life (strategy B Tables 2 and 3). Strategies including 12 month boosters result in more cases averted, compared to those without, because the duration of protection is assumed to be longer following the booster. Catch-up strategies result in further case reductions, but the effects are limited in this model assuming direct protection only. Assuming the vaccine costs £40 per dose, early infant vaccination (strategy A) would cost £103·7 million for a single birth cohort and provide NHS savings of £30·1 million, resulting in a cost per QALY gained of £162,800. The cost per QALY of a late infant schedule (strategy B) remains high (£164,100), despite the greater number of cases avoided, due to the additional cost of calling children for vaccination at 6 months. The vaccine would need to cost around £9 per dose for these routine infant strategies to be considered cost-effective (<£30,000 per QALY). Catch-up campaigns are least cost-effective because these involve immunising a large number of people in ages where disease incidence is relatively low. 3.2 Impact of vaccination allowing for herd immunity Routine early infant vaccination with a vaccine that offers 60% protection against carriage is estimated to prevent more cases than a strategy allowing for direct protection only (Fig. 2). However, a considerable disease burden remains because carriage is low in young children  thus the herd immunity effects generated are limited. Implementing a one-off large-scale catch-up could reduce the annual number of cases considerably. Ten years after the implementation of routine infant vaccination with 1–17 year old catch-up (strategy E), the annual number of cases is estimated to fall by 71%. Over time, however, cases increase as vaccine immunity wanes, and the catch-up cohorts age. Routinely vaccinating adolescents could result in a sustained reduction in cases in the long term, but is likely to be more effective in the short term in combination with catch-up. Adolescent vaccination either with, or without, catch-up is the most favourable decision economically (though still above a £30,000 per QALY threshold at £39,000 and £40,200 respectively, Table 4). In the base case, this model suggests early routine vaccination (strategy A) could be cost-effective at a willingness to pay of £30,000 per QALY if the vaccine were to cost £15 per dose. Routine adolescent vaccination with catch-up (strategy G) could be cost-effective at £32 per dose. 3.3 Scenario and sensitivity analysis Results were most sensitive to changes in disease incidence and case-fatality; with higher incidence and case-fatality, vaccination prevents more cases and deaths and is more economically favourable (Supplementary Table 2). In the probabilistic analysis of the cohort model, the results were sensitive to changes in the parameters with the greatest uncertainty, such as the annual cost of care for those with sequelae and the quality of life loss for those individuals (Supplementary Fig. 2). More cases and deaths are averted through vaccination when the carriage prevalence is lowered (Supplementary Fig. 3) and the discounted cost per QALY is reduced to £69,700 and £86,300 for routine infant vaccination with (strategy E) and without (strategy A) large-scale catch-up, respectively. The predicted annual cases were similar when using a simple preferential population mixing structure or one using self reported leisure contacts. However, cases increased more quickly once immunity from large-scale catch-up had waned, in the model using self reported contacts (Supplementary Fig. 3). Reducing the proportion of vaccine-preventable strains from 100% to 75% in the dynamic model resulted in a higher cost per QALY gained (£131,800) for routine infant vaccination (strategy A). Models were also sensitive to varying vaccine effectiveness, duration of protection and particularly vaccine cost (Fig. 3 and Supplementary Tables 2 and 3). The discount rate has a large impact on the cost per QALY gained (Table 4). None of the strategies considered were cost-effective at a willingness to pay of £30,000 using Treasury recommended discount rates (at £40 per vaccine dose). Using differential discounting, with a lower discount rate for health benefits compared to costs (which allows for increasing value of health benefits), did result some in strategies appearing cost-effective (assuming a 60% vaccine efficacy against carriage). 4 Discussion These are the first models to comprehensively assess the potential impact of introducing vaccines which have the capacity to protect against capsular group B meningococcal disease in England. Our results indicate that introducing a ‘MenB’ vaccine, which provides direct protection only, into the routine infant schedule, could prevent 27% of meningococcal cases per birth cohort, and that this could be cost-effective at £9 per vaccine dose. Substantial impact occurred if the vaccine disrupted carriage as well as preventing disease. In this scenario, the most efficient programme in the short term appears to be routine infant vaccination with catch-up, which after 10 years could reduce annual cases by 71% and be cost-effective at £17 per dose. Models were sensitive to assumptions around: disease incidence and case fatality; vaccine cost, duration of protection and efficacy; quality of life losses from the disease; and the cost of caring for those with sequelae. Vaccination with MCC has shown that the impact of herd immunity effects can be extremely important when predicting the likely impact and cost-effectiveness of vaccination . We therefore used two types of model, including a transmission dynamic model, in order to appropriately capture potential herd immunity effects . Unlike previous meningococcal models, our carriage estimates were drawn from a recent systematic review and meta-analysis  and we assessed the impact of assuming different mixing patterns in the population using simple preferential mixing and mixing based on self-reported contacts . Our transmission models differ to others previously used to model meningococcal disease [43–45] as they are not capsular group specific. This is not required as the vaccines in development, unlike their predecessors, do not target the capsular group. We chose not to explicitly include N. lactamica as the role of this commensal in affording protection remains unclear, and if N. lactamica remains at the current level, any protection is implicitly included through the use of a case: carrier ratio. That said, our results are consistent with previous models in that a vaccine able to disrupt carriage can have substantial added benefits in terms of disease prevention. There are limited data available to inform some parameters, in particular relating to the nature of the new vaccines and the proportion, type and costs associated with sequelae following meningococcal disease. These results are nevertheless important because decisions regarding vaccination are starting to be considered and the scenarios presented cover the most likely features of the vaccine. We used hospital admission data from two time periods to calculate disease incidence. The ability of hospital data to accurately capture the ‘true’ number of cases in unknown, but other sources such as laboratory reports are likely to be less complete, and we feel the true incidence probably lies within the range explored. Assumptions regarding the proportion of survivors with sequelae come from a recent systematic review of meningococcal meningitis only , so we may be underestimating the proportion of sequelae following meningococcal disease. Sequelae costs were based on dichotomising individuals into those with mild and severe sequelae, when in fact there is a considerable range in severity. Previous models have faced the same paucity of data [43,46] and we therefore chose to investigate parameter uncertainty in the cohort model through the use of probabilistic analysis, with particularly wide distributions for sequelae parameters. We did not consider possible replacement effects or adverse effects due to the loss of natural boosting through reduced carriage in the dynamic model (of meningococci or other nasopharyngeal flora) due to the paucity of data. The dynamic results here, therefore, may be optimistic. There was concern in the UK about the possibility of serogroup replacement following MCC introduction, though there is no evidence to suggest that this has occurred . However, serotype replacement has tempered the impact of the pneumococcal conjugate vaccine (introduced in the UK in 2006) . The results from our models, therefore, should be taken as an indication to the relative merits of different strategies, particularly as the models have a long time horizon over which many factors can change, including strain distributions, disease levels and population structure. In the UK, the Joint Committee on Vaccination and Immunisation can only recommend the introduction of a vaccine if it is deemed to be cost-effective . Our models indicate that a ‘MenB’ vaccine could substantially reduce disease in England, and be cost-effective if competitively priced. The results are of relevance, to other countries with similar epidemiology considering the introduction of one of the new meningococcal vaccines, though reapplication of our models with country specific data may be desirable in some circumstances (for example, if there are considerable inter-country differences in vaccine implementation costs). Our model results were sensitive to assumptions around the profile of the vaccine, disease incidence and case fatality and sequelae, including quality of life losses and costs of care. Further data on whether the new vaccines disrupt carriage should be forthcoming, and better information on the duration of protection will also emerge from the results of ongoing clinical trials [50,51], which will help to reduce some of the uncertainty in our models. Efforts are also being made to improve the information surrounding the impact of the disease in terms of QALY losses and care for those requiring sequelae [52–54]. 5 Conclusion Using cohort and dynamic models we have shown that a ‘MenB’ vaccine has the capacity to significantly reduce meningococcal disease in England and that vaccine programmes could be cost-effective if the vaccine is competitively priced.