The main results of this analysis, demonstrate numerous benefits of a vaccination strategy. These benefits include a reduced number of HZ and PHN cases, an increase in health-related quality of life as captured by the QALY, in addition to a reduction in hospitalisations, consultations and prescription costs for the vaccinated population. While sensitivity analyses illustrated that the model results are sensitive to some inputs, the vast majority of scenarios resulted in ICERs well below the cost-effectiveness thresholds commonly used in the UK .
This model takes into consideration a number of factors which contribute to its robustness. First, the model is population-based and therefore of direct relevance to decisions concerning vaccination, as this is done at the population-level. In addition, the model takes into account the ageing of the population over the duration of the model, adjusting for their mortality, the likelihood of contracting HZ and PHN and the efficacy of the vaccine. Though it was assumed that the vaccine would have lifetime duration of efficacy, the model is able to accommodate any changes to vaccine durability. Also this model incorporates epidemiological data that were obtained from a very large dataset (GPRD) , these input parameters for the model could be considered robust and representative of clinical practice in the UK.
Similar to other cost-effectiveness models, the main limitations of this study relate to the uncertainty surrounding some parameter estimates used in this model. One of these limitations concerns the assessment of disease severity in both HZ and PHN. The results of the severity split in HZ and PHN is quite different when the SPS and GPRD data are compared. For the base case we have opted for the SPS split for both HZ and PHN as we believe that this is the most accurate representation available from the literature, as special care was taken with regards to diagnosis in this study due to the use of the validated ZBPI questionnaire . Assessment of severity in GPRD was based on treatments prescribed  rather than actual pain measured and therefore is expected to be less reliable than SPS data, where measurements were geared to providing clinical benefits. Experts further confirmed that the majority of patients experience pain at HZ onset which confirms the higher validity of the SPS data.
Utility weights were available from several sources [13, 15, 29] and Oster et al.  utilities, obtained in a PHN population, were selected for the base case due to the focus on neuropathic pain. A recent publication investigating the quality of life in neuropathic pain reported utility weights of 0.67 for mild, 0.46 for moderate and 0.16 for severe pain , indicating that the Oster values are appropriate for this study. In addition a recent publication by Van Hoek et al.  arrived at similar utility scores for the different pain states, i.e. 0.78, 0.61 and 0.27 for mild, moderate and severe pain respectively. In the standard approach in economic evaluations, the utility gain from prevention is the utility that would otherwise be lost due to illness. An extension of the standard utility model and analysis of prevention interventions is offered by the utility-in-anticipation concept introduced by Cohen and Henderson . This concept acknowledges the fact that the utility resulting from preventive measures such as vaccination follows immediately after vaccination until the time when the outcomes was expected. Furthermore this utility will depend on the anxiety associated with both the perceived risk of infection and the perceived effectiveness of the vaccination in reducing that risk. This model does not include any such gains and therefore could underestimate the total utility gained from vaccination. Obviously, had this been included, this would have resulted in lower ICERs.
A 3.5% discount rate for both costs and benefits was used in the base case analysis as suggested by the current National Institute of Health and Clinical Excellence (NICE) guidelines. A lower discount rate of 1.5% for outcomes was selected in the sensitivity analysis to account for the long-lasting effects of the vaccine, resulting in lower ICERs of £9,893 per QALY gained for the third party payer perspective. This reflects the previous guidelines set by NICE, which recognised that differential discounting is appropriate in certain cases . This is because vaccination programmes accrue their cost in the present but may not observe their benefits until the future. This can be seen in the case of the vaccine preventing HZ where those vaccinated may likely be in their early 50s and 60s, but the incidence of HZ increases with age, thus the benefit of the vaccine may not be observed until the medium- to long-term. Discounting health benefits created by vaccines with long-term effects at the regular discount rates can negatively affect their true benefit by underestimating the cost-effectiveness of the vaccine .
Modelling the appropriate duration of vaccine efficacy is also a significant issue. Extensive SA were conducted to assess the impact of different duration of protection and showed that the highest ICERs were obtained when assuming a 10-year duration of efficacy, or utilising a waning rate. Most of the sensitivity analyses were still below £30,000/QALY.
Another limitation relates to the available vaccine efficacy data. Firstly, as the SPS trial did not include patients aged 50 to 59, the model assumed that the efficacy values for those aged 60 to 69 would be relevant for this younger population. Secondly, the SPS trial reported a 22% reduction in pain for those developing HZ , which may consequently reduce the pain associated with PHN, but as this effect was difficult to incorporate into the structure of the model as such, it was not taken into account directly.
Most of the resource use data were also taken from GPRD . Though we are confident that primary care resource use was accurately recorded, there is less certainty over the secondary care data, as there were only a few referrals or hospitalisations due to HZ or PHN. It is possible therefore that the estimated treatment costs were underestimated. The underestimation of costs does have the advantage that it does not favour the vaccine arm and therefore represents a conservative approach. A sensitivity analysis which varied all health care costs 20% above and below their base case value found that this had minimal impact on the ICERs. Furthermore, the model provides a conservative estimate of the value of herpes zoster vaccination. By not incorporating the common ocular and neurological complications (other than PHN) of HZ, including keratitis, iritis, retinal necrosis,  meningitis, encephalitis, and myelitis,  due to current lack of available data, the model may underestimate the potential health benefits and cost savings resulting from vaccination. Therefore future analyses, in the form of additional retrospective or prospective studies, would be of interest in order to model more accurately such potential disease pathways following herpes zoster and their impact on costs and quality of life.
With regards to a real-world vaccination strategy, it is worth nothing that the results presented in this analysis would only apply for the first years of a vaccination programme where a "catch-up" programme would be instituted for older patients. Following this period, a vaccination strategy would typically include younger cohorts (for instance, those aged 50-69) as other older adults will have already received the vaccine. As a result, in these later years, the cost-effectiveness of vaccination would improve, as illustrated in this study by the lower ICERs associated with relatively lower age groups.
A health economic evaluation of the new live attenuated vaccine against herpes zoster in England and Wales was recently published by Van Hoek et al.  The Van Hoek et al. model employs a different categorisation of HZ/PHN pain states by including a state of clinically relevant pain (CRP) to characterize both moderate and severe pain, while it assumes a limited duration of efficacy with the use of a waning rate,. In addition, there are differences in several input parameters such as HZ incidence rates, disease-specific utilities, and vaccine price applied. Even though a direct comparison of the base case results of the two analyses is not possible, the reported sensitivity analysis applying the maximum vaccine protection duration (100 years) for the cohort aged 65 in the Van Hoek et al. study (£5,660/QALY), produced almost identical results to our study (£5,583/QALY) when we set the vaccine price and population size equal.
In addition, a previous model by Edmunds et al.  had estimated the potential cost-effectiveness of vaccination prior to the availability of data on the clinical efficacy of such a vaccine. Structurally, this previous model differs from ours due to a different and less detailed utilisation of pain states and because it was not able to model the multiple efficacies (direct and indirect) now known of the vaccine. Despite these differences, as well as the use of several input parameters which relied on secondary data, considering a cohort aged 65, the base case resulted in ICERs of similar magnitude to our study, with values below £10,000 per QALY gained from a NHS perspective.
A third cost-effectiveness study by Pellissier et al.  has been published in the US, and while results cannot be directly compared to the UK due to differences in epidemiology and related health care costs, the general model structure has many similarities to the current model. Age-specific and severity-specific data were considered where possible. Vaccine efficacy was modelled using several dimensions, with both models allowing the use of a waning rate (however neither of the two studies included this feature in the base case analysis). In both studies, retrospective database analyses were performed to inform economic inputs and the resulting ICERs were determined to be cost-effective using locally accepted thresholds. Therefore, it confirms the robustness of the methods used in this analysis.
There are a number of areas where additional research could further improve the accuracy of the model. Continued validation of the duration of efficacy of the vaccine, as mentioned above, remains one of the key areas for research.