We estimated the costs, health benefits, and cost effectiveness of vaccination of high risk groups with the 13 valent pneumococcal conjugate vaccine on top of the current risk based vaccination programme with the 23 valent polysaccharide vaccine. This was done because the existing programme with the 23 valent polysaccharide vaccine is likely be continued despite the potential introduction of a risk based programme using the 13 valent pneumococcal conjugate vaccine. In addition our risk estimates for pneumococcal disease were estimated in the current situation in which a risk based programme using the 23 valent polysaccharide vaccine is already in place (albeit with a low uptake of vaccination).
As infants are already vaccinated with the 13 valent pneumococcal conjugate vaccine, we restricted our analysis to high risk patients aged 2 years and older. The perspective was from that of the National Health Service, as recommended in the United Kingdom.9
Model and population
We developed a cohort model to determine the cost effectiveness of vaccinating specific high risk groups with the 13 valent pneumococcal conjugate vaccine. Groups included in this analysis were based on a recent study among patients admitted to hospital in England with culture confirmed invasive pneumococcal disease, which compared the prevalence of clinical risk factors in the general population with that in patients admitted to hospital with invasive pneumococcal disease.7
The study sample comprised 22
298 patients admitted to hospital between April 2002 and March 2009 with an admission record in the hospital episode statistics database for England that could be linked with the dataset of the national invasive pneumococcal disease laboratory held at the Health Protection Agency.7
In the current analysis we differentiate between people who are immunocompromised, such as those with HIV, asplenia, or splenic dysfunction or who respond poorly to the vaccine, such as people with chronic kidney disease; and those in immunocompetent risk groups such as patients with chronic heart, liver, or respiratory disease and people with diabetes.7
The analytical time frame of the study was until 2021 (we assume that after this time the additional benefits of vaccination would be negligible). However, we extrapolated the long term effects of invasive pneumococcal disease over the full lifetime of the participants in each cohort—that is, until death or 100 years.
Incidence of invasive pneumococcal disease and mortality risks
Using the most recent data available we estimated age group and risk group specific incidences. Firstly, we calculated age specific incidences of invasive pneumococcal disease for the general population, including cases confirmed by polymerase chain reaction and culture from the epidemiological year 2009-10 (in this paper we refer to epidemiological years, which run from July to June, unless stated otherwise).8
These incidences were subsequently used to estimate the incidence of invasive pneumococcal disease in high risk people using the prevalence of clinical risk factors among the general population and the prevalence among the linked patients admitted to hospital with invasive pneumococcal disease.7
From the same databases we estimated the age specific share of meningitis and empyema to the total invasive pneumococcal disease burden to allow the inclusion of specific costs related to these outcomes. We also obtained age group and risk group specific case fatality ratios for invasive pneumococcal disease from this same study.7
Invasive pneumococcal disease sequelae
Invasive pneumococcal disease may lead to long term sequelae, especially in the case of meningitis. We obtained the risk of different types of sequelae from a recent meta-analysis.9
As patients can have multiple sequelae, we assigned all possible combinations on the basis of the prevalence of the individual conditions and reweighted them such that the overall risk to develop any sequela was equal to the pooled prevalence of 31.7% as estimated by the meta-analysis. We obtained the losses in overall quality adjusted life years (QALYs) using the most severe QALY weight in the combination.
Non-bacteraemic pneumococcal pneumonia
To assess whether to include an effect of the 13 valent pneumococcal conjugate vaccine on non-bacteraemic pneumococcal pneumonia in the base case we looked at the impact of the seven valent pneumococcal conjugate vaccine on the overall incidence of non-bacteraemic pneumonia in high risk children. For this we obtained the number of episodes of non-specified pneumonia (ICD J18.X, mentioned in any diagnostic code) and the number of deaths for the same cases (within 30 days of admission) for the years 1997-98 up to 2009-10 (data from 2002-03 to 2009-10 were used for deaths) from the hospital episode statistics database in children aged less than 5 years. Next, we divided individual cases into risk or non-risk groups based on the same ICD codes (see appendix 9 in supplementary file) as used for invasive pneumococcal disease, and we calculated incidences. An interrupted time series analysis showed that the incidence of pneumonia requiring admission to hospital in non-high risk children aged less than 5 years (that is, those eligible for vaccination) was significantly reduced after the introduction of the seven valent pneumococcal conjugate vaccine, whereas the incidence in high risk children of the same age was not significantly reduced (see appendix 1 in supplementary file). Based on the striking difference between risk and non-risk groups, and the additional uncertainty about the contribution of Streptococcus pneumoniae
to non-bacteraemic pneumonia, particularly in high risk children, we decided not to include an overall impact on non-bacteraemic pneumonia in the base case analysis for the high risk groups. We did, however, explore the potential impact of including an effect against non-bacteraemic pneumonia in specific analyses. For this we used the data on age specific incidence for all cause pneumonia for the year 2010 from hospital episode statistics and projected these forward assuming the same incidence as in 2010. Next we assumed that S pneumoniae
would be the causal agent in 42% of the patients in high risk groups admitted to hospital with non-bacteraemic pneumonia on the basis of the results of the two most recent UK studies available.10
We then assumed that the contribution of the vaccine serotypes to pneumococcal pneumonia would decline in line with the herd effect of the infant vaccination programme on invasive pneumococcal disease.
In virtually all countries the introduction of the seven valent pneumococcal conjugate vaccine was followed by a large reduction in invasive pneumococcal disease owing to vaccine serotypes in vaccinated and unvaccinated age groups, with the indirect benefits in some age groups partially offset by a concomitant increase in invasive pneumococcal disease due to non-vaccine serotypes.4
This was also the case in the United Kingdom in which the seven valent pneumococcal conjugate vaccine was introduced in September 2006 with a vaccination schedule of 2, 4, and 13 months, and catch-up vaccination for children aged up to 2 years.8
In April 2010, the 13 valent pneumococcal conjugate vaccine replaced the seven valent vaccine in the infant vaccination programme.
To predict the future decrease in invasive pneumococcal disease due to vaccine serotypes in unvaccinated age groups, we divided the serotypes into those covered by the seven valent vaccine and those included in the 13 valent vaccine but not in the seven valent pneumococcal conjugate vaccine. In both cases we used age group specific (2-4, 5-14, 15-44, 45-64, and >64 years) UK data on incidence of vaccination before and after the introduction of the seven valent vaccine. The prevaccination period included the incidence data for the years 2000-06, whereas the post-vaccination period included data up to four years after the introduction of the vaccine (2006-10). Using the age group specific annual incidence (adjusted for underlying trends in case ascertainment) we fitted a Poisson regression model adjusting for the population size to predict the future reduction in cases of invasive pneumococcal disease due to the vaccine serotypes (see appendix 2 in supplementary file).
We consequently used the predicted annual decrease in vaccine serotypes to predict the incidence of the additional serotypes (except for serotype 3, see below) in the 13 valent vaccine—that is, we assumed that the herd effects for the additional serotypes in this vaccine would be similar to those observed for the serotypes in the seven valent vaccine after the introduction of the routine infant vaccination programme using the seven valent pneumococcal conjugate vaccine in 2006.7
The only difference was that we delayed the herd effects for the six additional serotypes in the 13 valent vaccine by one year as the introduction of the vaccination programme using the 13 valent pneumococcal conjugate vaccine was not combined with a catch-up programme. This assumption is supported by the most recent data from the Health Protection Agency, which show no indication of any herd effect yet in people aged 5 years and older, 15 months after implementation of the routine infant vaccination programme using the 13 valent pneumococcal conjugate vaccine.12
Furthermore, in the Netherlands, where the vaccination programme using the seven valent pneumococcal conjugate vaccine was launched without a catch-up, herd effects were not observed in the first year after implementation in contrast with the United Kingdom.13
We did not include serotype replacement effect in the model as we assumed that it would not affect the incremental cost effectiveness ratio because changes in invasive pneumococcal disease due to non-vaccine serotypes are expected to be the same irrespective of the implementation of the risk group programme.
Vaccine efficacy, number of vaccine doses, duration of protection
Although the efficacy of the seven valent vaccine in healthy infants is well established, the available data for risk groups and adults is scarce, with most studies reporting data on immunogenicity rather than efficacy.3
Data on the efficacy of the 13 valent pneumococcal conjugate vaccine is limited12
; the current licence for the use in infants and children from 6 weeks to 5 years of age and adults aged 50 years and over was based on immunogenicity rather than efficacy data14
(see appendix 3 in supplementary file for an overview of available data).
Considering the limited data available, we carried out a formal elicitation of expert opinion on vaccine related variables to construct a probability distribution that represents the experts’ knowledge and uncertainty.15
The objectives of the elicitation were to estimate the efficacy of the 13 valent pneumococcal conjugate vaccine (against invasive pneumococcal disease and non-bacteraemic pneumococcal pneumonia) and the duration of protection after one dose of the vaccine (as in the base case analysis) or two doses of the vaccine. Importantly, recent data from our group show that the serotype 3 component of the 13 valent pneumococcal conjugate vaccine seems to be ineffective against invasive pneumococcal disease caused by this serotype.16
Therefore, in the model we also assumed no protection against disease or carriage for serotype 3.
Specific details on the method of elicitation can be found in appendix 4 in the supplementary file. Briefly, we asked five members of the Pneumococcal Subcommittee of the Joint Committee on Vaccination and Immunisation to give an estimate for the efficacy of the 13 valent pneumococcal conjugate vaccine in risk groups based on the available efficacy data for the seven valent pneumococcal conjugate vaccine and immunogenicity data for both the seven valent and the 13 valent pneumococcal conjugate vaccines. We used the estimates to create distributions for vaccine effectiveness using the Sheffield elicitation framework.15
Final distributions can be found in table 1.
Table 1 Variables used in economic model
Life years and QALY estimates
As the life expectancy between the general population and high risk groups differs,17
we calculated specific background mortality for people at high risk (and for the general population for validating purposes). Data were gathered from the Royal College of General Practitioners database (including 0.8 million patients; more than 1% of the UK population) over a period of six years (2005 to 2010). We grouped the patients by risk factor (based on Read codes mapped to ICD-9 codes) and calculated the number of person years and deaths in the high risk group. Using these data we calculated background mortality (see appendix 5 in supplementary file). We also calculated the mortality for non-risk groups and validated these against life tables from the Office for National Statistics.19
In addition to life years gained we also calculated QALYs gained by vaccination. For patients admitted to hospital for invasive pneumococcal disease, we used losses in QALYs of 0.0079 per case for bacteraemia and 0.0232 per case for meningitis.20
We assumed that non-bacteraemic pneumococcal pneumonia resulted in a QALY loss of 0.006 per case.21
In addition to acute losses in QALYs, we also linked specific losses in QALYs to the sequelae due to meningitis based on a Dutch study22
(see table 1 for specific losses in QALYs).
All costs are reported in pounds sterling at 2009-10 prices. Where necessary we inflated these using the hospital and community health services pay and price index.23
As the perspective was from that of the healthcare provider, we included only direct costs. We used recommended procedures to estimate the costs for patients admitted to hospital with invasive pneumococcal disease. The NHS healthcare resource group software was used, which combines procedure codes and ICD-10 diagnostic codes to output the most relevant healthcare resource group code. We subsequently assigned these codes a cost from the National Schedule of Reference Costs for NHS trusts. As the patients included in our analysis are all high risk, we included only those for which it was likely that the invasive pneumococcal disease episode was the main cause for admission to hospital—defined as those patients who had a primary diagnostic code related to an invasive pneumococcal disease code (see appendix 6 in supplementary file). Table 1 displays the costs and probabilities related to invasive pneumococcal disease. The costs of hospital admission for non-bacteraemic pneumococcal pneumonia were based on reference costs for pneumonia. We used the weighted average costs based on the number of non-elective admissions for pneumonia without complications (NHS reference costs code WADZ11C). Patients who had meningitis without sequelae were assumed to have a single outpatient appointment after discharge; we obtained the cost of treatment and care for patients with sequelae after meningitis from a previous cost effectiveness analysis.24
The total cost per dose of 13 valent pneumococcal conjugate vaccine was estimated at £56.61, consisting of the price of the vaccine (£49.10) and administration costs (£7.51).
Scenario and sensitivity analysis
We carried out univariate, threshold, scenario, and probabilistic sensitivity analyses. In the univariate sensitivity analyses, relevant variables were based on the 5% and 95% quantiles to explore the impact of uncertainty around each variable. A threshold analysis was done in which we varied the vaccine price to investigate the effect on the incremental cost effectiveness ratio.
In specific scenario analyses we explored the impact of changes in vaccine efficacy, vaccine waning, delaying the herd effect of the infant vaccination programme using the 13 valent pneumococcal conjugate vaccine, assuming life expectancy of the general population (rather than using the life expectancy of people in high risk groups), and the effect of discounting.
For the probabilistic sensitivity analyses, we generated variables using Monte Carlo sampling, with outcome values generated by running the model 5000 times using Latin hypercube sampling. When quantitative data about uncertainty around variables were available we used log normal and β distributions (see table 1 for specific distributions). When only a single point estimate was available, we assumed a normal distribution with a coefficient of variation of 0.25. For all the sensitivity analyses it was assumed that the vaccination programme would be launched in 2012-13 (two to three years after the infant programme).
Outcome measures and cost effectiveness analysis
The simulation model tracks the incidence of invasive pneumococcal disease and non-bacteraemic pneumococcal pneumonia, the number of deaths, costs, QALYs, and life years. We calculated the net costs, life years gained, and QALYs by summing all the costs, life years, and QALYs and calculating the differences for the evaluations with and without vaccination. The incremental cost effectiveness ratio was calculated by dividing the net costs by either the life years gained or QALYs gained. Health effects and cost were both discounted at 3.5% according to the UK guidelines.25
In the analyses we compared the possible impact of vaccination using the 13 valent pneumococcal conjugate vaccine with that of the current situation. Currently, adults aged more than 65 years and people in at risk groups aged 2 years or more are recommended to be vaccinated with the 23 valent polysaccharide vaccine26
; however, uptake of the vaccine is relatively low, especially in those aged less than 65 years (see appendix 7 in supplementary file).7
We assumed that the 13 valent pneumococcal conjugate vaccine, would be used in addition to the 23 valent polysaccharide vaccine.
Finally, we assumed that the uptake of the 13 valent pneumococcal conjugate vaccine would be similar to the annual influenza programme in the United Kingdom, at 34.5% in the age group 2-16, 53.6% in the age group 16-65, and 72.4% in the age group 65 and older27
and that vaccination with the 13 valent pneumococcal conjugate vaccine would be offered irrespective of previous vaccination with the 23 valent polysaccharide vaccine.