The cost-effectiveness of using pneumococcal conjugate vaccine (PCV13) versus pneumococcal polysaccharide vaccine (PPSV23), in South African adults

Streptococcus pneumoniae (pneumococcus) remains an important cause of morbidity and mortality. Pneumococcal vaccination is part of the South African pediatric public immunization program but the potential cost-effectiveness of such an intervention for adults is unknown. This study aimed to compare the cost-effectiveness of two widely used pneumococcal vaccines: pneumococcal conjugate vaccine (PCV13) and pneumococcal polysaccharide vaccine (PPSV23) in South African adults, 18 years and older. Four analyses were carried out in a) both the private and public health care sectors; and b) for the HIV-infected population alone and for the total mixed population (all HIV-infected and -uninfected people). A previously published global pharmacoeconomic model was adapted and populated to represent the South African adult population. The model utilized a Markov-type process to depict the lifetime clinical and economic outcomes of patients who acquire pneumococcal disease in 2015, from a societal perspective. Costs were sourced in South African rand and converted to US dollar (USD). The incremental cost divided by the incremental effectiveness (expressed as quality-adjusted life years gained) represented the incremental cost-effectiveness ratio for PCV13 compared to PPSV23. Results indicated that the use of PCV13 compared to PPSV23 is highly cost-effective in the public sector cohorts with incremental cost-effectiveness ratios of $771 (R11,106)/quality-adjusted life year and $956 (R13,773)/quality-adjusted life year for the HIV-infected and mixed populations, respectively. The private sector cohort showed similar highly cost-effective results for the mixed population (incremental cost-effectiveness ratio $626 (R9,013)/quality-adjusted life year) and the HIV-infected cohort (dominant). In sensitivity analysis, the model was sensitive to vaccine price and effectiveness. Probabilistic sensitivity analyses found predominantly cost-effective ICERs. From a societal perspective, these findings provide some guidance to policy makers for consideration and implementation of an immunization strategy for both the public and private sector and amongst different adult patient pools in South Africa.


Introduction
(including adults), reducing pneumococcal disease incidence even among unvaccinated individuals.
Data on the true incidence of pneumococcal infections in South Africa are uncertain [1], but given the extent of the HIV epidemic in South Africa, pneumococcal disease is a significant problem [1]. A study before the era of routine PCV vaccination of children, documented that the incidence of invasive pneumococcal disease (IPD, consisting of pneumococcal meningitis and bacteremia) was much higher in HIV-infected patients compared to HIV-uninfected patients (214 cases versus 6.5 cases per 100,000 population, respectively) [11]. Furthermore, comprehensive antiretroviral therapy (ART) rollout during the period of the study did not significantly change the incidence of IPD in HIV-infected adult patients despite a stable prevalence of HIV infection [11]. Conversely, in another South African study, where 89% of IPD cases over 5 years of age were HIV co-infected, an increased incidence of IPD with increasing age was reported in both HIV-uninfected and HIV-infected persons, with an additional peak in the HIV-infected group in the later childhood years. Moreover, 39% of HIV-uninfected persons with IPD had chronic conditions and 19% of HIV-infected persons had chronic conditions predisposing them to IPD [12]. In total, 93% of all the cases had a condition predisposing them to pneumococcal disease according to the US Advisory Committee on Immunization Practices (ACIP) list of conditions requiring pneumococcal vaccination [12].
The mortality from pneumococcal disease remains high. A South African surveillance study found that the case-fatality rate (CFR) for meningitis caused by S. pneumoniae was 55%, while that for bacteremia was 23%, among adolescents and adults (aged 15 and older) [13]. Amongst other risk factors, increasing age and HIV infection were associated with increased mortality in these patients [13].
The high morbidity and mortality resulting from pneumococcal infections indicate a need for reducing the burden of IPD in HIV-infected patients via means other than ART alone. Vaccination in adults and children, providing direct and indirect protection to the community, is one way of reducing this burden [13]. However, a study on the vaccination of HIVinfected Ugandan adults with PPSV23 did not find it to be effective in preventing disease caused by S. pneumoniae compared to placebo [14]. Nevertheless, the authors later reported that despite a persistent excess of all-cause pneumonia in these vaccine recipients in a 6-year follow-up, they had paradoxically a 16% lower overall mortality risk in the PPSV23 vaccinated group compared to the placebo group [15]. Conversely, a second study showed that PCV7 vaccination in mostly HIV-infected adults in Malawi was effective in prevention of recurrent IPD [16]. The Community-Acquired Pneumonia Immunization Trial in Adults (CAPiTA) study was a randomized, double-blind, placebo-controlled trial of 84,496 Dutch adults (�65 years). The study evaluated the efficacy of the PCV13 vaccination in preventing vaccine-type strains of pneumococcal community-acquired pneumonia (CAP), non-invasive and non-bacteremic CAP, and IPD [17]. Results indicated that PCV13 could reduce CAP by 45.6%; non-invasive and non-bacteremic CAP by 45%; and IPD by 75%. This trial is the basis for the vaccine efficacy data used in this current study [17].
PCV13 and PPSV23 are indicated and used in adults in South Africa. PPSV23 is recommended in South Africa as potentially beneficial to any individual and very effective in young otherwise healthy individuals, targeted at high-risk groups when there are cost considerations [1]. There is currently no National Department of Health policy on adult pneumococcal vaccination in South Africa, although a national guideline is in development by expert clinicians representing the various national societies of interest in South Africa. The only national recommendation for both PPSV23 and PCV13 vaccination in adults in South Africa is a section on vaccination included in the recently updated management guideline for communityacquired pneumonia in adults in South Africa [18]. The recommendations largely follow the Centers for Disease Control and Prevention recommendations for pneumococcal vaccination in adults in the United States.
Given this, the question arose whether an adult vaccination strategy would be cost-effective in the South African setting. The South African health care market consists of two tiers, the public health care sector and the private health care sector [19]. The provision of health care services in South Africa is divided along socioeconomic lines, with only approximately 50% of the health expenditure coming from the government, but approximately 86% of services being provided by the public health care sector, in public health care facilities (clinics, hospitals). The private sector caters to a small population with medium to high levels of income, in private health care facilities. This sector consists of those covered by medical insurance provided by medical schemes and those paying out-of-pocket for health care services. In general, PPSV23 is available and reimbursed in both the private and public sectors. PCV13 is available in the private sector and whether it is reimbursed at all, or how it is reimbursed (for example from medical scheme savings or a special benefits package), depends on the individual medical scheme the person belongs to. PCV13 is not available in the public sector.
The objective of this study was therefore to estimate the cost-effectiveness of the use of PCV13 compared to PPSV23 in South African adults in both the private and public health care sector from a societal perspective. Due to the high prevalence of HIV in South Africa, and the association of HIV infection with the incidence of pneumococcal disease, the cost-effectiveness of targeted pneumococcal vaccination of HIV-infected individuals was also investigated. Four scenarios were, therefore, created: i) A mixed private health care sector population (consisting of HIV-infected and HIV-uninfected individuals-the complete private health care sector population in South Africa); ii) A mixed public health care sector population (consisting of HIVinfected and HIV-uninfected individuals-the complete public health care sector population in South Africa); iii) A model for the HIV-infected population in the private health care sector; and iv) A model for the HIV-infected population in the public health care sector.

Model description
A previously published global pharmacoeconomic model [20] was adapted and populated to represent the South African population of adults aged 18 years and older. A probabilistic Markov-type model was developed in Microsoft1 Excel to depict the lifetime clinical and economic outcomes of alternative strategies for adult vaccination against pneumococcal disease. The study included both the private and public health care sectors and distinguished between i) the mixed population cohort (including both HIV-infected and -uninfected patients), as well as ii) a cohort of HIV-infected patients. Accordingly, we report on four theoretical South African population cohorts (private sector, public sector, HIV-infected private sector, and HIV-infected public sector).
The characteristics of each cohort of patients were stratified by age and risk of pneumococcal infection. Patients were classified into three risk categories: low-, moderate-and high-risk of acquiring pneumococcal disease ( Table 1).
The study's cohort of patients then entered the model via two potential scenarios: vaccination with PPSV23 or vaccination with PCV13. The vaccination of patients was dependent on their age, risk profile, and vaccination history.
The vaccination strategy was based on supplier coverage calculations. For patients in the private sector mixed and HIV-infected models, vaccination was assumed to be performed only for high-risk patients in the 18-49 and 50-64-year age groups (40%). Overall, 30% of patients in the 65-74 age group (all risk groups) were considered to be vaccinated, while 20% of all patients aged 75-99 (all risk groups) were considered to be vaccinated. For the public sector mixed and HIV-infected models, vaccination was assumed to be performed only for high-risk patients in the 18-49 and 50-64-year age groups (1%). For subsequent age groups, 1% of all patients in each risk group were considered to be vaccinated.
As a model base case, in order to calculate the actual number of patients vaccinated based on the above vaccination strategy, the following assumptions were made: No revaccination after completion of the vaccine regime; the effectiveness of vaccines are linked to risk stratification of patients, i.e. vaccines are less effective in higher risk patients (for example, HIV-infected patients).
Expected outcomes were evaluated for each person in the model population on an annual basis, from model entry, through to the end of the modelling horizon. The modelling horizon was set to a lifetime horizon of 82 years and was modelled from a societal perspective, considering both direct medical and indirect costs. Indirect costs considered were loss of paid productivity due to the disease. Furthermore, in accordance with the South African published Pharmacoeconomic Guidelines [26], the discount rates of 5% per annum for both costs and benefits were used. In each year, pneumococcal-related outcomes were projected for each person in the model cohort based on age, risk profile, vaccination status, vaccine type (PPSV23 or PCV13), and time since vaccination.
A schematic diagram of the model is shown in Fig 1. The model consists of several modules, including disease, death and costs, which impact on one another. The inputs required for the disease module include disease rates (incidence), model population, vaccine coverage and vaccine effectiveness. Inputs required for the death module include case-fatality rates. Inputs required for the cost module include direct medical and indirect unit costs, vaccine coverage and vaccine effectiveness. Outcomes include the number of IPD and non-bacteremic pneumonia (NBP) cases; case-fatalities for IPD and NBP; life years (unadjusted and quality-adjusted); medical care costs; indirect costs, and vaccination costs. Model health states included death, low-risk, medium-risk, high-risk, IPD, inpatient NBP and outpatient NBP. A schematic diagram for the Markov model, indicating the transitions between model health states, is shown in the supplementary material in [20].
Bacteremia in the study's model included all cases of invasive pneumococcal disease, except for meningitis, which was handled separately in the model. Non-bacteremic pneumonia in the model included all cases of pneumonia caused by S. pneumoniae, which are not included under the definition of bacteremia.
A cost-effectiveness threshold, or willingness-to-pay (WTP) threshold, of approximately $13,889 (R200,000) was used and was based on three times the GDP per capita for South Africa, based on previous funding decisions in the private sector [27,28,29]. According to this threshold, a vaccine with incremental cost-effectiveness ratio (ICER) less than $13,889 (R200,000) per quality-adjusted life year (QALY) is considered as providing good value for money (cost-effective), whereas treatments with an ICER greater than $13,889 (R200,000) are typically viewed as a poor use of resources (not cost-effective) [30].

Data inputs
The majority of data was sourced from the literature. However, where data points were not available in the literature or where there was uncertainty regarding some inputs, an advisory board consisting of eight key opinion leaders was set up to obtain consensus answers using a modified two round Delphi panel method [31]. For numeric responses provided by the Delphi panel, the coefficient of variation (CV) was calculated. The CV was used to determine whether consensus was achieved [32]: • CV between 0 and 0.5-good consensus, no additional round.
• CV between 0.5 and 0.8-less than satisfactory degree of consensus, possible need for an additional round.
• CV greater than 0.8-poor degree of consensus, definite need for an additional round.
For dichotomous responses (Yes or No), consensus was achieved when 70% of the panelists answered Yes or No for sensitive variables and 60% of the panelists answered Yes or No for non-sensitive variables. If consensus was not achieved after the first round, a second round was undertaken. Before the second round, the first round's results were discussed between the panelists. If no consensus was achieved after the second round, disclosure of this fact and sensitivity analysis on the parameter under question would be performed.
Clinical data. The population was categorized in five age bands (18-49, 50-64, 65-74, 75-84 and 85-99 years), and further stratified into three risk groups (low, moderate and high) as illustrated in Tables 2-5, which show the base case model inputs for each of the four models. The model allowed for patients to move between different risk groups as they aged in a dynamic way.                    Cost effectiveness of pneumococcal vaccines for South African adults  Cost effectiveness of pneumococcal vaccines for South African adults     The number of persons in the private health care sector per age group was calculated from Council for Medical Schemes data for 2014 [33]. The number of persons in the public health care sector per age group was calculated by subtracting the number of persons in the private health care sector from the total South African population for 2014 [34]. For the private sector HIV population, the HIV prevalence (24 per 1,000 beneficiaries [35]), together with the total population from the Council for Medical Schemes [35], were used to calculate the number of patients in the HIV-infected private sector model. The number of patients in the HIV-infected public sector model was calculated by subtracting the private health care sector HIV-infected population from the total HIV-infected population [36]. The age distributions for both the private and public sector HIV-infected populations were obtained from the Actuarial Society of South Africa (ASSA) model [36].
The annual incidence of bacteremia, meningitis, and all-cause pneumonia per 100,000 persons, assuming no vaccination and no herd effects, by age and risk group, was calculated from Nunes et al. [11] for the established highly active ART (HAART) era. The incidences of meningitis, bacteremia and hospitalized and out-of-hospital all-cause non-bacteremic pneumonia was further split between risk groups using the relative risks (RR) from Kyaw et al. [37] (RR = 8.65 for the moderate-risk group and RR = 44.4 for the high-risk group). For the two HIV models, the incidences were also calculated from Nunes et al. [11], but using the incidences for the HIV-infected only patient group for all risk groups and by not adding any additional risk using the RR from Kyaw et al. [37], as the risk was already for an HIV-infected group. Further to this, the percentage of cases of bacteremia and meningitis that were due to serotypes included in the PCV13 and PPSV23 vaccines, were estimated from data on IPD received from the National Institute for Communicable Disease [12]. It was assumed that these percentages were representative for all four models (private and public health care sector mixed models, and HIV-infected private and public health care sector models).
Indirect (herd) effects were obtained from Delphi panel consensus by asking what the percentage reduction in pneumococcal disease incidence in the adult population is, given the potential herd effects due to the widespread use of PCV13 in young children.
The annual number of all-cause deaths in the general population was obtained from the Statistics South Africa death notification findings [38]. For the HIV models, the rate per 100 persons per age group was multiplied by an additional factor of 3.8 to reflect the higher risk associated with HIV-infected persons compared to HIV-uninfected persons [39]. For the mixed models, only the high-risk group was multiplied with this additional factor.
The CFRs for bacteremia and meningitis were calculated separately [40]. The CFR was calculated by dividing the number of cases with a fatal outcome for each age group by the total number of cases for that age group. It must be noted that the CFR for bacteremia is for bacteremic pneumonia only. Consensus was obtained on this matter during the Delphi panel meeting using the consensus definitions explained above. As there is an additional risk of death for HIV-infected individuals, the CFRs for bacteremia and meningitis for the high-risk group was multiplied by an additional hazard ratio (HR) of 1.69, which is the additional risk of death for HIV-infected individuals compared to HIV-uninfected individuals in the mixed models [40]. For the HIV models, the CFR for each age and risk group was multiplied by this additional HR [40], as all individuals in these models had the higher risk of mortality compared to HIV-negative individuals. No information was found regarding the CFR for persons hospitalized with all-cause non-bacteremic pneumonia in South Africa. Instead, the ratio of the 30 day bacteremic pneumonia mortality (19.5) to the 30 day non-bacteremic pneumonia mortality (5.1) from Lin et al. [41] was used to calculate the non-bacteremic pneumonia CFR from the bacteremic pneumonia CFR. The CFR for outpatient care for all-cause non-bacteremic pneumonia was set to zero, as per the US publication [42]. The same health-state utilities and disutilities as used in the US publication were assumed for all four models, as these values were not available for South Africa [43,44]. A systematic review on health utilities in pneumococcal disease found no studies on health utilities for individuals with pneumococcal disease in sub-Saharan Africa [45]. These utilities and disutilities were used to calculate the QALYs in the model.
For all effectiveness inputs, except for the effectiveness of PCV13 against all-cause NBP, the same effectiveness values as used in the US model were assumed for all four models.
The effectiveness of initial vaccination with PPSV23 against vaccine-type IPD for immunocompetent individuals aged 50 years and older, and for immunocompromised individuals aged 65 years and older, was based on Smith et al. [46]. It was assumed that effectiveness of PPSV23 in immunocompetent individuals aged between 18 and 49 years was the same as for persons aged 50 years [46]. For immunocompromised individuals aged between 18 and 50 years, the initial effectiveness was based on Shapiro et al. [47], while for those aged between 51 and 64 years, effectiveness was estimated by interpolating between values for individuals aged 50 years and 69 years [48]. The rate of decline of protection since vaccination for immunocompromised individuals aged between 18 and 64 years was based on Smith et al. [46]. The effectiveness of vaccination with PPSV23 against all-cause NBP was assumed to be zero. This is based on various published sources and assumptions used in other published economic studies [46,48].
The effectiveness of PCV13 against vaccine-type IPD for immunocompetent individuals was based on the results of the per-protocol population of the CAPiTA study [17]. The perprotocol population was used as this corresponds to immunocompetent individuals throughout the follow-up period. The effectiveness against IPD was anchored on individuals with a mean age of 73 years in the CAPiTA study. Protection was assumed to be stable over the first 5 years of the model, based on the follow-up period in the CAPiTA study (mean of 3.97 years follow-up). The rate of change in the effectiveness of PCV13 for those younger and older than 73 years was equal to 50% of the value for PPSV23. The rate of decline in effectiveness of PCV13 over time (after the initial 5 years) was assumed to be 50% of PPSV23 values. These rates were obtained from an expert panel, including members of the US CDC pneumococcal workgroup. PCV13 effectiveness in immunocompromised individuals was assumed to be 78% of the values for the immunocompetent individuals, based on results from a 9-valent pneumococcal conjugate vaccination in children (with and without HIV) [49]. PPSV23 and PCV13 effectiveness were assumed to be 0% after year 16 of the model time horizon.
The effectiveness of PCV13 against all-cause NBP for immunocompetent individuals was estimated using the effectiveness of PCV13 against vaccine-type non-bacteremic and noninvasive CAP as per the CAPiTA study [17], as well as the percentage of all-cause NBP that is due to serotypes contained in PCV13. To localize the PCV13 effectiveness values for all-cause pneumonia to the South African context, two values were required: the proportion of all-cause pneumonia due to S. pneumoniae and the serotype coverage for PCV13 from IPD. The proportion of all-cause pneumonia due to S. pneumoniae was set to 100.0%, as the values provided for mortality and incidence of all-cause pneumonia is only for pneumonia caused by S. pneumoniae, and not for pneumonia caused by any other pathogens, as the definition requires. The value of serotype coverage for PCV13 from IPD (42.5%) was obtained using the Delphi method consensus. The rate of change with age and rate of decline over the modeling horizon were estimated similarly to that described for the effectiveness of PCV13 against IPD. The initial PCV13 vaccine effectiveness in immunocompromised individuals was assumed to be 65% of the values for immunocompetent individuals, based on Klugman et al. [49]. The waning method was based on absolute levels, and not on the rate of decay. This means that values represent effectiveness in the corresponding year; decline in the interval is estimated via a linear function.
Cost data. Unit costs were obtained in South African rand (ZAR), and the model was also run using this currency. Conversion of the currency to US$ (USD) was performed for ease of interpretation and results are shown using both currencies. Conversion was performed using the daily conversion rate from ZAR to USD for 30 November 2015 (where 1 USD = 14.4 ZAR) [50].
Direct medical costs taken into consideration included the cost of the vaccine (PCV13 and PPSV23) and its administration; the cost per event of hospitalization and medication for the treatment of bacteremia, meningitis and all-cause pneumonia; as well as the cost per event of outpatient care for the treatment of all-cause pneumonia. For PPSV23 vaccination, an additional administration fee equal to one consultation fee was added since this vaccine requires a prescription. Microcosting was used to calculate event costs for bacteremia, meningitis and all-cause pneumonia. The resource use per event was based on the literature [51][52][53][54][55] and Delphi panel input. Unit costs for the private sector were sourced from the Department of Health (DoH) Database of Medicine Prices [56] and the National Reference Price Lists (RPL). In the public sector, the Uniform Patient Fee Schedule (UPFS), tender prices and National Health Laboratory Services (NHLS) prices were used [57]. Event costs were obtained by multiplying resource use per event with the unit cost for each element. All costs not in 2015 values were inflated to 2015 values using medical services inflation factors derived from Statistics SA Consumer Price Index publications for the relevant years. Cost details for the private health care sector models are shown in Table 6, while cost details for the public health care sector models are shown in Table 7. Details on resource use and costs for events are shown in S1-S4 Tables.
The human capital approach to calculation of productivity loss costs was used. This considers that productivity loss due to morbidity is represented by lost wages during the period of illness [58]. To calculate the costs related to loss of productivity when an employee was sick, the percentage of persons in the workforce in each age category, the average daily wage for that age category, and the number of work-loss days for each disease, per age and risk group, were calculated. The number of persons in the workforce per age group, as well as the average monthly wage of persons in the workforce, was obtained electronically from Statistics SA (N. Roux, personal email communication, July 9, 2015). Levinsohn et al. [59] indicated that HIVinfected persons were 7.9% more likely to be unemployed, thus for the two HIV models, that percentage was deducted from the percentage in the mixed models.
The number of work-loss days per age group and per disease were assumed to be the same as the number of days required for hospitalization for the disease, plus the additional workloss days for recovery at home after discharge, as per the Delphi panel inputs. For out-of-hospital non-bacteremic pneumonia cases, the number of work-loss days at home after discharge was used as per the Delphi panel inputs. Only paid employment was considered by multiplying by the percentage of persons in the workforce. As the duration of work-loss due to disease was short (between approximately 5 and 20 days), it is reasonable to consider that there would not be enough time to substitute the sick worker with a new person and to train that person to take over during the absence. However, the number of days would still have a significant impact on productivity and therefore should be considered in the calculation.

Results
The actual number of adults vaccinated were calculated according to the model input variables. The number of adults vaccinated in the base case are as follows: • Public sector mixed population PCV13 immunization uptake: 57,957 (0.18%) • Private sector mixed population PCV13 immunization uptake: 217,930 (3.38%)

Public health impact
For the mixed scenarios (i.e. including HIV-infected and uninfected patients) in the public and private health care settings, we estimated that vaccination with PCV13 would lead to a reduction of 313 cases (-0.08%) and 716 cases (-0.99%) of IPD, respectively, compared to PPSV23. PCV13 vaccination would also be associated with a reduction in NBP in the public and private health care settings both for patients who are hospitalized and for outpatients (Table 8). Furthermore, PCV13 would lead to a reduction in disease-related deaths (-160 in the public sector [-0.07%] and -354 [-1.17%] in the private sector). Amongst the public health care sector adult cohort in which approximately 31 million adults were included in the model, 1.9 million (6.2%) were classified as high-risk patients. The public sector high-risk age distribution was skewed towards a younger population with 12.6% of patients in the 18-49-year age-band, 8.5% in the 50-64-year age-band and only 1.2% in the 85-99-year age-band. This contrasted with the private sector where 6.5 million adults were included in the model, 98,425 (1.52%) were classified as high-risk patients and they were fairly evenly distributed per age-band (1.8% to 4.3%). It is most likely that these demographic differences have led to the differences seen in mortality benefits.
For the HIV-positive model cohorts the results showed a reduction in IPD cases in both the public and private sectors of 208 (-0.07%) and 241 (-2.90%) respectively, with the use of PCV13. There was also a reduction in NBP in both these scenarios for both hospitalized and outpatients (Table 8). Furthermore, the results indicated that PCV13 use would lead to a reduction in disease-related deaths of 133 in the public sector (-0.05%) and 134 (-2.13%) in the private sector.
According to the above results, all strategies considered in the base case indicated that PCV13 would have an improved public health impact compared to PPSV23.

Cost-effectiveness analysis
The incremental discounted cost-effectiveness results for the four analyses under base case assumptions are shown in Table 9. The undiscounted results are shown in S7 Table in the      The private sector mixed population shows that while one could expect a discounted saving of approximately $1.8 million (R26 million) related to medical care costs over the model lifetime, additional vaccination costs of approximately $3.47 million (R50 million) would be needed to implement the PCV13 vaccination strategy. When including all costs related to the

Sensitivity analysis
Both one-way and probabilistic sensitivity analyses (PSA) were performed to subject the model to uncertainty of the input parameters.
One-way sensitivity analysis. One-way sensitivity analysis was performed by altering single base case model data inputs to higher or lower values. The results for the one-way sensitivity analyses are shown in Figs 2-4. In the figures, high indicates when a parameter was increased, while low indicates when a parameter was reduced. It shows the deviation from the base case ICER when the parameters are adjusted up or down in the model.
One-way sensitivity analysis showed that vaccine price and vaccine effectiveness were the more sensitive inputs to the model. For the HIV private sector model, all sensitivity scenarios resulted in dominant ICERs, except for vaccine price (high) and vaccine effectiveness (low), which resulted in ICERs of $204/QALY (R2,943/QALY) and $63/QALY (R910/QALY), respectively. Therefore, a graph is not shown for this scenario.
One-way sensitivity analysis is also important where significant assumptions need to be made. This is the case for the utility values used in the model. Using local utility values would be advisable, due to possible differences in social weights between countries [61]. However, these values are not available for South Africa and therefore health-state utilities and disutilities were used from a US model. As can be seen in Figs 2-4, it appears that the model was not sensitive to these parameter assumptions.
The updated vaccine prices as of 22 October 2018 were as follows [62,63]: According to Statistics South Africa [64], the average life expectancy at birth for 2019 is 61.5 years for males and 67.7 years for females. These values are relevant to the mixed population (HIV-positive and HIV-negative individuals). Sensitivity analysis was performed to reduce the time horizon in the mixed public and private health care sector models to 43 years (61-18 = 43). The following results were obtained when the time horizon was reduced from 82 to 43 years in the mixed public and private health care sector models: mixed public health care sector ICER/QALY of $962 (R13,854) and mixed private health care sector ICER/QALY of $629 (R9,055).
According to Statistics South Africa [64], the average life expectancy at birth for 2019 without HIV/AIDS is 65.6 years for males and 72.7 years for females. Life expectancy according to Johnson et al. [65], for HIV-positive patients on ART, is between 70% and 86% of HIV-negative adults of the same age and sex. Therefore, a life expectancy of approximately 46 years (65.6 Probabilistic sensitivity analysis. For the PSA, the following probability distributions were used: • For incidence rates, vaccine effectiveness and case-fatality rates, a beta distribution was used, with the number of cases and non-cases in the cohort as inputs. This is relevant as the value of these parameters are probabilities ranging between 0 and 1 [66,67].
• For indirect effects and utility values, a uniform distribution was used, with the minimum and maximum value of these parameters as inputs [66].
• For costs, a log-normal distribution was used, with the mean cost per case and the standard error of the cost per case as inputs. This is relevant, as the value of the costs range from 0 to infinity [66,67].
The PSA ICER scatterplots are illustrated in Figs 5, 7, 9 and 11 and show the differences between PCV13 and PPSV23. The quadrant explanation and detailed results for the PSA are provided in the supporting information. The cost-effectiveness acceptability curves are illustrated in Figs 6, 8, 10 and 12, where the net monetary benefit > = 0 means that direct medical costs (medical care costs plus vaccination costs) and indirect costs associated with PCV13 is less than the value of additional benefit achieved.
For the public health care sector mixed model, the PSA and cost-effectiveness acceptability curve indicated that, with a WTP threshold of $6,944 (R100,000), the probability of having a net monetary benefit > = 0 would be 0.963. With a WTP threshold of $13,889 (R200,000) this probability would be 0.972. The ICERs obtained from the PSA indicate that the ICERs are mostly (97%) cost-effective.
For the private health care sector mixed model, the probability of having a net monetary benefit > = 0 was 0.958 at a WTP threshold of $6,944 (R100,000), and 0.969 at a WTP Cost effectiveness of pneumococcal vaccines for South African adults threshold of $13,889 (R200,000). The majority of ICERs obtained from the PSA were costeffective or dominant (73% below WTP and 24% dominant).
For the public health care sector HIV-infected model, the probability of having a net monetary benefit > = 0 would be 0.980 at a WTP threshold of $6,944 (R100,000), and 0.988 at WTP threshold of $13,889 (R200,000). The results from the PSA indicated that 99% of ICERs obtained were cost-effective or dominant (98% below WTP and 1% dominant) with 1% dominated.  For the private health care sector HIV-infected model, the probability of having a net monetary benefit > = 0 would be 0.973 at a WTP threshold of $6,944 (R100,000), and 0.977 at a WTP threshold of $13,889 (R200,000). The results from the PSA indicated that 98% of ICERs obtained were cost-effective or dominant.

Discussion
This study was performed to evaluate the cost-effectiveness of immunization of an adult population against pneumococcal disease in South Africa by comparing PPSV23 and PCV13. Four patient cohorts were evaluated. Firstly, given the existence of both a strong private and public health care system in South Africa, the analysis was performed for both these sectors Cost effectiveness of pneumococcal vaccines for South African adults separately. Secondly, given the high prevalence of HIV infection in South Africa, we considered the cost-effectiveness in the two sectors for only the HIV-infected population and secondly for the total South African mixed population (all HIV-infected and uninfected people). These four patient cohorts were then stratified into three groups according to the risk of contracting pneumococcal disease, namely, low-risk, moderate-risk and high-risk patients. The cost-effectiveness results are based on specific, realistic, vaccination strategies for each of the four patient cohorts and their disease risk characteristics. The base case focus was on vaccinating more of the high-and moderate-risk population and less or none of the low-risk population. Results were then presented based on the incremental population impact of immunizing only a proportion of the stated risk pool. The base case results indicate that the use of PCV13 in favor of PPSV23 is highly cost-effective in both public-sector cohorts with ICERs of $771 (R11,106)/QALY and $956 (R13,773)/QALY for the HIV-infected and mixed population, respectively. The private sector cohort showed similar highly cost-effective results for the mixed population (ICER $626 (R9,013)/QALY) and in the HIV-infected risk pool it showed that PCV13 dominated PPSV23. Even when indirect costs (productivity loss costs) were excluded, all four models show highly cost-effective results.
Before the CAPiTA trial [17], many cost-effectiveness studies were performed based on assumptions and calculations of vaccine effectiveness in adults [68]. Our study used the vaccine effectiveness data based on the CAPiTA trial [17]. Because the CAPiTA trial did not include HIV-infected patients, the PCV13 vaccine efficacy had to be estimated for this cohort of patients. In a study by French et al., PCV7 was effective in preventing recurrent IPD in adult HIV-infected patients [16]. Specifically, the vaccine efficacy against IPD caused by vaccine serotype or serotype-6A was 74% (95% CI: 30%-90%). Smith et al., using an expert panel and Delphi technique, estimated effectiveness of PCV13 in HIV-positive patients for both IPD and NBP [46]. These values are somewhat lower compared to the CAPiTA results for IPD, but almost two times more effective for NBP. In the current study, we opted to keep vaccine effectiveness the same for both the HIV-infected and HIV-uninfected patient cohorts in our base case (based on CAPiTA effectiveness) and subject it to sensitivity analysis.
When comparing our results to other studies in the post-CAPiTA era, the cost-effectiveness study performed by Mangen et al. in the Netherlands [20] closely resembles the South African cost-effectiveness model architecture in terms of age groups, risk stratification and vaccination strategies. Mangen et al., however, compared a single dose of PCV13 to no vaccination and results imply that, in that scenario, PCV13 is cost-effective [20]. Other studies in many other countries show similar results, although the vaccination strategies are diverse [69][70][71][72]. It must be noted that these studies are diverse in the population to be vaccinated, vaccination strategies used (sequential use of PCV13->PPSV23, PCV13, revaccination, etc.), comparator (PPSV23 or no vaccine), seroepidemiology and treatment cost. For example, four studies found vaccination with PCV13 to be cost-effective compared to vaccination with PPSV23 [61,70,72,73], and two studies found vaccination with PCV13 to be cost-effective compared to no vaccination [20,72]. In contrast to this, three studies found vaccination with PCV13 in sequence with vaccination with PPSV23 to not be cost-effective compared to PPSV23 alone [69,74,75]. Two of these studies [53,58] reported results for a population of 50 years and older and this might be the reason for the results. In the second study [58], when analyzing patients of 65 and above, the results were cost-effective. The results from the UK study [59] is not surprising given the very low burden of disease related to pneumococcal disease. From the above, it appears that there is a global body of evidence that supports vaccination with PCV13 as a cost-effective intervention when compared to no vaccine or PPSV23. It appears that sequencing PCV13 with PPSV23 is not cost-effective. The present study's results support the literature, and further differentiates between different risk populations such as HIV-infected patients and mixed populations. This additional information is of particular importance when considering policy implementation strategies in countries like South Africa where there is a high incidence of HIV/AIDS.
As mentioned before, the price of PCV13 remains the most sensitive parameter in the model. In our base case, we took a conservative approach and assumed that the private and public sector price for PCV13 will be the same, but that an administration cost will be incurred in the private sector, but not in the public sector. This conservative approach was followed because a final price for PCV13 for adults was not agreed with the public sector at the time of this publication. However, if one compares the pediatric public sector price for PCV13, it is feasible that one will expect a significantly lower price for PCV13 than what was used in the base case ($18 (R266) vs. $55 (R794) per pediatric dose in the public and private sector, respectively, [60]). When using the pediatric public sector price for PCV13 in our model, both the public sector analyses become dominant. Also, updating the models to the vaccine prices available in October 2018 resulted in both public sector analyses becoming dominant. For both private sector analyses, the ICER/QALY increased, but remained cost-effective.
The model is sensitive to vaccine effectiveness and given the lack of PCV13 vaccine effectiveness data for HIV-infected people, this remains an important shortcoming in our study. However, as was shown in the sensitivity analysis, by decreasing the sensitivity by 10%, vaccination in this group of people remains a cost-effective investment suggesting our results are robust despite the lack of this data.
Vaccine effectiveness was assumed to be stable for the initial 5 years of the modeling horizon (no waning), based on stable vaccine effectiveness during the follow-up period (mean 3.97 years) observed in a post hoc analysis of CAPiTA [76]. As with other studies, another limitation is the fact that no empirical evidence is available after this time period.
The indirect effect of pediatric vaccination in South Africa and its impact on the cost-effectiveness of PCV13 vaccination in adults remains uncertain. The present study's sensitivity analysis showed, as can be expected, that there is an indirect correlation between the benefits of pediatric vaccination and the ICER. This makes intuitive sense as the more impact the indirect effect has, the lower the burden of pneumococcal disease and therefore the lower the potential impact of adult immunization.
A further limitation of the study is that unpaid lost productivity due to illness, informal health sector costs, such as patient-time costs, unpaid caregiver-time costs and transportation costs, as well as additional non-medical costs, such as future consumption unrelated to health, social services, legal or criminal justice, education, housing and environmental costs were not considered [77].
Probabilistic sensitivity analyses indicated that ICERs were predominantly cost-effective, with only a small percentage of ICERs being dominated or not cost-effective. There is only minor uncertainty regarding the cost-effectiveness of the scenarios at a WTP threshold of $13,889 (R200,000).
The model that was used in this study is a static model and the main limitation (as opposed to a dynamic model) is that the probability of disease exposures is constant over time and unaffected by vaccination. This is probably not realistic if there are high vaccination rates. In the dynamic transmission model, the force of infectivity can change over time due to changes in the size of the susceptible population. A recent systematic review paper [78] of cost-effectiveness analysis publications for adult PCV13 included only static models. The dynamic transmission model requires more data on the transmission aspects and this data is scarce, which leads to many assumptions. This is one of the reasons why most adult pneumococcal vaccine modeling is static. A static model can account for pediatric herd effect.
While it is preferable that local utility values should be considered in economic appraisals, the lack of access to local data required the use of international data. Literature clearly demonstrates statistically significant inter-country differences for health utility values [79]. However, under one-way sensitivity analysis, the use of US-based utility values does not impact on the results from a societal perspective.
The uptake of PCV13 in both the public and private sector is conservative in the model. From a societal perspective, the results are therefore sufficiently robust to provide some guidance to policy makers for consideration and implementation of an immunization strategy for both the public and private sector. It further allows consideration to implement immunization strategies for different patient pools, notably different ages and risk profiles. Given the resource constraints experienced by South Africa, the results of this study provide some insights into the pneumococcal immunization interventions that will provide value for money.
Supporting information S1  Table. Cost-effectiveness of PCV13 versus PPSV23 vaccination in South African adults in the public and private health care settings for the mixed and HIV-positive populations (undiscounted). ICER, incremental cost-effectiveness ratio; Δ, incremental difference (PCV13 -PPSV23); USD, United States dollar; ZAR, South African rand. a Costs included in the calculation of the ICER are direct medical costs (medical care + vaccination costs) + indirect costs. (DOCX) S8