Appendix 1: Technical report for Cost-effectiveness of Pneumococcal Vaccines for Adults in the US

Model Structure

We developed a deterministic, static mathematical model to project the incidence and outcomes of pneumococcal disease over the lifetime of a closed cohort of persons in the United States starting at 50 years of age. The model had a lifetime time horizon, a payer perspective, and used a 3% discount rate for costs and benefits. Costs were expressed in 2012 dollars.

Given the risk of pneumococcal disease can be elevated by certain chronic and immunocompromising conditions, the cohort was initially stratified into three health states based on pneumococcal disease risk groups as shown in Table S1 according to the prevalence of these groups in the U.S. population at age 50 (Weycker et al. 2010).We tracked these health states together with disability health state due to pneumococcal disease,until death or maximum age of 100 and projected the incidence of pneumococcal disease and all-cause mortality events on an annual basis. At each year, we first projected the portions of the cohort that would be in the disability state or the death state. The disability state was created to capture long-term disability that may occur subsequent to a pneumococcal disease episode (such as neurologic complications, amputations, etc). For the remaining cohort, we projected changes each year in the distribution of persons in each of the health states defined in Table S1 based on Weycker et al. (2010) with simplifying assumptions and linear extrapolation between the ages listed in the Table S9(distribution of risk groups) . We assumed persons experiencing pneumococcal disease remained susceptible to pneumococcal disease in the future, except for those in the disability state or dead. Pneumococcal disease events and all-cause mortality events projected include invasive pneumococcal disease (IPD), non-bacteremic pneumococcal pneumonia (NBPP) outpatient, NBPP inpatient, IPD- and NBPP-related disability (new cases), IPD- and NBPP-related disability (from a previous cycle), IPD-related death, NBPP-related death, death of those previously disabled, and death due to other causes. Finally, Figure 1 depicts the probabilistic structure utilized to distribute pneumococcal disease events, all-cause mortality events, and changes in pneumococcal disease risk group events.

Observed Incidence of Pneumococcal Disease with Vaccination

Annual age- and risk-specific rates of IPD (bacteremia and meningitis), NBPP (both inpatient and outpatient), and corresponding case-fatality rates were based on Weycker et al (2010) (Table S2). In that study, IPD rates were derived from the Active Bacterial Core Surveillance (ABCs) for calendar year 2006 and Kyaw et al (2005). NBPP rates were derived using data from a population-based study of the impact of 7-valent pneumococcal conjugate vaccine (PCV7) on the burden of community-acquired pneumonia (CAP) (Nelson et al 2008), and assuming 30% of all cases of CAP were pneumococcal in etiology. Inpatient rates were further adjusted down to exclude CAP episodes that were bacteremic (7%), and thus were included in IPD rates.

The 30% assumption is consistent with the assumption used in literature such as Smith et al (2012), who in turn cited Metersky et al (2010), Fry et al (2002), and Carbonara et al (2009). Based on expert opinion and literature review, Huang et al (2011) also estimated that 20% (10-25%) of outpatient pneumonia and 30% (20-40%) of inpatient pneumonia were attributable to pneumococcus in 2004 in the United States.

Estimating the Incidence of Pneumococcal Disease in the Absence of Vaccine Effects in the Population

Estimating the impact of vaccination on reducing pneumococcal disease in the cohort depends on the background incidence of pneumococcal disease in the absence of vaccine effects. Given that there is currently a recommendation for vaccination of adults with 23-valent pneumococcal polysaccharide vaccine (PPSV23) in most countries, background incidence data do not exist. As a result, we needed to back calculate what the incidence of pneumococcal disease would be in the population had vaccination not been introduced. Table S2 was the observed incidence rates of pneumococcal diseases from Weycker et al 2010. Table S11 shows our estimatesof the hypothetical “no vaccination” incidence rates of pneumococcal diseases for the U.S. population 50 and older per 100,000. The following describes our methods for estimating these incidence rates.

Two factors may affect observed disease burden: 1) the use of PPSV23 among adults, 2) the introduction of PCV7 and then PCV13 in children. To account for the effect due of PPSV23 vaccination in the adult population, we followed the method proposed by Fry et al (2002). In particular, we applied PPSV23 effectiveness, observed vaccination uptake levels, and observed serotype coverage of PPSV23 to estimate incidence rates of IPD and NBPP (Table S3). Serotype coverage for NBPP was assumed to be the same for IPD. Because data about vaccine uptake level over time were also not available, we assumed that vaccine uptake level remained fairly stable over time and utilized observed year 2006 vaccine uptake rates by age category (National Health Interview Survey, 2006). To account for elderly who might be vaccinated after 74 years of age, we assumed 8% were vaccinated at age 75 years and 7% at age 85 years (Fry et al, 2002). To be consistent with the available age categories for vaccine uptake rates and estimated disease burden, mean PPSV23 effectiveness over an age category was used.

Pilishvili et al (2010) reported an indirect effect on IPD due to the widespread use of PCV7 in young American children. It is possible that PCV13 may have a similar herd effect on reducing the IPD disease among persons not vaccinated with PCV13. Therefore, the disease burden estimated with the method by Fry et al (2002) might need to be further adjusted by a possible additional indirect effect.

Pilishvili et al (2010) has also shown that serotype distributions changed after the introduction of PCV7 among children. Within a few years post-PCV7 introduction, there was a significant reduction in the proportion of IPD among adults caused by PCV7 serotypes and a slight increase in the proportion of IPD caused by non-PCV7 serotypes. PCV13 was first recommended for infants and young children in February 2010, and changes in adult pneumococcal disease epidemiology since PCV13 introduction similar to the trends observed post-PCV7 introduction have been documented (Moore et al 2012). For modeling purposed, we used the Pilishvili et al 2010 data to project changes that may occur once the full impact of PCV13 use in the pediatric population has been realized.

Because there may be an inherent relationship between the size of indirect effect and the changes in serotype distribution, we devised an algorithm to simultaneously project the size of the indirect effect and the changes in serotypes. The projected indirect effect was used to adjust the disease burden estimated in the absence of PCV13 adult vaccination. The projected change in serotype distribution was used to project future adult pneumococcal disease incidence. This algorithm was applied in sensitivity analysis.

Algorithm for Projecting Indirect Effect and for Serotype Coverage in PCV13 era

The algorithm is summarized as the following: Pilishvili et al (2010) reported IPD cases per 100,000 for 2006-2007 by each serotype for two age groups, 18-64 and 65 and older. To obtain projected IPD cases by each serotype, we multiplied the reported incidence for each type by an assumed percentage change in incidence due to PCV13 indirect effect for each serotype. The percentage change in total number of IPD cases across all types before and after PCV13 introduction provided the estimate for the size of the PCV13 indirect effect. The serotype coverage for the vaccine-types was calculated as a proportion between numbers of IPD cases for vaccine-types and total number of IPD cases of all types. These projections were done for each age group separately.

Note that this algorithm projected correlated changes in both disease incidence and serotype distribution, consisting of both reductions in disease incidence for the PCV13 serotypes and projected increases in disease incidence for non-PCV13 serotypes, as well changes in the projected serotype distribution associated with remaining disease. However, the model did not require the above algorithm to be used for future serotype coverage assumptions. Serotype coverage projections could be varied individually in the model, for example, if no data is available to inform the relationship between indirect effect and serotype coverage change.

It is expected that the wide spread PCV13 use in children is likely to have some degree of indirect effect in adults. If the magnitude of the indirect effect is large, the need for PCV13 use in adult may be dampened (e.g. Musher, 2012). Since the full magnitude of the indirect effect is still unknown, we assumed in the base case that there was no indirect effect or serotype changes due to PCV13 (Table S4, 2nd column), favoring PCV13 use in adults . In a sensitivity analysis, we assume that widespread use of PCV13 in children will cause some degree of changes in serotype coverage of PCV13-but-non-PCV7-types as well as non-vaccine types (Table S4, 4th column). Specifically, we assumed PCV13 was likely to exert the same magnitude of change on vaccine serotypes as observed with PCV7 reported by Pilishvili et al (2012). The algorithm projected no change in IPD incidence due to PCV7 serotypes, 87.5% reduction in IPD incidence due to PCV13-but-non-PCV7 serotypes, and 28.8% increase in IPD incidence due to non-PCV13 serotypes. The projected indirect effect was a 23.1% reduction in overall IPD incidences among 50-64 years old and a 26.6% reduction among 65 years and older. The corresponding projected serotype coverage was 67.8% and 20.6% for PPSV23 and PCV13, respectively, among 50-64 years old, and 56.4% and 17.7% for PPV23 and PCV13, respectively, among 65 years and older. This sensitivity analysis corresponded to PCV13 use in adults when PCV13 in children has reached its full potential herd effect. Another sensitivity analysis explored a situation where indirect effect is only 50% of the projected full indirect effect (Table S4, 3rd column).

Although the model allowed serotype coverage for IPD and NBPP to be different, the calculated indirect effect and serotype coverage for IPD was applied to NBPP, given the absence of data from which to derive NBPP specific reductions due to PCV13 introduction in children. However, in one sensitivity analysis, we only assumed indirect effect for IPD but not for NBPP (Table S4, 5th column).

Projected Vaccine Uptake Rates

The model assumed vaccine uptake rates vary by the health state (pneumococcal risk) and by age group, but not by vaccine. The assumed vaccine uptake rates are based on self-reported PPSV23 coverage at year 2008 by National Health Interview survey (Table S9, Projected Vaccine Uptake for age-50cohort):

Estimating Pneumococcal Disease Incidence Rates, Mortality Rates, and Rates of Disability due to Pneumococcal Disease

For each vaccination strategy, the incidence rates at each cycle year were estimated by multiplying the estimated "no-vaccination" pneumococcal disease incidence rates with (1 minus the projected vaccine uptake rates * projected serotype coverage rate * projected vaccine effectiveness). In the base case, we assumed projected vaccine uptake rates did not differ between the two vaccines. Projected serotype coverage differed between the two vaccines, but did not differ among IPD, NBPP outpatient, or NBPP inpatient cases. Projected vaccine effectiveness differed between the two vaccines and between IPD, and NBPP, but not between NBPP outpatient and NBPP inpatient cases.

Weycker et al (2010) provided estimates for case-fatality rates by pneumococcal disease risk status as well as age group. Annual mortality rates were then estimated by multiplying estimated incidence rates for IPD and NBPP inpatient cases times the case-fatality rates.

Following Smith et al (2012), disability after IPD was modeled using meningitis as a proxy, taking into account that not all meningitis is disabling but other IPD syndromes can be. The disability rate after NBPP was assumed to be 50% of that after IPD in the base case. Meningitis incidence rates were based on Weycker et al (2010). The probability of death after being disabled was assumed to be 20%, 30%, and 40%, respectively, for patients who were previously in healthy state, immunocompetent with comorbidities state, and immunocompromised state.

Sequential Use of PPSV23 and PCV13 Vaccines

We defined each vaccination strategy as a sequence of vaccinations at different age points (more than 1 year apart). This would be considered a re-vaccination strategy. The model allowed any number of years in between two vaccination age points with a maximum three vaccination administrations with specification age at vaccination points (referred to as vaccination age points) over the modeling horizon. At each vaccination age point, if two different vaccines were used within a year, it was considered a sequential use of the vaccines. In this paper for the sequential PCV13—PPSV23 regimen, the two vaccines were assumed to be administered 2 months apart, although the time can be adjusted in the model.

Because the combined effectiveness of the sequential regimen is not known, disease reduction associated with each vaccine were modeled independently and then combined additively (except for the first year) to approximate the incidence rate and mortality rate for the combination use of vaccines. For the year in which the two vaccines were administered, the disease incidence rate and mortality rate reduction associated with the second vaccine used was proportioned by the time with second vaccination. In the first year, PCV13 vaccine was assumed to provide protection for 52 weeks, PPSV23 (administered 8 weeks after PCV13) was assumed to provide protection for 44 weeks. Starting in the week of PCV13 vaccination (week 1), it is assumed that individuals are protected against disease caused by the 13 serotypes in PCV13. Starting in week 9, the benefit of the 11 additional unique serotypes contained in PPSV23 is added to the PCV13 benefit.

Estimation of Vaccine Effectiveness

Due to the absence of data comparing PPSV23 to PCV13 efficacy in adults, assumptions around PCV13 effectiveness against IPD and NBPP as well as PPSV23 effectiveness against NBPP were based on expert opinion using estimates generated from a Delphi panel engaged in 2011 by United BioSource Corporation for healthy and immunocompetent with comorbidities groups (Table S5 & S6) (see Appendix A2 for description of Delphi methods).

PPSV23 effectiveness against IPD for the healthy group was based on Moberley et al (2008). Ten studies involving 35,483 participants were included for this outcome with 15 events in the vaccinated group and 60 events in the control group. PPSV23 reduced the risk of all IPD with a pooled estimated odds ratio (OR) of 0.26 (95% confidence interval (CI): 0.15 to 0.46; random effects model). This translates into a protective vaccine efficacy of 74% (95% CI: 56 to 85%) for the average follow-up time through three years. Therefore, we assumed the first-year effectiveness was 82.7% and waned at a rate of 8.27% per year for 10 years. This assumption corresponds to a 74% average effectiveness for the first three years after receiving PPSV23 vaccination.

The model assumed that effectiveness for immunocompetent people with comorbidities was a portion of that for the healthy people. Specifically, Smith et al (2012) assumed that vaccine effectiveness for immunocompetent persons with comorbidities was 85% of that of healthy persons in the base case for PPSV23 against IPD. We utilized similar relative vaccine effectiveness estimates based on estimates derived by the 2011 Delphi panel. The Delphi panel provided first-year effectiveness estimates for 50-64 year old immunocompetent people with comorbidities against IPD and NBPP (except for PPSV23 against IPD) (Table S6). The vaccine effectiveness for this group relative to the healthy group was used as an estimate of vaccine effectiveness among immunocompetent persons.

Effectiveness estimates presented at the June 2012 ACIP meeting (Stoecker C, June 20, 2012 ) were used for the immunocompromised risk group (Table S7).

We assumed that effectiveness decreased as the age increased. Based on Smith et al (2008 & 2012), we assumed the effectiveness for 65-year-old and 80-year-old persons to be 89.2% (=83%/93%) and 72% (=67%/93%) of that for a 50 year old, respectively. Because the model allowed up to three vaccinations at any age between 50 and 100, linear extrapolation on the effectiveness was applied for the age in between the three age groups (50, 65, and 80).

Recent studies suggested that PPSV23 induces antibody responses that remain above the levels of unvaccinated adults for 5 to 10 years or more (Grabenstein, Manoff 2012). In the base case we assumed PPSV23 waned over a period of 10 years.

Given the absence of long-term effectiveness data for PCV13 in adults, we assumed vaccine effectiveness for PCV13 to wane over a period of 15 years. While all vaccine-waning patterns were assumed to be linear in the base case, the model allowed other waning patterns to be used. For revaccination, effectiveness was assumed to take on the waning pattern and the effectiveness of the newly administered vaccine multiplied by a revaccination effectiveness loss factor. In the base case, we assumed no effectiveness loss at revaccination due to lack of data. We conducted 2 sensitivity analyses using the following waning patterns: 1. PCV13 and PPSV23 were assumed to have equivalent waning rates of (a) 10 years or (b) 15 years; 2. PCV13 was assumed to wane over a period of 20 years, rather than 15 years as in the base case.

We assumed vaccine effectiveness to be the same for NBPP inpatient and NBPP outpatient cases in the base case. The vaccine effectiveness by health states and by age for both vaccines against IPD and NBPP are summarized in Table S10.

Mortality Probabilities Due to Causes Other Than Pneumococcal Disease

The model accounts for mortality due to pneumococcal disease as well as other causes. Mortality rates are assumed to vary, depending on the three health state categories. Mortality probabilities were calculated by first obtaining the contribution of death from each health state category in the general population based on CDC 2006 death rates (National Vital Statistics Reports, Vol. 57). This calculated contribution of death from each health state category (Table S8) was then applied to each of the health states.

The year 2006 life table provided the total number of death by age for the whole population. Together, the number of deaths at age x for each health state was estimated. The number of people who survived to age x for each health state was estimated using transition probabilities among non-fatal health states (i.e., distribution of risk groups) (Table S9). This information allowed estimation of the probability of dying at age x for each of the health states. Subtracting from the probability of dying at age x the probability of death due to IPD and NBPP and the probability of dying after being disabled due to pneumococcal disease for year 2006-2007 (Weycker, 2010), the probability of death due to other causes for each of the health states was obtained. While the estimations were cross-sectional, we assumed they were a proxy for the background mortality probability overtime as the cohort aged.