Our results show that a routine PCV program is expected to prevent, over the first 5 years of life of a birth cohort, approximately 1040-1650 cases of primary endpoint pneumonia, 470-740 hospitalizations due to severe pneumococcal diseases, and 30-50 pneumococcal deaths, as well as avert 740-1180 DALYs, depending on the valences of PCVs (PCV7 to PCV13) used. We found that vaccine price per dose is the main driver of cost-effectiveness, followed by discount rate, case-fatality rate, and vaccine efficacy. Since the UNICEF purchase price of PCV7 or of other vaccines in the pipeline are unknown, in the present analysis we chose to use a base-case value of $3.5, which has been suggested as a possible long term cap price by AMC; we also evaluated the implications of lower and higher costs, including another potential AMC cap price of $7.0 [61
]. Using a general heuristic of an incremental cost-effectiveness ratio less than GDP per capita ($360) as a proxy for "good value for money" [62
], the results demonstrated that, from the societal perspective, all PCVs examined in the present study-PCV7, PCV9 or PCV10, and PCV13-would be considered very cost-effective
up to a unit vaccine price of $1.24. Using a higher threshold of three times GDP per capita ($1080), all PCVs were shown to be cost-effective
up to a per-dose vaccine price of about $4.20.
Note that the cost-effectiveness results for each of the PCVs were calculated using the "no vaccination" baseline comparator. Evaluating each PCV using a comparator of the currently used PCV7 or of the next lower valence of PCV (competing choice analysis) would have been less meaningful to this study, since our analyses do not assume increased programmatic costs or adverse events for higher-valence PCVs. More importantly, using the same baseline comparator of "no vaccination" with which other immunization programs are commonly evaluated allows for comparison not only of current and prospective PCV programs, but also of PCVs against other immunization programs that will ultimately compete for limited financial resources.
Despite the increasing global level efforts to introduce PCVs into developing countries with high burdens of pneumococcal diseases, no published study has provided a detailed assessment on the potential impact of a PCV in an individual, low-income GAVI-eligible country using local data. Our study provides such an assessment for PCVs in The Gambia. In doing so, we took advantage of the availability of recent local data on incidence of pneumonia [9
] as categorized by the recent WHO standards [25
], disease burden of pneumococcal meningitis and sepsis [1
], clinical consequences of pneumococcal diseases [9
], serotype distribution [47
], and medical and non-medical costs. Also, our study extends the base-case results for PCV9 to project the effect of PCVs of different valences--PCV7, PCV10, and PCV13. Our findings provide more detailed information on the distribution of the health outcomes across different types of pneumococcal diseases as well as project the potential impact of PCVs based on comprehensive sensitivity analyses that take into account possible waning of vaccine immunity, serotype replacement, and herd immunity.
The results of this analysis are likely a conservative estimate of the health benefits of PCV intervention for the following five main reasons:
• First, we assessed only three main types of pneumococcal diseases, excluding the potential role of PCVs in the prevention of otitis media and serious clinical syndromes, such as cellulitis or septic arthritis, that have also been attributed to S. pneumonia
infection but for which local data are lacking. Although the burden of these clinical syndromes is relatively small in The Gambia (e.g., according to the 2004 GBD study [1
], the annual DALYs for otitis media for the entire population was <500, in contrast to ~58,640 for lower respiratory disease), exclusion of such syndromes underestimates the burden of pneumococcal diseases.
• Second, we assessed the potential impact of PCVs on childhood mortality by simulating vaccine effects on disease-specific mortality. While a 16% reduction in all-cause
mortality by PCV9 was reported separately by Cutts et al. [8
], we relied on incidence, vaccine efficacy, and case-fatality data related specifically to the three major pneumococcal syndromes, as discussed previously. However, given the plausible "hypothesis that pneumonia contributes to many more deaths than are directly attributed to pneumonia in studies using verbal autopsies, and that community-acquired bacteraemia is a greater cause of childhood mortality than previously recognized" [8
], our base-case estimates of vaccine effects may be an underestimate.
• Third, we follow the recent WHO standards [25
] in distinguishing all-cause pneumonias, and assume 35% efficacy against primary endpoint pneumonia while also assuming a 1.5% increase
in non-primary endpoint pneumonia, based on our interpretation of trial results [9
]. Although a negative efficacy value of -1.5% against the latter category may seem small, because more than 80% of estimated cases fall within the category of non-primary endpoint pneumonia, the impact on the aggregate population benefit may be non-negligible. Indeed, in our analysis, the number of deaths due to non-primary endpoint pneumonia is higher among vaccinated than in non-vaccinated children (283 vs. 276), offsetting the number of primary endpoint pneumonia deaths averted by PCV9, 42, by about 17%. Our approach that explicitly considers the impact of PCV on non-primary endpoint pneumonias therefore provides a more conservative assessment of vaccine benefits than most previous discussions, which have focused solely on the reduction in primary endpoint pneumonia.
• Fourth, our base-case analysis conservatively assumed the reported value of 22% for PCV9's efficacy against all-serotype pneumococcal meningitis and sepsis [8
], rather than using the theoretical efficacy, for example, of 57% for PCV9, which is computed by multiplying the serotype coverage of PCV9 (62%) by the vaccine efficacy against cases caused by vaccine serotypes (92%) [8
• Finally, our model does not capture explicitly any herd immunity effects over time among non-vaccinated populations. Given that routine immunization programs using PCV7 have been shown to reduce overall pneumococcal carriage rates in children as well as reduce the incidence of invasive pneumococcal disease in adults through indirect effects [65
], we may have underestimated the impact of a PCV program.
The weaknesses and limitations of our study relate to both data gaps and a model structure that could not capture indirect effects. Country-specific data were not available for every parameter, and data quality was variable. While our scenario analysis showed that indirect net benefits could lead to more attractive cost-effectiveness ratios, our intention was to gain insight into possible outcomes rather than generate a precise estimate. Caution is needed in interpreting such results due to multiple factors that limit extrapolation of results obtained in developed country settings to developing countries (e.g., differences in immunization strategy, structure and mixing patterns of the population, overall coverage achieved in the population, serotype distribution, and co-morbidities, etc.) [67
We compared our model-predicted outcomes of hospitalizations and deaths with those from Cutts et al. [8
]. While Cutts et al. has reported a 15% reduction (per-protocol analysis) in "first admission" due to all clinical conditions among children receiving PCV9 [8
], our base-case analysis predicted an 8% reduction in all hospitalizations. When we adjusted the base-case coverage to 100%, model-predicted reduction in hospitalizations increased to 9%. The remaining difference would presumably be due largely to the fact that Cutts et al. reports only first admissions [8
] while our model captures all hospitalization events (including multiple hospitalizations due to repeat episodes by different serotypes) using incidence data from Enwere et al. [9
]. Obviously, under this condition, differences between vaccinated and non-vaccinated groups would be less pronounced. Regarding mortality, as previously discussed, while Cutts et al. has reported a reduced overall mortality by 16% among vaccinated children [8
], our analysis relied exclusively on disease-specific data and did not incorporate reported vaccine effects on overall mortality. As a result, while our model predicted a 9% reduction in disease-specific mortality under a PCV9 intervention, model-predicted reduction in overall mortality over a 5-year horizon was not remarkable ( < 1%). This suggests that the mortality results between the two studies are not directly comparable, primarily due to the difference in choice of mortality endpoints.
While Sinha et al.'s study [24
] does not report country-specific results, according to the under 5 mortality strata of The Gambia, the incremental cost-effectiveness ratio of PCV9 for the country falls between $69 and $138 per DALY averted (in 2000 international dollars). This is based on the study's setting-specific secondary analysis, in which vaccine intervention (at $5 per dose) was credited for reducing hospitalizations and outpatient visits for non-fatal pneumococcal disease cases [24
]. The corresponding value from our base-case analysis, $670 per DALY averted (in 2005 US$), is approximately 5 to 10 times higher than that from the previous study. Although there are multiple possible sources of this observed discrepancy (e.g., differences in type and year of currency, model structure, and modeling process), one main source is the fact that Sinha et al.'s health outcome measurement is based on the vaccine's effect on all-cause mortality (i.e., an absolute reduction of 7.4 deaths averted per 1000 vaccinated children) [8
] while our study's DALY measurement is based on the vaccine's effect on disease-specific morbidity and mortality (i.e., reduction in disease incidence, hospitalizations, outpatient visits, and deaths due to pneumococcal diseases). Indeed, when we transposed into our model the assumptions made about reduction in all-cause mortality by Sinha et al. and further standardized on some modeling processes (e.g., age-weighting in DALY calculation) and vaccine price, the incremental cost per DALY averted decreased from $670 to $40; this implies that analytic choices about endpoints and assumptions about key parameter values primarily account for the observed differences.
Although our analysis suggests PCVs would be considered a cost-effective investment over a plausible range of future vaccine prices, the base-case estimates of the incremental cost-effectiveness ratios ($570, $670, and $910 per DALY averted for PCV7, PCV10, and PCV13, respectively) are higher than those of other new vaccines that have been introduced in African countries. For example, the costs per DALY averted estimated for a routine hepatitis B vaccination program (with per dose vaccine price of $0.32) in The Gambia was $28 (in 2002 US$) [68
], and the corresponding value for the Hib program in Kenya (using a pentavalent diphtheria-tetanus-pertussis-hep B-Hib vaccine at a net cost of $2.46 per dose) was $38 (in 2004 US$) [69
]. One of the main drivers of these differences is vaccine price per dose or unit cost of an intervention. For example, when we used the same vaccine price as that for hepatitis B, $0.32, a PCV program's costs per DALY averted decreased to about $80-$140. To a lesser degree, differences in disease burden account for differences in outcomes.
Our study provides insights into the research priorities for accelerating control of pneumococcal disease burden using vaccines. First, our study highlights the importance of future clinical and microbiological studies on the natural history of S. pneumoniae
and etiology of diseases that are caused by the pathogen. It is well discussed that the burden of pneumococcal diseases is often underestimated even in large-scale clinical trials [9
] due to the challenges in isolating the pathogen. Also, little is specifically known about whether natural infection affords any significant protection against reinfection. Once more relevant evidence becomes available through future studies, the burden of pneumococcal diseases might be estimated more accurately. Second, our study also highlights the necessity of more in-depth research on the efficacy of PCVs. As noted previously, a comparison between our study and a previous study [24
] shows that the cost-effectiveness profile of a PCV can be highly sensitive to choice of endpoint against which vaccine efficacy is measured. Accordingly, while some hypotheses are available regarding the mechanism by which reduction in all-cause child mortality would be associated with PCV9 intervention [8
], more investigation about the impact of PCVs on the all-cause child mortality would be crucial. Third, for any given endpoint, the serotype distribution of S. pneumoniae
, vaccine serotype-specific efficacy, and possible cross protection against vaccine-related serotypes would be key determinants of PCVs' overall efficacy. As described previously, we observed a large discrepancy between the reported efficacy of PCV9 against, for example, all-serotype pneumococcal meningitis and sepsis (22%) and the theoretical efficacy (~57%), which is calculated as the product of serotype coverage of PCV9 and the vaccine efficacy against invasive diseases caused by vaccine serotypes. Additionally, while our use of surveillance data indicating no role of serotype 7F in local pneumococcal diseases [47
] resulted in identical model outcomes for PCV9 and PCV10, other local surveillance data have indicated some role by serogroup 7 (including serotype 7F, which is included in PCV10 and PCV13) [46
]. Given that vaccine efficacy is one of the most influential parameters, future clinical studies on this particular subject is warranted. Fourth, on a related note, it is crucial to monitor any trends in potential serotype replacement following large-scale introduction of a PCV into a local setting, since replacement disease would potentially diminish the long-term effectiveness of an immunization program. While our base-case analysis did not consider serotype replacement explicitly, our scenario analysis results show that even a moderate level (e.g., 25%) of serotype replacement could have a large impact on the cost-effectiveness of PCVs in The Gambia. Furthermore, routine immunization of Dutch infants using PCV7 has recently been found not to be cost-effective, due largely to the impact of serotype replacement [70
]. Recent surveillance data from different regions have reported an increased incidence of invasive pneumococcal diseases due to non-vaccine serotypes among young children, following the introduction of PCV7 [71
]. Though confounding factors may contribute to such findings, such findings still highlight the importance of further surveillance and research regarding the presence, magnitude, and speed of replacement by specific serotypes. Lastly, our findings from the scenario analyses suggest that an effective surveillance system should also carefully monitor actual take-up rates of PCVs and post-immunization incidence rates of pneumococcal diseases in order to capture potential indirect effects.
Although The Gambia would receive financial support for a PCV program from the GAVI Alliance according to the co-financing scheme, the country's healthcare expenditures (~$15 per capita per annum [19
]) must also accommodate the costs for the two new vaccines (Hepatitis B and Hib) already introduced as well as the six traditional EPI vaccines. Further, the country is also considering introducing another new vaccine, a meningococcal group A conjugate vaccine [74
]. Accordingly, in order to provide an optimized set of immunization services in the context of the entire health care infrastructure in The Gambia, it is crucial to assess affordability as well as cost-effectiveness (i.e., value for money) of a PCV program. Technically, affordability of a program can be defined according to whether a fixed (single or shared) budget for a specified period for the program can accommodate the total financial (not economic) costs required to implement the specific program [68
]. Thus, an accurate and relatively precise estimation of the total financial costs of introducing and sustaining a PCV program from the perspective of a program provider (e.g., local government) as well as an assessment of plausible budget levels would be essential steps for an affordability analysis. A few previous studies illustrate how information on both affordability and cost-effectiveness of a health program can be visually presented, under a single fixed budget [68
]. If a fixed budget designated exclusively for a PCV program is not available, an affordability analysis for PCV should take a more comprehensive approach, with consideration for other vaccination programs' affordability and shared budget constraints. Also, other complicating issues such as non-additivity of vaccination program costs in the presence of combination vaccines and uncertainty surrounding financial costs should be considered [68