Modelling the cost-effectiveness of introducing the RTS,S malaria vaccine relative to scaling up other malaria interventions in sub-Saharan Africa

Introduction

Following the adoption of the Millennium Development Goals (MDGs) in 2000, the concurrent targets of increasing the proportion of children under 5 years that sleep under a bed net and that have access to appropriate antimalarial drugs1 have been associated with an estimated 37% decline in malaria case incidence and 60% decline in malaria mortality.2 ,3 These gains have been achieved with substantial investment in malaria which has more than doubled over the last decade, from less than $1 billion in 2005 to over $2.5 billion in 2015.2 However, since 2011, the acceleration in funding has slowed, plateauing in recent years.2 Funding gaps therefore remain an inevitable issue for future control and elimination efforts and thus optimising the use of available resources is paramount. Within this context, new interventions must be evaluated not only on their direct cost-effectiveness, but comparative to the other intervention options.

In early 2015, the final results from the large phase III trial of the RTS,S malaria vaccine across 11 sites in Africa were published.4 Reported efficacy against clinical and severe malaria disease in the presence of high use of bed nets for prevention and alongside a high level of access to care was moderate. The trial reported 36.3% (95% CI 31.8% to 40.5%) efficacy against clinical disease in children aged 5–17 months over 4 years under a four-dose schedule and 32.2% (13.7% to 46.9%) efficacy against severe disease.4 Nevertheless, in the high transmission sites contributing most of the disease episodes, there was a significant public health impact, with between 1000 and 6000 cases estimated per 1000 population over 4 years of follow-up in the six higher transmission sites.4 To estimate the wider public health impact and cost-effectiveness of the vaccine in settings representative of current levels of ongoing malaria transmission, a WHO working group was formed to compare the outputs from four mathematical models parameterised using the trial data. The results from this comparison demonstrated that the vaccine could have a substantial public health impact across settings ranging from 10% to 60% parasite prevalence in 2–10 year olds.5 Furthermore, assuming a midrange cost of $5 per dose under a four-dose schedule, the vaccine was considered to be highly cost-effective (incremental cost-effectiveness ratio (ICER) of $44–$279 per disability-adjusted life year (DALY)) within these same transmission levels.5

Existing malaria interventions are also highly cost-effective.6 Although there are substantial differences between the methodologies used to make these estimates, a 2011 study estimated the median ICER per DALY at $27 (range $8.15–110) for long-lasting insecticide-treated nets (LLINs) and $143 (range $135–150) for indoor residual spraying (IRS), while for seasonal malaria chemoprevention (SMC), the cost per case averted has been estimated at $68 (95% CI $62 to $75), similar to comparable metrics for LLINs and IRS.7 Thus, in areas in which coverage of these interventions is not yet universal, it is important to understand the relative cost-effectiveness of the full suite of interventions and where the RTS,S malaria vaccine could contribute. Importantly, this needs to take into account the diminishing marginal returns associated with the scale-up of interventions that may lead to a higher unit cost at high levels of coverage.8 ,9

Here, we use a well-established transmission model for Plasmodium falciparum malaria and its associated interventions10 to estimate the cost and impact of different intervention packages at varying levels of scale-up. We evaluate these packages over a wide range of transmission settings and use the estimates to derive the most cost-effective pathways for scaling-up malaria interventions in order to inform decisions about the introduction of the RTS,S malaria vaccine.

Methods

Estimating cost-effective scale-up

For each transmission strata, the cost-effective scale-up of interventions was estimated using the following steps: (i) start with 0% usage/coverage of all interventions, (ii) for each available intervention, calculate the ICER of scaling up to the next usage/coverage level, (iii) implement the step with the lowest associated ICER (the most cost-effective), (iv) repeat the process (i–iii) until the scale-up of all interventions has been maximised or elimination is achieved. This process is summarised in figure 1. It should be noted that while this approach results in the most cost-effective next level of intervention coverage, the resulting intervention packages are not necessarily maximally efficient for a given budget since this reflects a gradient descent optimisation (stepwise) rather than a multidimensional optimisation approach. Nevertheless, it is used here to illustrate the pathways at each point in the absence of defined budget limits.

Request permission

Figure 1

Schematic of the cost-effective scale-up pathway. For each transmission strata, the cost-effective scale-up of interventions was estimated, starting with no intervention coverage then scaling up coverage based on the most favourable incremental cost-effectiveness ratio. Scale-up ceased when all interventions were at full coverage or elimination had occurred.

Results

Figure 2 shows the scale-up pathways for the combination of the four interventions across a range of transmission levels characterised by their baseline PfPR_2–10 (in the absence of interventions other than treatment of clinical cases). For the majority of settings, across the full-range of baseline PfPR_2–10, for vector bionomics characteristic of the three main species found in Africa (A. gambiae s.s., A. arabiensis and A. funestus) and for seasonal and non-seasonal settings, LLINs appear first in the most cost-effective pathway. At baseline PfPR_2–10 <5%, LLINs and RTS,S are predicted to reduce ongoing transmission to pre-elimination levels. If scale-up of these interventions is unable to achieve pre-elimination transmission levels, our results suggest that LLINs should be scaled up to very high usage levels (75%) prior to introducing additional interventions if cost-effectiveness is the single deciding factor. After this level has been achieved, in non-seasonal settings, IRS is the second-most cost-effective intervention at lower baseline PfPR_2–10 (5%< PfPR_2–10 <65%), while RTS,S is estimated to be the second most cost-effective intervention in settings with baseline PfPR_2–10 ≥65%. In seasonal settings, SMC is generally the second most cost-effective intervention to introduce prior to IRS or RTS,S.

Request permission

The results shown in figure 2 assume a single unit cost that does not change with increasing coverage. However, as higher coverage levels are sought, costs tend to increase as it becomes more difficult to fill the remaining coverage gaps and to access the hardest-to-reach populations. Figure 3 shows our estimated empirical production functions for IRS, SMC and vaccination (based on DTP3 data) alongside the previously published estimated for LLIN usage.27 All four functions have a similar shape, with increasing costs at high coverage. However, for IRS and SMC, the recorded coverage was consistently high (>80%) and the limited number of data points mean that this function is uncertain.

Request permission

Including the non-linear production functions shown in figure 3 leads to a substantially more complicated pattern of scale-up pathways (figure 4). With these additional non-linearities, alternative interventions are always introduced before LLIN usage is increased to the maximum level due to the high estimated cost of achieving high LLIN usage. In non-seasonal settings, LLINs are estimated to be the most cost-effective initial interventions. Across the majority of settings with baseline 5%<PfPR_2–10<60%, IRS appears second with RTS,S also introduced once moderate levels of IRS coverage have been achieved. In seasonal settings, for baseline PfPR_2–10<50%, LLINs remain the first most cost-effective intervention. However, in settings with higher baseline transmission, SMC and RTS,S appear earlier. Following this, the second and third most cost-effective additional interventions vary by setting, with IRS being favoured in locations where the vector bionomics are more amenable to insecticidal control, whereas RTS,S or SMC are favoured in settings with less favourable vector bionomics.

Request permission

Figure 5 illustrates the translation of these generic results to locations in sub-Saharan Africa using estimates of vector species presence, baseline PfPR_2–10 and seasonality. Across the majority of settings, LLINs are the first most cost-effective intervention. Using the results from figure 4, the LLIN usage at which a second intervention is estimated to be more cost-effective than further LLIN scale-up is shown in figure 4A. In the majority of settings, we estimate a switch is cost-effective at 55–65% LLIN usage, although in some seasonal areas in West Africa and in areas with high estimated baseline PfPR_2–10, other interventions are estimated to be more cost-effective at lower levels of LLIN usage. This threshold is similar to the levels of usage that have now been achieved in many parts of Africa (figure 5B).

Request permission

While the results vary for different assumed unit costs for each intervention, in a sensitivity analysis of these costs, we found that the order of scale-up is generally maintained (figure 6A). Since the price of the RTS,S vaccine has not been released, we additionally assessed the sensitivity to the assumed price per dose. Figure 6B shows the proportion of scenarios in which RTS,S, at a given price per dose, is estimated to occur before other interventions in the scale-up pathway. In general, RTS,S remains late in the pathway. However, this pattern changes if the price per dose drops below US$3 where the relative cost-effectiveness becomes comparable to IRS and SMC. However, even at a very low cost per dose (<US$1.00), LLINs are estimated to remain a more cost-effective intervention than RTS,S in approximately half of the settings.

Request permission

Discussion

Our analysis demonstrates that LLINs remain the most cost-effective first intervention to reduce malaria transmission across the broad range of transmission settings observed in sub-Saharan Africa. The high consumption and competitive market for LLINs has driven costs down over the last 10 years.21 This, coupled with their dual effect of personal-level and community-level protection,28 makes them highly cost-effective. This finding was consistent for all outcome measures considered (see online supplementary materials S1–S4). Furthermore, our results indicate that, based on cost-effectiveness considerations, the RTS,S vaccine should be considered a secondary intervention alongside the two other WHO-recommended malaria interventions for this region—SMC and IRS. The recommended schedule for the RTS,S vaccine that will be tested in pilot implementation is for four doses given in children aged 5–27 months, who constitute a small subset of the exposed population. The vaccine offers partial protection to this group over a duration of ∼4 years.4 ,14 As a result, this vaccine does not have the benefit of inducing herd-immunity in the population and is considerably more expensive per person than the other interventions considered here (at the assumed cost of $5 per dose), lowering its relative cost-effectiveness. Thus, other than in high transmission settings where there is a high burden of disease in young children, we find that RTS,S enters the cost-effectiveness scale-up pathway later on when other potential options for reducing transmission and/or protecting from disease are already maximised.

While increasing usage of LLINs is identified as the most cost-effective first intervention, once these levels reach 50–60%, we estimate that the three alternative interventions—IRS, SMC and RTS,S—become increasingly competitive when comparing the relative cost-effectiveness. This level of LLIN usage is similar to the levels reported for many countries in sub-Saharan Africa in 2015,3 suggesting that context-specific cost-effectiveness considerations may become increasingly important as investment in current or new interventions are considered. The inclusion of IRS as a secondary vector control option could further reduce onward transmission through providing additional protection for those that do not consistently use nets or over the periods between net distribution rounds in which the integrity of the net and/or efficacy of the insecticide has decayed. However, the trial data on the combination of these two vector control interventions remain inconclusive29 ,30 and hence close monitoring would be required to fully understand the operational impact of their combined use. Furthermore, any recommendations in favour of vector control must be made in the light of current and potential future insecticide resistance, the effects of which were not included in this analysis. In a small number of settings, characterised by a high level of seasonal transmission and intense transmission, we identified either SMC or IRS as the most cost-effective first intervention. SMC and IRS are temporally targeted at the peak transmission season, and this therefore increases their cost-effectiveness relative to non-seasonally targeted interventions.31 A comparison of the impact of pairwise combinations of interventions is included in online supplementary material S5.

We explicitly excluded treatment scale-up options from the cost-effectiveness analysis for several reasons. First, treatment has been previously shown to be highly cost-effective.6 Therefore, equitable access to treatment for severe disease is an ethical priority and universal scale-up of treatment coverage is important to preserve. Second, while increasing treatment coverage, and therefore costs, affects the absolute cost-effectiveness of other interventions (with lower treatment coverage making them more cost-effective), the relative cost-effectiveness when comparing interventions (and therefore scale-up pathways) will be influenced far less. Third, the ability of a country to increase coverage of treatment will depend critically on health system capacity and hence vary geographically. Thus, a simple unit cost approach is unlikely to be appropriate. For other interventions, the commodities (particularly LLINs) are purchased through Global Fund pathways for which there is a coordinated tendering process. We have therefore adopted a unit costing approach for the interventions considered in this broad-scale comparison.

Systems-level inefficiencies,32 overallocation27 and systematic under-representation in hard-to-reach populations33 ,34 may all contribute to diminishing marginal returns when investing in increasing the coverage of an intervention to very high levels. Our results demonstrate that it is important to capture these non-linearities when considering the relative costs-effectiveness of introducing new interventions such as the RTS,S malaria vaccine. With the simple assumption that costs associated with increasing the usage or coverage of an intervention increases linearly, we observe a very clear picture of cost-effective scale-up, with LLIN usage increased to the maximum level (75%) in nearly all settings before any other intervention is implemented. However, the inclusion of non-linear productions functions that capture the increasing cost associated with achieving high levels of coverage of any given intervention leads to a more complicated picture of the cost-effective scale-up pathway. While there was considerable uncertainty in our estimated production functions for each intervention, general patterns in scale-up remained fairly robust to this. However, further data on these patterns are critical to inform local planning. Subtle differences in the inflection points of the production functions affect when a switch between interventions is made. This is especially of note for the SMC production function where lack of data lead to considerable uncertainties in the resultant production function.

While cost-based uncertainties were explored in this analysis, we did not additionally explore the uncertainty in model structure or parameterisation due to the computational complexity in undertaking such an analysis. Clearly, model parameters and structures could affect the relative impact of interventions as well as the combinations of interventions needed to reach pre-elimination levels. Determining the cost-effective scale-up in a stepwise manner informed by the ICER always chooses the next most cost-effective option. This is analogous to a gradual scale-up of interventions over time, where future options are considered, given an established intervention landscape. However, for a given spend, the resulting intervention package estimated from a stepwise approach may differ from a global optima if all scale-up options were considered in unison.

Context-specific challenges to scaling-up a given intervention will always be present and cannot be represented in this style of analysis. To this end, policy decisions must take into account such challenges when considering recommendations in a specific setting. While a number of countries have achieved the high levels of LLIN usage that our analysis suggests to be cost-effective,3 in other settings, there may be barriers to achieving and sustaining the levels in the long term, nuancing the decision to target LLIN scale-up. Nevertheless, the ambitious targets set by the WHO for universal coverage may be integral in driving trends in net usage upwards, even if the target cannot be reached.