Modeling the public health impact of malaria vaccines for developers and policymakers

Population Data on the population of each country (birth cohort, one-year olds, 0-4 year olds, 0-39 year olds, and the total population) were drawn from the UN Population Division [20], and are regularly updated with the latest data, thereby accounting for changes in demographic trends. Users may select the age range of the target population consistent with the expected vaccination strategies of potential malaria vaccines (Table 2).

Vaccine coverage The MVM included historical and projected vaccine coverage for Bacille Calmette-Guerin; first, second, and third doses of diphtheria-tetanus-pertussis (DTP1, DTP2, DTP3); and measles-containing vaccine for each country. Users may set the maximum vaccine coverage for each country equal to historical data for any of the above vaccines, or they may set maximum vaccine coverage at any level from 0–100%. Data on the 2005–2007 coverage rates were obtained from the WHO and United Nations Children’s Fund (UNICEF) [21]. Future coverage levels for this analysis were based on projections from WHO (Lara Wolfson, WHO ICE-T v4.0, Oct 2007, unpublished data).

Supply estimates are based on user inputs describing each manufacturer’s anticipated capacity, timing of expected increases in capacity, and year of vaccine availability, assuming development success and regulatory approval. Demand estimates in the form of total doses required per year are based on the size of the target population in each country, the maximum level of coverage and time taken to reach this level, and the number of doses in a regimen. Along with the above inputs and reference data, the model outputs include the year the vaccine is available for use in each country (from both supply and approval perspectives), the number of doses demanded, and the number of doses available to meet demand in any given year.

For the demonstration scenario, it was assumed that supply would not be constrained by manufacturing capacity and that all countries would require vaccine prequalification by WHO. Forty countries in sub-Saharan Africa that experienced a malaria disease burden in 2006 of 100 deaths per year or greater, or that experienced a malaria mortality rate of 10 deaths per 100,000 per year or greater, were included. The criteria were intended to allow inclusion of large countries with many cases at a relatively low rate, and small countries with few absolute cases but which have a significant rate relative to the population size. The time horizon modeled in the demonstration scenario was 10 years of vaccine use, beginning in 2016 (see Additional file 1).

Demand in the demonstration scenario (Figure 2) assumed a vaccine regimen of three doses delivered to a target population of infants and a wastage rate of 10% (see Additional file 1). No product preference or maximum acceptable price was modeled. Years between vaccine approval and country adoption of the modeled malaria vaccine, maximum coverage, and the time to reach maximum coverage were benchmarked from historical data, based on the uptake of Hib vaccine from 2001 to 2010 (either on its own, in the form of a tetravalent vaccine with DTP, or as a pentavalent vaccine with DTP and hepatitis B). WHO data on Hib coverage served as a reference point for the number of years post vaccine availability that each country began implementation, each country’s maximum coverage (DTP3 coverage was selected to represent maximum), and the time it took for Hib3 to match DTP3 levels [22]. Since malaria is more commonly recognized as a health threat than Hib, and because countries have gained experience in vaccine implementation since Hib introduction began, unadjusted data from Hib introduction might therefore underestimate the uptake of a malaria vaccine. The data were adjusted such that all countries that did not adopt Hib in the first two years of its availability were modeled as adopting a malaria vaccine two years earlier than they adopted Hib. The time each country took to reach maximum coverage was not adjusted. If countries did not introduce Hib by 2010, they were excluded from the demonstration scenario. The exception was Nigeria, which appears to be moving more quickly to adopt more recent vaccines than it did with Hib [23]. Nigeria was assumed to adopt the malaria vaccine midway through the time horizon analyzed, and reach maximum coverage in two years (see Additional file 2).

Disease burden Malaria-specific morbidity and mortality rates were obtained from the WHO for both the entire population and the population under 5 years of age [34] and are updated annually. In situations where no data specific to the under-5 population were available, the total population’s malaria morbidity and mortality rates were applied. Projections of future disease burden were based on each country’s current morbidity rate or current mortality rate multiplied by the country’s population forecast for the selected population. Options for changing the future disease burden, such as from implementation of other interventions, are described below. Estimates of vaccine impact on disease burden are based on the levels of transmission in each country, as described below.

Malaria transmission In the MVM, the intensity of transmission is represented by the entomological inoculation rate (EIR), defined as the product of the human biting rate of mosquitoes and the proportion of mosquitoes that are infectious. EIR is measured in the number of infective bites per person per annum (ibpa). Two sources of data on transmission were incorporated into the MVM, allowing the user to choose between them. Data from WHO were in the form of the percentage of each country’s population at high or low risk for malaria infection [34]. Alternatively, the Malaria Atlas Project (MAP) data were in the form of P. falciparum parasite prevalence [35]. The WHO source data were not provided in the form of EIR, which is the required input for transmission for the Swiss TPH model. The data were converted to EIR using methodology endorsed by both MAP and an external expert panel (see Additional file 3). Fortunately, much progress has been made recently in understanding the relationship between different measures of transmission [36]. The MVM calculates the percentage of each country’s population that falls into each of five EIR levels (0, 0.1, 1, 10, and 100 ibpa). Since transmission can vary multifold between settings and the means of measurement are imprecise [37], the EIR levels listed above were selected to clearly reflect this wide variation. The user is also free to input future changes in levels of transmission for countries, for example based upon the expected impact of other malaria interventions.

Estimates of the impact of vaccines on disease burden were based on the parameters of and simulations run in the previously published, dynamic model of the Swiss TPH, and integrated with country-specific transmission data in the MVM. Pre-erythrocytic vaccination was simulated as described in previous publications [33, 38], assuming that vaccination leads to both a reduction in the proportion of inoculations from the bites of infected mosquitoes and the resulting blood-stage infections. This leads to values of efficacy against infection higher than the proportion of clinical episodes prevented by such a vaccine [39]. Blood-stage vaccination was simulated by assuming that the vaccine reduced blood-stage parasite densities while transmission-blocking vaccines reduce the proportion of blood feeds that result in infection of the vector [33]. Duration of protection against infection is modeled as an exponential decay in vaccine efficacy as previously described [40], and the decay rate of efficacy against infection is the key variable that was chosen as a user input in the MVM. Decay rate of efficacy against infection describes the time after vaccination at which the vaccine protection against infection is half of its initial value. Vaccination scenarios were paired with non-vaccination comparator simulations to provide impact estimates of the outcomes found to be of greatest interest from the first version of the MVM: the numbers of cases (uncomplicated and severe), deaths, and disability-adjusted life years (DALYs) averted through vaccine use over time periods of interest.

MVI, Applied Strategies, and Swiss TPH agreed on possible combinations of input parameters anticipated to be of interest to vaccine developers and policymakers, resulting in over 100,000 scenarios to be simulated (Table 4). Each scenario was run multiple times, and mean frequencies of events were computed in order to reduce stochastic variability in the estimates of health effects. The results of these simulations are stored in look-up tables as reference data in the MVM, drawn upon as necessary according to the particular combinations of inputs for specific populations in the scenario chosen by the user (see Additional files 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 and 16).

Estimates of the public health impact of a vaccine are generated based on the type of vaccine, initial efficacy and decay rate of vaccine efficacy against infection, mode of vaccine delivery, and transmission setting. The model outputs from this module are the total number of events averted (uncomplicated cases, severe cases, deaths, and DALYs) in the total population or those less than 5 years of age through use of the vaccine.

For the demonstration scenario, although none of the current vaccine candidates in trials is expected to meet this target, a pre-erythrocytic vaccine with an efficacy of 85% against uncomplicated malaria cases was assumed in light of the strategic goal for malaria vaccine development set in 2006 [19]. The decay rate of efficacy against infection was four years. These values of efficacy and its decay were selected for the demonstration scenario as they were expected to be broadly consistent with targets set by the international community [19]. Routine infant immunization (via the Expanded Program on Immunization) was selected as the mode of delivery. As described above, the user enters the projected transmission in each country in the form of EIR into the MVM. For the demonstration scenario, three different projections were created to reflect decreases in transmission from current levels, potentially due to implementation of existing interventions. The projection applied to the demonstration scenario used the MAP-derived EIR distribution across countries, shifting one-quarter of the population at each EIR level (0, 0.1, 1, 10, and 100 ibpa) into the next lower category (e.g., from EIR of 10 down to 1). The other two projections shifted either one-half or three-quarters of the population at each level to the next lower level of EIR. It was assumed that the percentage of the population residing in each EIR level decreased in a linear fashion to the next lowest level over a period of five years, and remained constant for the remaining five years modeled.

This module does not include pre-entered reference data.

Based on the user inputs of vaccine price and implementation costs, the module generates the total investment required to achieve the public health gains estimated in the public health impact module, and calculates the cost per case, severe case, death, and DALY averted. Depending on the assumed financing scenario, for example, if donors support a portion of the purchase and implementation costs, the model provides a break-down by donor and country contributions for each year of vaccine use. A discount rate can be used to calculate the present values of the total investments, and can be summed to estimate the net present value.

The demonstration scenario assumed a vaccine price remaining constant at $5/dose, including insurance and delivery to the airport of a country as specified by the consignee. Inflation was not included. This price is broadly consistent with the assumed cost of other new vaccines for low-income countries, such as pneumococcal conjugate and human papilloma virus vaccines (as there is no price yet determined for any potential malaria vaccine). In addition to the vaccine price, the implementation cost modeled was $0.33/dose and injection equipment costs were $0.07/dose, totaling $1.20/fully immunized child. Costs were based upon a simulation study of the cost of introducing a malaria vaccine into the routine immunization system in Tanzania [41], and were also consistent with results from a similar study following the introduction of a pentavalent DTP-hepatitis B-Hib vaccine in Ethiopia [42]. Results from the demonstration scenario are presented without discounting and with a discounting of up to 5%, to cover a range around the most commonly used rate of 3% in the literature [43].

Table 6 summarizes the assumed values and vaccine characteristics for the demonstration scenario that were described throughout the methods section. Assumptions related to cold chain and personnel were not separately costed for the demonstration scenario, but instead based upon published analyses which built upon experience with existing vaccines (e.g. DTP-HepB-Hib) [42, 44].