The authors have declared that no competing interests exist.
Conceived and designed the experiments: EMS NC KG. Performed the experiments: EMS JS AB. Analyzed the data: EMS KG NC. Wrote the paper: EMS KG NC JC CD JS. Conceived and designed the modeling study: TS JC. Designed the field studies: TB AYB SK. Supervised and conducted the intervention field trial: JS AYB WO. Responsible for data management: JS AYB WO.
Tools that allow for in silico optimization of available malaria control strategies can assist the decision-making process for prioritizing interventions. The OpenMalaria stochastic simulation modeling platform can be applied to simulate the impact of interventions singly and in combination as implemented in Rachuonyo South District, western Kenya, to support this goal.
Combinations of malaria interventions were simulated using a previously-published, validated model of malaria epidemiology and control in the study area. An economic model of the costs of case management and malaria control interventions in Kenya was applied to simulation results and cost-effectiveness of each intervention combination compared to the corresponding simulated outputs of a scenario without interventions. Uncertainty was evaluated by varying health system and intervention delivery parameters.
The intervention strategy with the greatest simulated health impact employed long lasting insecticide treated net (LLIN) use by 80% of the population, 90% of households covered by indoor residual spraying (IRS) with deployment starting in April, and intermittent screen and treat (IST) of school children using Artemether lumefantrine (AL) with 80% coverage twice per term. However, the current malaria control strategy in the study area including LLIN use of 56% and IRS coverage of 70% was the most cost effective at reducing disability-adjusted life years (DALYs) over a five year period.
All the simulated intervention combinations can be considered cost effective in the context of available resources for health in Kenya. Increasing coverage of vector control interventions has a larger simulated impact compared to adding IST to the current implementation strategy, suggesting that transmission in the study area is not at a level to warrant replacing vector control to a school-based screen and treat program. These results have the potential to assist malaria control program managers in the study area in adding new or changing implementation of current interventions.
Important progress has been made in the past decade in reducing malaria morbidity and mortality in Kenya, but it is not obvious which additional tools and strategies should be the next priority to include in the package of malaria control interventions in a given area to keep transmission levels low, especially given the threat of resistance of the parasite and vectors to antimalarial drugs and insecticides
OpenMalaria, a stochastic simulation modeling platform
Rachuonyo South District in Homa Bay County of Nyanza Province, Kenya is a highland fringe area with altitude between 1,400 and 1,600 meters. Ethnicity is predominantly Luo and homesteads are distributed broadly across a rolling landscape intersected with small streams and rivers. The area is characterized by generally low malaria endemicity with marked seasonal and inter-annual variations in transmission
The main malaria control methods are currently mass-distribution of LLINs, annual indoor residual spraying (IRS) with pyrethroids, and prompt and effective treatment
Rachuonyo South is one of a number of field sites of the Malaria Transmission Consortium (MTC), a project with the goal of enabling operational program managers to achieve optimal implementation of transmission-reducing malaria control techniques. Active between 2009 and 2012, MTC surveys provided detailed entomological studies of species composition and biting behavior
The study proposal received ethics approval from the Ethical Review Committee (ERC) of the Kenya Medical Research Institute (KEMRI) Nairobi under proposal number SSC 2163, the London School of Hygiene & Tropical Medicine ethics committee (#6111), and from Centers for Disease Control and Prevention (with exempt status).
A team at the Swiss Tropical and Public Health Institute (Swiss TPH) and Liverpool School of Tropical Medicine (LSTM) developed the OpenMalaria platform comprising stochastic simulation models of transmission of malaria based on the simulation of infection in individuals. These models are able to evaluate the impact (cost-effectiveness, clinical, epidemiological and entomological) of numerous intervention strategies for malaria control
The scenario describing the current intervention mix was parameterized using a previously-published model of malaria epidemiology and control in Rachuonyo South District, validated with observed data from the site-specific MTC studies described above
Combinations of interventions for the experiment were chosen in collaboration with malaria control personnel in the study area to correspond to a 2011–2012 intervention evaluation trial
LLIN use (%) | IRS coverage (%) | IRS deployment month | School-based IST coverage (%) | IST frequency (per school term) | Fevers receiving an antimalarial (%) | |
56 | 70 | Alternating April/June | 28 | |||
28 | ||||||
80 | 90 | Alternating April/June | 28 | |||
56 | 70 | April | 80 | 2 | 28 | |
80 | 90 | April | 80 | 2 | 28 | |
56 | 70 | April | 28 | |||
56 | 70 | May | 28 | |||
56 | 70 | June | 28 | |||
56 | 90 | April | 28 | |||
56 | 90 | May | 28 | |||
56 | 90 | June | 28 | |||
70 | Alternating April/June | 28 | ||||
90 | Alternating April/June | 28 | ||||
56 | 28 | |||||
80 | 28 | |||||
40 | 1 | 28 | ||||
40 | 2 | 28 | ||||
80 | 1 | 28 | ||||
80 | 2 | 28 |
*Represents the base case scenario as parameterized in Stuckey et al. 2012
Each intervention strategy was simulated in a population of 100,000 individuals. To simulate the status quo prior to interventions, simulations were run for one human life span to induce an “equilibrium” level of immunity. Forward simulations of each intervention combination were made using an ensemble of 14 model variants for malaria in humans to address model uncertainty
Malaria case management costs were based on a societal perspective; direct costs to the health systems are considered, as well as direct expenditures associated with malaria episodes at the household level. Indirect costs, including productivity loss due to illness, were not accounted for. While the latter tend to dominate the economic cost of illness
Treatment costs are evaluated following a model of malaria case management developed for endemic settings and is described elsewhere
On the provider side, cost per episode covers drugs, diagnosis, medical personnel, facility charges, and other consumables. In addition to the first-line antimalarial as per national malaria guidelines, a portion of uncomplicated cases were assigned to treatment with sulfadoxine pyrimethamine (SP) given evidence on moderate uptake of AL, the first line artimisinin combination therapy (ACT) in the study area
Direct patient costs associated with a malaria episode include travel expenses to and from healthcare facility and other consumables (i.e. water, food, etc) and were based on the multi-country literature review. Spending on consumables is generally considered negligible; only a few studies recorded these data with an average of $0.20 per visit
Both average and marginal health system costs were calculated for each outcome. The average cost includes all costs involved in delivering a health intervention, including the use of spare capacity or slack in the system, health care resources diverted from other uses, and existing health sector resources shared with other health programs. In the marginal analysis only costs of drugs, diagnosis, and patient spending per visit were considered, as broader savings to the health system including labour and capital costs would not be immediately affected by changes in consumption of medical services due to lower diseases burden achieved by control interventions
A sensitivity analysis was conducted for the costs of test and cost per ACT dose by varying costs −50%/+100%, and for proportion of fevers that access medical care by varying access −/+50% (
Sensitivity analysis | ||||
Parameter | Unit | Value per unit | Lower value | Upper value |
Paracheck® rapid diagnostic test | $0.62 |
$0.31 | $1.24 | |
Coartem® (Artemether-lumefantrine) | $0.0898 |
$0.045 | $0.1769 | |
Proportion of the most recent episode of fevers in children under five within 2 week recall seeking medical care | 0.6183 |
0.309 | 0.927 |
All costs are in 2012 USD.
A general approach for costing malaria interventions using secondary data was applied as outlined by Kolaczinski et al
Sensitivity analysis | ||||||
Intervention | Unit | Distribution method | Economic cost per unit | Marginal economic cost per unit | Lower value | Upper value |
Net delivered | Mass campaign through community organizations |
$8.52 | $8.37 | $4.26 | $17.04 | |
Person protected | Annual mass campaign |
$0.73 | $0.34 | $0.34 | $1.46 | |
Child screened | School-based distribution |
$6.32 | $2.89 | $3.16 | $12.63 |
All costs are in 2012 USD.
The simulated effectiveness of malaria control interventions and intervention combinations was evaluated by calculating the mean and inter-quartile range (IQR) of all model variants and seeds for each intervention combination for the difference in disease burden over a five year period from the start of intervention deployment compared to the mean of the simulations of the base case scenario with no interventions other than the existing case management system. Outcomes evaluated include decrease in parasite prevalence, number of uncomplicated episodes, hospitalizations and deaths averted in the general population. In addition to indicators for severity of illness, the overall population burden averted in terms of disability adjusted life years (DALYs) is calculated by combining mortality and morbidity measures as described by Murray and Lopez
Estimates of effectiveness of control interventions and intervention mixes are combined with the added costs of implementing these control measures. Treatment cost savings, or the reduction in cost to the health system due to the reduction in cases seen by the system, achieved by implementing the control strategy, are used to offset implementation costs and thus cost effectiveness ratios are calculated based on net rather than total intervention costs.
The cost savings to the case management system and households (CM) associated with implementing each intervention combination (IC) instead of a scenario without interventions (NO) are computed as DCcmNO−DCcmIC, where DCcmNO are the direct costs (DC) of case management in the scenario without interventions and DCcmIC are the direct costs of case management in the case of each intervention combination. These cost savings are subtracted from the direct cost of implementing each intervention combination (DCint) to give a net intervention combination cost (NC) computed as follows: NC = DCint−(DCcmNO−DCcmIC). Cost effectiveness is evaluated in two ways. The first is by calculating the average cost effectiveness ratio (ACER), as the net cost (NC) of the intervention divided by the net effects (NE) of the intervention. The second is by calculating the incremental cost effectiveness ratio (ICER), which follows the same methodology for calculating the ACER, except the net costs and net effects of each intervention combination are calculated against the currently implemented strategy.
Both marginal and average cost-effectiveness ratios over a five year reference period are reported to illustrate the likely short-term financial impact of the intervention, as well as the longer-term impact associated with the intervention including structural changes in health care delivery in response to lower disease burden achieved by the program. Cost effectiveness ratios are reported without discounting of future costs and benefits due to the short implementation time frame of the study and the recommendation from the revised GBD study
Compared to an intervention scenario with no malaria control outside of routine case management, and after five years of implementation, the intervention combination with LLIN use by 80% of the population, 90% of households covered by IRS with deployment starting in April, and IST of school children using AL with 80% coverage twice per term result in the largest simulated reduction in all-age parasite prevalence (99%, IQR 99.1–99.3%), average averted cases of uncomplicated malaria per person (7.46, IQR 7.44–7.48), hospitalizations averted (thousands)(3.96, IQR 3.95, 3.98), deaths averted (1,541, IQR 1,535, 1,551), and DALYs averted (thousands) (77.6, IQR 77.3–78.2) (
Proportion reduction in all-age parasite prevalence, year 5 | Uncomplicated episodes averted per person | Hospitalizations averted (thousands) | Deaths averted | DALYs averted (thousands) | ||||||
Mean | IQR | Mean | IQR | Mean | IQR | Mean | IQR | Mean | IQR | |
LLIN 56%+IRS 70% | 0.96 | (0.95, 0.96) | 7.04 | (6.97, 7.10) | 3.78 | (3.74, 3.83) | 1.42 | (1.40, 1.44) | 71.48 | (70.77, 2.37) |
LLIN 80%+IRS 90% | ||||||||||
LLIN 56%+IRS 70%+IST 80% twice per term | ||||||||||
LLIN 80%+IRS 90%+IST 80% twice per term | ||||||||||
LLIN 56%+IRS 70% April start | ||||||||||
LLIN 56%+IRS 70% May start | 0.95 | (0.95, 0.96) | 6.98 | (6.90, 7.05) | 3.75 | (3.71, 3.82) | 1.40 | (1.39, 1.42) | 70.48 | (69.81, 1.50) |
LLIN 56%+IRS 70% June start | 0.96 | (0.96, 0.97) | ||||||||
LLIN 56%+IRS 90% April start | 0.76 | (0.73, 0.78) | 6.12 | (5.85, 6.25) | 3.26 | (3.06, 3.45) | 1.21 | (1.14, 1.27) | 61.31 | (58.17, 4.18) |
LLIN 56%+IRS 90% May start | ||||||||||
LLIN 56%+IRS 90% June start | ||||||||||
IRS 70% | 0.53 | (0.50, 0.55) | 2.89 | (2.06, 3.33) | 1.66 | (1.35, 1.97) | 0.60 | (0.49, 0.74) | 30.33 | (25.56, 7.05) |
IRS 90% | 0.66 | (0.63, 0.67) | 3.63 | (2.95, 3.99) | 2.10 | (1.86, 2.37) | 0.74 | (0.66, 0.85) | 37.62 | (33.56, 2.63) |
LLIN 56% | 0.95 | (0.95, 0.96) | 7.00 | (6.93, 7.05) | 3.76 | (3.72, 3.82) | 1.41 | (1.39, 1.43) | 70.86 | (70.11, 1.95) |
LLIN 80% | 0.94 | (0.93, 0.94) | ||||||||
IST 40% once per term | 0.09 | (0.05, 0.16) | 0.28 | (−0.83, 1.05) | 0.16 | (−0.17, 0.63) | 0.07 | (−0.04, 0.26) | 3.40 | (−2.12, 12.75) |
IST 40% twice per term | 0.14 | (0.11, 0.20) | 0.46 | (−0.66, 1.20) | 0.25 | (−0.08, 0.68) | 0.10 | (−0.001, 0.32) | 5.21 | (−0.33, 14.80) |
IST 80% once per term | 0.16 | (0.13, 0.21) | 0.53 | (−0.59, 1.27) | 0.29 | (−0.05, 0.74) | 0.12 | (−0.01, 0.35) | 5.93 | (−0.84, 16.08) |
IST 80% twice per term | 0.22 | (0.19, 0.27) | 0.78 | (−0.27, 1.46) | 0.42 | (0.08, 0.82) | 0.16 | (0.05, 0.36) | 8.14 | (2.34, 18.06) |
Compared to a scenario with no interventions outside the existing case management system, the mean and inter-quartile range of the impact of different intervention combinations (
Simulation results indicate that increased coverage of vector control has a larger impact than adding an IST intervention to the current control strategy. However, adding the highest IST coverage and frequency to the current strategy could reduce parasite prevalence by an additional nine percentage points (
White lines represent the simulated median value, blue boxes represent the inter-quartile range, and capped bars represent the upper and lower adjacent values for simulated results for each intervention combination using an ensemble of 14 model variants and five random seeds. Choice of intervention combinations is based on the criteria of simulated reduction in parasite prevalence greater than the strategy currently implemented in the study area.
Despite moderate levels of self-reported LLIN use, simulations indicate LLINs, and not IRS, account for the majority of impact on parasite prevalence. Removing LLINs and continuing only with a higher level of IRS coverage resulted in a similar number of averted uncomplicated cases compared to the IST interventions (
Simulated cumulative DALYs averted after five years compared to the no intervention scenario by net program costs for different implementation strategies of
Depending on coverage level and frequency, without vector control interventions, simulations suggest IST could reduce annual average parasite prevalence in the population by 9–22% (
Total delivery costs and net health system costs for implementing each intervention combination can be found in
Tornado diagram of the change in the ACER of an intervention with 80% LLIN use, 90% IRS coverage, and 80% IST coverage twice per term in relation to variation in component costs.
Five intervention combinations simulated more averted DALYs than the currently-implemented intervention combination (
Simulated cumulative DALYs averted in a population of 100,000 individuals after five years compared to the no intervention scenario by net program costs for the intervention combinations with a better simulated health outcome than the currently implemented malaria control strategy, ranked in descending order of ACER. Black dots represent the mean simulation results across 14 model variants and five seeds. Circles represent the of simulated DALYs averted by net program costs with different assumptions of input costs of the case management system and malaria control interventions in the study area represented in
Average ACER | Marginal ACER | Average ICER | Marginal ICER | |||
Current strategy: LLIN 55%, IRS 70% | 4.29 | (4.22, 4.33) | 6.30 | (6.28, 6.31) | ||
LLIN 55%, IRS 90% May start | 5.27 | (5.21, 5.31) | 6.75 | (6.74, 6.75) | 235.46 | 111.58 |
LLIN 55%, IRS 90% June start | 5.11 | (5.06, 5.13) | 6.62 | (6.61, 6.62) | 50.24 | 24.27 |
Add IST to the current strategy | 6.13 | (6.09, 6.14) | 7.02 | (7.01, 7.03) | 66.03 | 30.55 |
LLIN 80%, IRS 90% | 7.39 | (7.38, 7.40) | 8.92 | (8.92, 8.92) | 53.75 | 48.06 |
LLIN 80%, IRS 90%, IST 80% twice per term | 9.06 | (9.04, 9.05) | 9.59 | (9.58, 9.60) | 65.05 | 48.27 |
The mean and inter-quartile range of the average cost effectiveness ratios (ACER) compared to a scenario with no interventions outside the existing case management system, and incremental cost effectiveness ratios (ICER) compared to the currently implemented strategy for different intervention combinations with more simulated DALYs averted than the currently implemented strategy. ACERs and ICERs are calculated using costs reported in
Cost effectiveness analyses based on health outcomes simulated by transmission models can compare many more intervention effects than can static models or field trials. In these simulations, interventions simulate a decrease in vector population and a corresponding decrease in transmission that allows for mass community effects of interventions. In particular, such models can explore the effects of intervention scenarios by transmission level and coverage level whereas in single field studies all the effects of different interventions cannot be captured.
Increased coverage and use of vector control interventions has a larger simulated impact on all malaria indicators than adding IST to the currently implemented control strategy. There could be additional impact of IST programs not captured in this analysis, including improved school performance and decreased anemia
Despite moderate observed use in the population, simulations show LLINs and not IRS account for the majority of impact on disease burden. Changing the timing of IRS implementation did not have a large impact on parasite prevalence. This could be due to the simulation experiment design, which models implementation of IRS programs rolled out over a 60 day period culminating in the target proportion of individuals protected. Because the start date of implementation was varied by 30 days at a time, implementation could overlap enough to prevent observing a substantial difference between scenarios. Rather than changing the timing or coverage of IRS, the study area may benefit from adding new vector control interventions, particularly those targeting exophagic and exophilic vectors.
The simulation results for the effect of the currently-implemented strategy on parasite prevalence in the study area have been previously validated and found to be in the range of the effects observed in the field
While interventions were chosen to correspond to those in the hotspot-targeted intervention study, simulated implementation was assumed for the whole population rather than target hot spots because OpenMalaria does not incorporate an explicit spatial element. Therefore results cannot be matched against intervention trial results for validation purposes. However, findings from this experiment can help put the trial results in the broader context of what could be expected from community-wide implementation of combinations of interventions.
While simulations of the scenarios describing the effects of the intervention combinations in reducing malaria burden account for uncertainty by employing an ensemble of 14 model variants and multiple random seeds, uncertainty in the costing model is limited to a one-way sensitivity analysis. A probabilistic sensitivity analysis exposing the model to changes in assumptions of inputs to the case management and intervention unit costs is being conducted for publication elsewhere, and will assist in clarifying the uncertainty inherent in these predictions.
Despite vector behavior in the study area favoring outdoor biting, IRS had a lower health impact than expected when simulated as a stand-alone intervention when compared to LLINs. The IRS model parameterization has deterrency and killing effects of half that of LLINs, due to simulated action only on post-prandial indoor resting mosquitoes, in contrast to the both pre- and post- prandial killing effect of LLINs. A model update will allow the effect of IRS to be simulated on both states of the mosquito feeding cycle, and the parameters for effectiveness of IRS should be updated based on experimental hut data. It is also worth noting the lower cost per sachet of insecticide assumed in the costing model compared to the average unit costs reported in the recently released UNITAID report on malaria vector control commodities
Results of this experiment have the potential to assist malaria control program managers in the study area in deciding on adding new or changing the implementation of current interventions. All the simulated intervention combinations can be considered cost effective in the context of levels of health expenditure in Kenya. Malaria is the number six contributor to the burden of disease in Kenya, both overall and in children under five
Findings from this study indicate there are several combinations of interventions that could result in a greater health impact per dollar spent than the currently implemented strategy in the study area. However, increasing LLIN use and IRS coverage and initiating a school-based IST program will require investment in several elements not included in this analysis. Firstly, the unit costs of scaling up or introducing some programs will vary by implementation strategy more than others. For example, the majority of the economic cost of the LLIN program implemented by training existing community organizations on distribution is represented by the marginal cost of procuring nets (
Secondly, additional costs will be incurred by determining the appropriate strategy for achieving programmatic goals. Several scenarios in this experiment assume LLIN use of 80%, which is an ambitious target that will depend not only on universal coverage but a large behavior change communications component. Understanding of the behavioral determinants for why nets existing in households currently remain unused will be critical to achieving this goal. In addition to increased personnel and commodities, increasing coverage of IRS will require continued monitoring of insecticide resistance in the vector population, as well as understanding why households remain unsprayed, whether it is due to rejection by household members or the inability to logistically access hard to reach households. Implementing a school-based IST program as intensive as twice per school term over an extended period of time could result in a change in adherence rates as well as an increased risk of selecting for drug resistance, elements which may impact the effectiveness of the intervention if community acceptability is not assessed.
Thirdly, the study does not allow for any economies or diseconomies of scale for the costs of commodities and program delivery, assuming costs will grow linearly with scale up. In practice this will likely not be the case; increasing intervention coverage from 70% to 80% may be more expensive than scaling up from 50% to 60%.
Assessing the epidemiologic impact and cost effectiveness of different intervention combinations is a necessary element in considering a change of malaria control policy, but it is by no means the only criteria with which to base a recommendation for policy change. Changes in implementation, whether this includes new strategies to increase coverage and use of existing interventions or the addition of a new intervention, will have implications on acceptability by the individuals and communities receiving the interventions, the personnel involved in service delivery, the natural environment into which additional insecticides could be introduced, and the systems of surveillance and monitoring for indicators of malaria and other febrile illnesses, to name a few. Conducting a health impact assessment, drawing on existing frameworks
(DOCX)
(DOCX)
We would like to thank the community of Rachuonyo South District for their support and cooperation. We acknowledge colleagues and partners at the Kenya Medical Research Institute (KEMRI)/US Centers for Disease Control and Prevention, Kisumu, Kenya and the Swiss Tropical and Public Health Institute, Basel, Switzerland for helpful discussions. We are also grateful to the MalariaControl.net volunteers for donating computing resources to run simulations. Preliminary results of this study were presented at the American Society of Tropical Medicine and Hygiene Annual Meeting on November 15th, 2013 in Washington DC, USA, Scientific Session 66 – Mosquitoes: Vector Biology - Epidemiology II (Abstract Number: 1523; Presentation Number: 520). This article has been approved by the Director of KEMRI.