Background
Scale up of malaria prevention—mainly with the mass distribution of long lasting insecticidal nets (LLINs) and indoor residual spraying (IRS) of the interior lining of the wall of the houses—have brought a remarkable reduction in the global burden of malaria in the last decade [
1,
2]. Empirically, the effectiveness and the cost-effectiveness of both LLINs and IRS, for malaria prevention, are well-established [
3‐
7]. However, evidence also indicates that neither LLIN nor IRS—alone—will be sufficient to reach and maintain the interruption of transmission in highly malarious regions of Africa [
8‐
10].
In Ethiopia, LLINs and IRS are usually implemented separately in different districts or different villages [
11,
12]. While LLINs and IRS in some villages have been implemented simultaneously within the same households [
12‐
16], little is known about the effect and cost-effectiveness of the combined use of LLINs and IRS [
12,
17]. Moreover, are the additional costs of the combined interventions reasonable from a provider’s perspective given the combined benefits? [
7].
Mathematical models by Yakob et al. [
18], Okumu et al. [
19], and Chitnis et al. [
20] shows that there might be some additional protective value by a combined implementation of LLINs and IRS compared to either of them alone. A review of cross-sectional data from 17 different countries in sub-Saharan Africa also shows that people in households which use both bed-nets and IRS are about 36% (95% CI 7%–53%) better protected compared to households which only use one of the interventions in medium malaria transmission areas [
16]. Similarly, studies from Kenya [
21] and Tanzania [
22] also report positive results of combining LLINs and IRS. Kleinschmidt et al. based on literature search and a cross-sectional survey from Bioko island of Equatorial Guinea, conclude that the increased resource use of the combined intervention is justifiable because of additional effectiveness compared with each intervention alone [
23]. On the other hand, randomized trials from Benin [
24], Gambia [
25], and Sudan [
26] report that there is no added effect in the combined implementation, compared with each intervention implemented separately.
However, none of those studies estimated the effect at a general population level, nor did they attempt to evaluate the cost and cost-effectiveness of the interventions. In the battle against malaria the need for transparent evidence based on randomized controlled trials, which integrates robust decision modelling, is critical to allocate scarce resources appropriately [
27]. Such evidence will be useful to guide the selection of the packages of interventions for malaria elimination programs. Specifically, the pressing questions in this line of inquiry are; first, what are the additional effects of combining both LLIN and IRS compared with singleton interventions or the routine practice? Second, is the value of added protection substantial enough to justify the additional resources (i.e. cost) required for a combined implementation? Therefore, the aim of this study was to compare the cost-effectiveness of combined implementation of IRS and LLIN, compared with LLIN alone, IRS alone and routine practice in Ethiopia.
Scenario analysis with literature-based cost-effectiveness model
The overall incidence of malaria in the study area was low during the study period compared with historical data and national average estimate [
44]. Even though the interaction of weather changes and malaria incidence is complicated, a likely explanation for such a low incidence, in addition to the intensive intervention of the research project, could be the atypical weather during the study period. During the years 2015 and 2016, the study area was seriously stricken by drought which was related to an
El Nino event. Meteorological data from the study area show that rainfall decreased from 909 mm to 673 mm during 2011–2014 to 471 mm in 2015. At the same time, the average annual high temperature in 2015 (29 °C) was elevated with 2 °C compared to 2014 (27 °C).
Furthermore, the trial finding unexpectedly showed that the incremental protective effectiveness of either the combined intervention or the singleton intervention was not significantly different from the routine practice. While this result has been observed also in a few other studies [
17], a majority of the empirical literature concludes that LLIN and IRS have substantial protective-effectiveness against malaria [
17]. While we believe the internal validity of these results are good for the timing and context of this trial, the generalizability of the effectiveness and cost-effectiveness of combined implementation of LLIN and IRS are more uncertain.
We therefore performed a literature-based scenario analysis of cost-effectiveness for two reasons. First, we wanted to reduce a limitation of the trial-based evaluation—poor external validity. Second, we wanted to give decision-makers more flexibility to interpret results subject to a broader set of contexts.
In the literature-based cost-effectiveness analysis, we changed the input values from the trial-based analysis for malaria risk in the area (annual malaria probability) and protective-effectiveness of the interventions. We defined annual malaria probability as the probability of acquiring a malaria episode per person within a given year. We applied the annual parasitic incidence (API) measured as malaria probability per annum per 1000 population at risk. The API is the most common and reliable estimate of malaria probability in a specified geographic area [
45]. In Ethiopia, about 17% of the districts on average have API lower than 5; and 43% of the districts have 5–100 API while nearly 7% has API greater than 100 [
46]. Nearly 33% of the district are malaria-free. Based on the World Health Observatory data, the average API for Ethiopia is 58 per 1000 population at risk in 2015, And therefore we assume a base-case annual malaria probability of 5.8% (i.e. background malaria risk in the area) [
44]. For intervention arms, we multiply the annual malaria probability by the protective-effectiveness of the interventions to estimate the transition probability in the corresponding arms with the presence of the interventions [
34].
Regarding effectiveness, we utilised a systematic review to assume that the most likely values for protective effectiveness are 40% (35–45%) for LLINs alone and 28.5% (23.5–33.5%) for IRS alone [
17]. For the combined intervention, we calculated protective-effectiveness as the multiplicative combination of the individual risk of malaria of the singleton interventions (LLIN and IRS) [
47], yielding a protective-effectiveness of 57%.
One-way sensitivity analysis
To test the robustness of model conclusion to these assumptions, we performed one-way sensitivity analyses on the literature-based cost-effectiveness model, in addition to the PSA. We did this for different level of protective-effectiveness of the combined interventions (47.1–67.1%) and at different level of annual malaria incidence (1–20%), and present results in a tornado diagram, where also the variables time horizon, cost, proportion of cases diagnosed, proportions of cases treated, probability of mortality from severe malaria were included. We evaluated the incremental cost-effectiveness values against the willingness to pay thresholds of less than or equal to 1 times GDP per capita.
Discussion
This cluster randomised controlled trial found no significant difference in the effects of malaria prevention. The effectiveness of all the three intervention arms was the same as the routine arm, and the economic evaluation inevitably shows that the current routine practice dominates all the prevention alternatives since they are all more costly. When generalising key inputs from the trial and replacing them with literature-based assumptions, the economic evaluation shows that both—LLINs alone with ICER of USD 207 and the combined intervention with ICER of USD 1403—are likely to be ‘cost-effective’ compared to a willingness to pay threshold of 3 times GDP per capita per DALY averted. At a willingness to pay threshold of 1 times GDP per capita, only LLIN alone is likely to be cost-effective, while IRS is dominated by LLIN (more costly but less effective).
This study is the first of its kind in Ethiopia which compared the cost-effectiveness of malaria prevention interventions. Our literature-based analysis yield higher ICERs for both the combined intervention and LLIN alone compared to previous studies on malaria prevention [
7]. For example, Goodman et al. analysing the cost-effectiveness of malaria in sub-Saharan Africa, found an ICER (in 1995 USD per DALY averted) ranging only from 32 to 58 for ITN and from 16 to 29 for IRS [
48]. Another study by Morel et al. [
49] examined the cost-effectiveness of mixes of curative and preventive interventions, and reported an ICER (in 2005 international dollar per DALY averted) ranging from 10 for case management with artemisinin-based combination therapy to 96 for combination of the four interventions together (i.e. IRS, insecticide-treated net (ITN), case management with artemisinin-based combination therapy (ACT), and intermittent presumptive treatment in pregnancy). A systematic review of studies published between 2000 and 2010 [
6] reported a median ICER of 27 per DALY (range 8.15–110) for LLIN/ITN and 143 per DALY (range 135–150) for IRS.
The relatively high ICERs in this study compared to other studies in Africa can be partly explained by the differences in malaria burden, the incremental costs of interventions, and unique malaria dynamics in Ethiopia. In the last 15 years, the incidence of malaria in sub-Saharan Africa decreased significantly [
2,
44,
50], while the cost of the interventions increased [
51]. The cost of the intervention increased mainly because of the replacement of DDT (Dichlorodiphenyltrichloroethane) with Propoxur, use of LLINs instead of ITNs, and the introduction of the ACT. All these three recent changes were not only associated with improving malaria prevention and control, but also with increased cost to the health system. Particularly IRS was costly in our study and this was mainly caused by the price of the insecticide. In theory, although insecticide resistance can be one of the factors which can affect the cost-effectiveness of IRS or combined intervention, it is less likely that it’s’ effects could influence the results in our trial. Our entomological study indicates that the efficacy of the insecticide (Propoxur) applied in all the study arms was similar and very potent in all study arms for the whole year [MalTrials Final report, Unpublished]. In addition, the difference in malaria epidemiology in Ethiopia compared with other places in African or elsewhere could also largely contribute to this disparity [
52]. The epidemiologic profile of malaria in Ethiopia is in a number of ways different compare to other African countries. For example, malaria transmission in Ethiopia is low to moderate, unstable, and seasonal while it is high, stable, and perineal elsewhere [
52,
53].
Practically, economic evaluation of malaria prevention interventions is complex [
7,
54]. Unlike typical cost-effectiveness evaluations, in some cases, the effects of combined interventions might be the same with the effect of individual interventions alone; and subsequently, the incremental effect could be negligible. In other cases, any of the intervention might not be effective at all—even compared with ‘doing-nothing’ [
17]. In our study also we found that the effectiveness of the combined intervention was the same with both singleton and routine interventions. This might be partly explained by the ‘counter-balanced effect’ between incremental health effect and cost saved resulting from adding IRS over high LLIN coverage or vice versa. On the other hand, the strong protective effect from active case finding and treatment by itself might dilute the ‘modest’ protective-effects from other preventive measures (i.e. LLIN and IRS). It is also important to note that in this trial—across all the four study arms—a weekly visit to each household was conducted in order to identify any febrile member of the household, and almost all febrile cases were tested with RDT, and if found positive, treated with the appropriate ant-malaria drug [
28]. Therefore, we strongly recommend further pragmatic trials from different setting from our study to estimate protective-effectiveness of the intervention.
In general, the cost-effectiveness of malaria prevention intervention is a function of the health benefit gained and the resources required to implement the intervention [
27]. In the one-way sensitivity analysis, first, we tested the effect of the background malaria incidence in the area on the cost-effectiveness of the interventions by varying the annual probability of malaria infection for an individual from 1 to 20%. On account of this, the ICER for combined intervention varied from about USD 8000 to USD 200 per DALY averted. Moreover, the annual malaria incidence should be at least about 13% in order for the combined intervention to be cost-effective compared with a willingness to pay threshold of 1 times GDP per capita per DALY averted for Ethiopia (USD 628). However, what the recent data from Ethiopian Ministry of Health indicates is that only a few areas in Ethiopia have malaria risk levels of such magnitude. Only about 5% of the districts in Ethiopia, mostly in Western lowlands and few in the Rift Valley, have annual incidence rates exceeding 13% [
46], and based on the results of this analysis should be the focus of attention for future prevention campaigns.
These findings should be interpreted in the light of at least two important issues about the dynamics of malaria control program at low incidence setting (i.e. at stages of elimination and eradication) should look like. First, malaria control program should not be a victim of its own success [
55]. When the malaria control program succeeds, malaria incidence will certainly reduce. In this case, such a versatile malaria prevention interventions like IRS and LLIN will not continue to be sufficiently competitive in terms of cost-effectiveness parameter, and LLINs and IRS will both appear to be not cost-effective [
55]. Therefore, it has been argued that for malaria prevention programs the willingness to pay thresholds should be expanded from the conventional level [
56]. Second, the need for disaggregate malaria data at a district level is crucial for better targeting of interventions and for local planning (micro-planning). In this regard, the National Malaria Control Program in Ethiopia has also recently stratified all districts based on annual malaria incidence into four groups (i.e. free, low, moderate, and high) and started conducting interventions based on the strata [
46].
The second parameter that we examined in the one-way sensitivity analysis was the protective-effectiveness of combined intervention. We found that the combined intervention (LLIN + IRS) need a minimum of 53% protective-effectiveness in order to be ‘cost-effective’ alternative (Fig.
5). It is important to remember that none of the interventions have an inherent degree of effectiveness. Rather, it is the manner how it is implemented, the identification of those areas where it is most suitable, and the proper use by the community which determine the effectiveness most. However, based on a recent systematic review [
17], it would be very challenging to achieve a protective-effectiveness of such level (53%) against the current supply side and demand-side barrier which reduces the effectiveness of both individual and combined interventions. The major demand side barriers for LLIN, observed in our visits, includes under-utilization, misuse, and lack of convenient sleeping space to hang-up the bed nets; while refusal, covering the wall of the house with a mud or other material, and rudimentary nature of the wall for some of the houses were challenges for IRS. The financial and human capacity of the district to execute the interventions, the price of the insecticide, and the quality of the LLINs can be considered as major supply-side barriers. IRS demands strong and very close supervision.
The costing analysis shows that the unit cost of IRS per person-year protected was predominantly influenced by the price of the insecticide, which alone accounted for about sixty percent of the cost. Regarding the cost of LLIN, in addition to the price of the bed nets, useful life-year (durability) of the bed nets was important parameters which determined the cost of LLIN per person-year protected. The life-year of the LLINs determines the frequency of the redistribution (refill). In Ethiopia, based on the National Malaria Program, LLINs were intended to serve for about 3 years and therefore the distribution campaigns are held once every 3 year [
11]. However, what we observed in our study was that the LLINs worn out faster, and had little effect after 1 to 2 years. Local production of the bed nets with low cost and better quality could reduce the price of the bed nets. A strong quality control mechanism in the production, procurement, and distribution of the nets can be considered not only to maintain the fabric integrity of the nets but also to maintain the insecticidal property. Above all, a well-coordinated information, education, communication and advocacy program promoting proper utilization of the LLIN could improve both effectiveness and longevity of the LLINs.
In this study, most of the cost items of malaria prevention interventions at the district level were identified, measured, and valued prospectively alongside the community trial using robust techniques [
27]. Yet, there are some caveats that deserve due consideration with respect to the data, generalizability, and relevance of this study. The first limitation was that the costing was done only from the local providers’ perspective and a few cost items incurred at national and regional levels (e.g. mass-media and communication costs etc.) were omitted. Although this might not have substantial implication when we compare the cost and cost-effectiveness of the prevention interventions, this might to some extent underestimate the actual unit cost of the interventions.
The second limitation to our model was that we were not able to account for health-loss from co-morbidities of severe malaria such as anaemia, convulsions, and long-term neurological sequel because of lack of accurate estimates about the magnitude of these events. Despite the effect on the ICER would be minimal since the probability of severe cases is rare in the cases of treated malaria [
36], this might underestimate the actual benefit of the prevention intervention slightly [
57].
A third limitation of this study is in the decision we made in choosing 1 times or 3 times GDP per capita per DALY averted as a willingness to pay thresholds for interpretation of the ICER results. Despite long-standing debate in economic evaluation literature on this issue [
58], it is particularly important for the evaluation of malaria prevention interventions in Ethiopia [
57]. It is difficult to precisely define the WTP threshold in Ethiopia due to the fact that the financing of health care in general and malaria programs, in particular, are complex. For example, the larger share of the funding (78.6%) for malaria is generated from different to external sources (UNICEF, PMI, GLOBAL FUND, WHO etc.) and most of which is also ear-marked for malaria (vertical program) [
59]. This kinds of cost-effectiveness evidence would be most relevant in a country where there is functional and established disease control priority-setting system which utilizes economic evaluation in decision making [
60].
Authors’ contributions
AH and BR undertook the data analysis and developed the first draft of the methodology section. The first draft manuscript was prepared by AH. All authors substantially participated in the conception of the research idea, design of data collection tools, and interpretation of the result. AH and TG coordinated and supervised the data collection for the trial. All the authors read and approved the final manuscript.