Abstract
Although malaria incidence in Africa decreased substantially between 2000 and 2015 due to large-scale control efforts relying mainly on the distribution of insecticide-treated nets, this decrease has stalled in recent years. To further reduce the residual transmission, new vector control tools are needed. Transfluthrin-treated eave ribbons are considered a promising spatial repellent to protect people when they are indoors but not under an insecticide-treated net (ITN), and potentially also while they are in the peridomestic area. We conducted semi-field and field studies (published elsewhere) to assess the effectiveness of eave ribbons, the odour-baited Suna trap and the combination of these two interventions (push-pull system) on reducing mosquito-human contact and vectorial capacity in an East African setting. Here we combine the semi-field and field data to parameterise the effects of the eave ribbons repellent, the trap and the push-pull system on the mosquito population in an individual-based simulation of malaria epidemiology.
This modelling framework accounts for effects that build up over time, such as reduced infectiousness of the human population, and fully maps the uncertainty of the intervention parameter estimation on the final results. This allows us to estimate the impact of the interventions on clinical malaria in two settings: Ahero in Western Kenya and the Kilombero valley in Tanzania. Our results suggest that the transfluthrin-treated eave ribbon may substantially decrease malaria case incidence in settings with low-transmission or with low ITN coverage, especially in regions where Anopheles funestus dominates among malaria vectors and primarily uses human hosts, such as in the Kilombero setting. However, by diverting mosquitoes from indoor to outdoor host-search, the transfluthrin-treated eave ribbon may reduce the community-protective killing effect of an insecticide-treated bed net deployed at the same house. Moreover, at high but incomplete coverage, it may increase malaria risk for people that remain uncovered and are not protected by the baseline ITN intervention. Therefore, we suggest deploying this spatial repellent only as a targeted intervention complementing insufficient indoor protection by ITNs and to continue ensuring the highest possible coverage and use of ITN. The much smaller effect sizes in the Ahero setting, likely due to the lower anthropophily of An. funestus, shows that assessment of the potential of a candidate vector control tool is only meaningful under specification of an entomological and epidemiological setting. There is a high degree of uncertainty in our estimates because it is unclear whether the spatial repellent primarily kills or disarms the mosquitoes, which were shown to stop host-seeking due to the repellent in semi-field experiments. Although further research is needed to elucidate the levels of repellency, killing and disarming induced by the volatile transfluthrin under different climatic conditions, transfluthrin-treated eave ribbons offer a promising supplementary tool to ITNs to protect people at times and locations when they are in or near the house but not under the net.
1 Introduction
Malaria incidence has been drastically reduced between 2000 and 2015, primarily due to the large-scale implementation of insecticide-treated nets (ITNs) and to a smaller extent due to indoor residual spraying (IRS) and treatment with artemisinin-based combination therapy [Bhatt et al., 2015], but progress has stalled since then [WorldHealthOrganization, 2020]. Even if coverage, use and formulation of ITNs and IRS could be improved further, residual transmission of malaria would still persist, in part because transmission occurs at times and places beyond the reach of these vector control tools. There is evidence that a substantial part of malaria transmission takes place outdoors and in the evenings; and that the space near houses (peridomestic area), usually within 10 m, where household members spend time before going indoors to sleep, is likely to play an important role in transmission [Killeen et al., 2006, Govella et al., 2010, Monroe et al., 2019].
Behavioural adaptations to avoid exposure to insecticides have been discovered in certain Anopheles species [Russell et al., 2011, Moiroux et al., 2012] and the species composition in many places changed drastically after scale-up of ITNs [Mwangangi et al., 2013]. In particular, the populations of Anopheles gambiae sensu stricto, once the dominant malaria vector in East Africa, became negligible while the sibling species Anopheles arabiensis remains very abundant [Bayoh et al., 2010]. Species of the Anopheles funestus complex have become increasingly important for malaria transmission in East Africa [Lwetoijera et al., 2014, Kaindoa et al., 2017].
Moreover, mosquito populations across Africa have become increasingly physiologically resistant to the insecticides used in ITN and IRS formulations [Liu, 2015, Hancock et al., 2020, Moyes et al., 2020]. This may further increase residual transmission and jeopardise past achievements, and calls into question a vector control strategy relying mainly on ITNs and IRS [Hemingway 2016]. Therefore, new tools for mosquito control need to be developed [malERA, 2017, Killeen and Moore, 2012, Killeen, 2014, Sougoufara et al., 2020].
To speed up their development, outdoor vector control tools, and spatial repellents in particular, are typically tested first in semi-field sites and then in field studies. Semi-field sites are large structures, typically more than 100 square meters in ground area, covered with a net fabric that allows for nearly realistic field conditions inside the structure in terms of airflow, temperature and humidity. Insectary-reared, disease-free mosquitoes can be released safely inside such semi-field sites and recaptured by human landing catch (HLC) [Service, 1993] in order to investigate their response to a vector control tool deployed inside the site.
Odour-baited traps for mosquito control are designed to kill mosquitoes after they are attracted by a synthetic human-odour mimic [Okumu et al., 2010], a principle against which insensitivity is unlikely to develop. A cluster-randomised controlled trial in Western Kenya showed a strong impact of mass-trapping with the Suna trap on malaria transmission and prevalence [Homan et al., 2016].
Spatial repellents are designed to repel mosquitoes from a defined area, without necessarily killing them [Achee et al., 2012]. However, most spatial repellents use a slowly evaporating insecticide as an active ingredient and therefore may also kill (immediately or after a delay), or disarm mosquitoes (inhibit hostseeking behaviour until the next night without killing) [Ogoma et al., 2012a]. Sisal fabrics treated with transfluthrin [Ogoma et al., 2012b, Govella et al., 2015, Ogoma et al., 2017] were recently applied as insecticide-treated eave ribbons and showed particular promise to protect against mosquito bites in the peridomestic area [Mmbando et al., 2018, Mwanga et al., 2019, Njoroge et al., 2021], especially due to their low cost and minimal need for user compliance.
Spatial repellents can also be combined with odour-baited traps to form a push-pull system, to mitigate the risk of spatial repellents pushing mosquitoes to unprotected humans. The rationale is to achieve a synergism similar to that of ecological push-pull systems used in agricultural pest control [Cook et al., 2006]. Push-pull systems were shown to reduce house-entry of malaria vectors [Menger et al., 2015] and recent studies assessed their impact on both indoor biting and outdoor biting in the peridomestic area [Mmbando et al., 2019, Njoroge et al., 2021, Denz et al., 2021].
The development of a new vector control intervention is not finished with the product itself. Its potential for reducing malaria incidence needs to be assessed and suitable intervention strategies for different settings need to be defined [WorldHealthOrganization, 2017]. In the absence of a predictive model, very large-scale and long-term trials on both entomological outcomes and malaria incidence would be needed, in particular because important effects on the mosquito population and malaria prevalence build-up over longer times and the relationship between malaria transmission and incidence is non-linear. Moreover, many trials would be needed to cover the great variability of entomological settings across Africa and the different existing control interventions. Models of malaria epidemiology that take into account the aforementioned aspects can be a powerful tool to answer these questions more quickly and at lower cost, thus greatly assisting product development. Accurate prediction of the effect of a new tool, however, requires relatively complex models that incorporate a detailed understanding of the tool’s modes of action, its potential interaction with existing interventions, and estimation of the corresponding key parameters.
The effectiveness of insecticide treated nets (ITN) relies on both direct prevention of mosquito biting (personal protection) and killing mosquitoes (community protection, in which sense an ITN can be seen as a human-baited mosquito killing station). A spatial repellent that pushes mosquitoes away from the house may therefore reduce the effectiveness of the ITN. Modelling can be used to better understand the interaction of the spatial repellent with the ITN and to find optimal strategies combining these two interventions.
We conducted large semi-field studies in Mbita, Western Kenya, [Njoroge et al., 2021] and Bagamoyo, coastal Tanzania, [Tambwe, 2018] evaluating different candidates and formulations of the spatial repellent and the push-pull system. A sisal fabric (hessian strip) impregnated with emulsified transfluthrin (2.5 g active ingredient / m2 of fabric) and hung at the eave of the house without covering the eave was identified as the best spatial repellent formulation [Njoroge et al., 2021] and will be referred to as ‘the spatial repellent’ in the rest of the paper. The chosen push-pull system consisted of this spatial repellent and the Suna trap, baited with the MB5 blend and carbon dioxide from molasses fermentation, and placed inside the peridomestic area approximately 5 m from the house and 0.3 m above ground. This proof of concept study found a strong reduction of HLC by An. arabiensis due to both the spatial repellent and the push-pull system, and related these findings to measurements of volatile trans-fluthrin concentrations by air sampling at different heights and locations in the semi-field system [Njoroge et al., 2021]. The Suna trap alone showed no effect on HLC in this setting [Njoroge et al., 2021].
From the same data, we estimated the differential effects of the spatial repellent on An. arabiensis in terms of repellency on the one hand and killing or disarming on the other, using a stochastic model of mosquito host-seeking behaviour in semi-field experiments, based on time-stratified human landing catch (HLC) data [Denz et al., 2021]. Distinguishing these differential effects is important to estimate community protection, which cannot be measured directly from semi-field experiments running over one night. However, it was not possible to distinguish between killing and disarming from such data, because both effects affect HLC [Denz et al., 2021]. Based on these findings, the impact of the spatial repellent, the Suna trap and the push-pull system on the ability of An. arabiensis to transmit malaria (vectorial capacity) was estimated, under assumptions of either killing or disarming [Denz et al., 2021].
A Latin-square block-randomised controlled trial over 12 houses and 17 weeks, which we conducted in Ahero, Western Kenya, assessed the chosen pushpull and individual interventions in terms of relative reduction of biting (protective efficacy) by An. arabiensis, An. funestus and other mosquito species, both indoors (CDC light trap) and outdoors (HLC) [Fillinger et al., 2023]. This study showed strong effects on indoor biting across all species for both the spatial repellent and the push-pull system; but no outdoor effect for the malaria vectors and even an increase of An. funestus, presumably due to mosquitoes being diverted from biting indoors. The Suna trap, baited with MB5 and 2-butanone [Mburu et al., 2017] instead of carbon-dioxide, in this field study, showed no effect in terms of protective efficacy and only a small effect on the catch of malaria vectors.
This modelling study estimates the potential impact of the spatial repellent, the Suna trap and the push-pull system on clinical incidence of malaria and other epidemiological outcomes, for two baseline settings: Ahero in lowland Western Kenya and the Kilombero Valley in Tanzania, for a range of coverage levels of these new interventions, different intervention scenarios, and assumptions of overlapping coverage with ITNs. Sensitivity analyses for important entomological and epidemiological baseline characteristics of these two locations are performed to account for parameter uncertainty and to allow for translation of the findings to other settings. We are not aware of other modelling studies on the impact of transfluthrin-treated eave ribbons, the Suna trap or the push-pull system on incidence of malaria.
An open-source, stochastic, individual-based simulation platform for malaria [OpenMalariaDevelopmentTeam, 2020] is used to model the intervention dynamics in order to include the heterogeneity of malaria in humans and the nonlinear relationship between malaria transmission and clinical disease. For the accurate parameterisation of the different intervention effects on both An. arabiensis and An. funestus, semi-field data sources [Njoroge et al., 2021, Tambwe, 2018] and field data sources [Fillinger et al., 2023], are combined. While the field data provide stronger evidence and include the different wild-type mosquito species, the data from well-controlled semi-field experiments are needed to estimate the killing/disarming effect and the killing after biting effect, which are both of great importance in predicting the community effect of the new interventions. Estimation of all intervention parameters relied on Bayesian inference. Random samples drawn from the corresponding posteriors were used as inputs to the simulation in order to capture the uncertainty in the intervention effects.
2 Results
2.1 Scale up of the transfluthrin-treated eave ribbons, the Suna trap and the push-pull system in the Ahero setting
Estimates of the incidence of uncomplicated malaria when the transfluthrintreated eave ribbons (spatial repellent), the Suna trap (trap) and the push-pull interventions are deployed with increasing coverage are shown in Figure 1 for the Ahero baseline setting, and different baseline assumptions with respect to transmission intensity (as measured by the entomological inoculation rate, EIR) and ITN coverage.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Ahero with either the spatial repellent (blue), trap (red) or push-pull (purple) intervention with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines representing the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Figure 2 shows the corresponding relative reduction of incidence.
Estimates of the relative reduction of malaria incidence under different transmission settings in Ahero with either the spatial repellent (blue), trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines representing the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), with solid edges indicating disarming and dashed edges the killing assumption for the spatial repellent and push-pull interventions. Note that the vertical axes are aligned per row, but not across all transmission settings.
At baseline, the low, mid-range and high transmission settings correspond to EIRs of approximately 13, 30 and 60 under the 60% ITN coverage scenario and to EIRs of approximately 8, 20 and 40 under the 80% ITN coverage scenario (see Figure 19 in the Supplementary information). Since it is currently unknown whether mosquitoes which stop host-seeking behaviour as a result of the spatial repellent (as shown in semi-field studies [Denz et al., 2021]), are disarmed or killed, separate scenarios for these two assumptions were considered. The median estimates for both the disarming and the killing scenario are reported together with 50% credible regions with respect to the combined uncertainty from both scenarios (lower bound: 25% percentile under killing assumption, upper bound: 75% percentile under disarming assumption).
Overall, all candidate interventions showed modest to good impacts on malaria incidence. The push-pull system performed best in all settings followed closely by the spatial repellent and with a larger gap to the Suna trap. However, there are considerable differences across the baseline settings and uncertainty is generally large, especially due to the open question on the killing and disarming effects of the transfluthrin-treated eave ribbon.
Figure 2 shows that for the setting with low transmission and ITN coverage of 60%, a 80% coverage with the push-pull system is estimated to reduce incidence in the median by about 15% under the disarming assumption and by about 30% under the killing assumption, with a combined 50% credibility interval ranging from 10% to 40% reduction. Similarly, an 80% coverage with the spatial repellent is estimated to reduce incidence in the median by about 10% under the disarming assumption and by about 35% under the killing assumption, with a combined 50% credibility interval ranging from 5% to 35% reduction. The trap is estimated to reduce incidence by only about 5% in this setting, with a narrow uncertainty interval. For the same transmission setting with ITN coverage of 80%, very similar incidence reduction by the push-pull system was estimated, while the estimates for the spatial repellent were about 5% lower as compared to the setting with 60% ITN coverage. However, the impact of the trap was estimated to be considerably stronger in this setting with high ITN coverage, with a relatively precise estimate of 10% incidence reduction. For the mid-range transmission setting, relative reduction of incidence was consistently about halved across all interventions and both ITN settings, as compared to the low transmission setting. The conservative estimates for the spatial repellent and the push-pull system under the disarming assumption were at about 5% incidence reduction in the median, while the trap achieved this level of reduction only under the high ITN scenario. For the high-transmission setting, estimates of relative incidence reduction were again slightly lower, but otherwise similar. Under the assumption of increased effective coverage with case management (CM), malaria incidence was lower overall, but patterns with respect to the interventions were consistent (see Figure 20 in the Supplementary information).
Under the disarming assumption, the spatial repellent and the push-pull system performed very similarly across all settings. Under the killing assumption, however, the push-pull system was superior at high coverage levels. In most settings, the effect of the spatial repellent stalled at about 50% coverage, while the push-pull system exhibited a nearly linear reduction over the whole range of coverage levels. The Suna trap performed better, both absolutely and in comparison to the other interventions, in settings with high ITN coverage, indicating a synergistic effect between the Suna trap and the ITNs.
2.2 Scale up of the transfluthrin-treated eave ribbons, the Suna trap and the push-pull system in the Kilombero valley setting
Figure 3 and 4 display that, in the Kilombero setting, both the push-pull system and the spatial repellent performed much better compared to the Ahero setting; and both interventions showed very similar effectiveness to each other.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Kilombero with either the spatial repellent (blue), trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines representing the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes do not start at 0.
Estimates of the relative reduction of malaria incidence under different transmission settings in Ahero with either the spatial repellent (blue), trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines representing the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), with solid edges indicating the disarming and dashed edges the killing assumption for the spatial repellent and push-pull interventions. Note that the vertical axes are aligned per row, but not across all transmission settings.
The Suna trap again did not perform well, but still better than in the Ahero setting. However, uncertainty was generally large and there were some differences across the different baseline assumptions with respect to transmission intensity (EIR) and ITN coverage. At baseline, the mid-range and high transmission settings corresponded to EIRs of approximately 22 and 48 under the 60% ITN coverage scenario and to EIRs of approximately 13 and 32 under the 80% ITN coverage scenario (see Figure 23 in the Supplementary information).
Figure 4 shows that for the setting with mid-range transmission and ITN coverage of 60%, a 80% coverage with either the push-pull system or the spatial repellent is estimated to reduce incidence in the median by about 30% under the disarming assumption and by about 70% under the killing assumption, with a combined 50% credibility interval ranging from 20% to 80% reduction. The trap is estimated to reduce incidence by only about 5% in this setting, with little uncertainty estimated. For the same transmission setting with ITN coverage of 80%, very similar incidence reduction by both the push-pull system and the spatial repellent was estimated. However, the impact of the trap was estimated to be stronger in the setting with high ITN coverage, with a relatively sharp estimate of 10% incidence reduction. For the high transmission setting, relative reduction of incidence was consistently about one third lower across all interventions and both ITN settings, as compared to the mid-range transmission setting. The conservative estimates for the spatial repellent and the push-pull system under the disarming assumption were at about 20% incidence reduction in the median for both ITN coverage levels. The trap was estimated to reduce transmission by only about 4% and 8% in the settings with 60% and 80% ITN coverage, respectively. Similar to the Ahero setting, under the assumption of increased effective coverage with case management (CM), malaria incidence was slightly lower overall, but patterns with respect to the intervention were consistent with the baseline CM coverage (see Figure 24 in the Supplementary information).
Under the killing assumption, the spatial repellent performs slightly better than the push-pull system at low coverage levels, while for high coverage levels the push-pull system is superior. In particular, under the killing assumption, the incremental effects of increasing the coverage with the spatial repellent over 60% is minimal across all settings. The same is only true in the mid-range transmission level for the push-pull system, while it continues to reduce incidence in an almost linear fashion in the high transmission setting. Under the disarming assumption there is no such saturation effect for either of the two interventions involving transfluthrin-treated eave ribbons. In settings with higher ITN coverage, the performance difference between the Suna trap and the other interventions is slightly smaller than in settings with lower ITN coverage. The Suna trap again performs better in settings with higher ITN coverage, indicating a synergistic effect between the Suna trap and the ITNs.
2.3 Intervention deployment strategy with respect to the baseline ITN intervention
Figures 5 and 6 show, for the Ahero and the Kilombero setting respectively, the impact of increasing coverage with the candidate interventions on the median malaria incidence under different deployment strategies with respect to the baseline ITN distribution in the population.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) in Ahero for different intervention scenarios with regard to allocation of the spatial repellent (blue), trap (red) or push-pull (purple) intervention first to people without ITN (minimal overlap, solid lines), irrespective of ITN ownership (random overlap, dashed lines) or to people with ITN (maximal overlap, dotted lines). All lines denote the median over all simulations under the disarming assumption for a setting with mid-range transmission intensity and ITN coverage of 60%.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) in Kilombero for different intervention scenarios with regard to allocation of the spatial repellent (blue), trap (red) or push-pull (purple) intervention first to people without ITN (minimal overlap, solid lines), irrespective of ITN ownership (random overlap, dashed lines) or to people with ITN (maximal overlap, dotted lines). All lines denote the median over all simulations under the disarming assumption for a setting with mid-range transmission intensity and ITN coverage of 60%.
The impact of the spatial repellent and the push-pull system was maximised with a complementary intervention strategy with respect to the ITN intervention (the new intervention was first given to people without ITNs: ‘minimal overlap’). When increasing coverage with the new intervention beyond the people not covered by the ITN intervention under this strategy, no incremental impact was estimated for the Ahero setting (Figure 5) and a considerably smaller incremental impact was estimated for the Kilombero setting (Figure 6). Under an intervention strategy giving the new intervention first to people already covered by the ITN intervention (maximal overlap), the impact of increasing coverage with the new intervention was marginal until reaching the baseline ITN coverage. This clearly indicates an antagonistic interaction of both the spatial repellent and the push-pull system with the ITN, due to the fact that the new interventions push mosquitoes away from the house and hence outside of the reach of the killing effect of the ITN. This effect is slightly stronger for the spatial repellent without the trap, explaining the fact that its impact is stalling earlier when increasing coverage as compared to the push-pull system. The small differences among the three allocation scenarios for the Suna trap intervention is due to the small diversion of An. funestus from indoor to outdoor host search implied by the Suna trap and due to stochasticity. For Figures 5 and 6, mid-range transmission intensity, 60% ITN coverage and a disarming effect of the interventions involving the transfluthrin-treated eave ribbon was assumed. Similar results were obtained under the killing assumption and for different assumptions on the transmission intensity and the ITN coverage.
2.4 Personal and community protection of the transfluthrintreated eave ribbons and the push-pull system
The impact of increasing the coverage with the spatial repellent or the push-pull system on malaria incidence among the different cohorts of people with respect to user status of the new intervention and ITN is shown in Figures 7 and 8 for the Ahero and the Kilombero setting, respectively.
Incidence per person per year among the people that received (a) ITN only, (b) ITN and the new intervention, (c) no intervention and (d) the new intervention only, under increasing coverage with the new interventions spatial repellent and pushpull system. All simulations are with respect to the Ahero setting with mid-range transmission intensity, 60% coverage with ITNs and random overlap between the two vector control interventions.
Incidence per person per year among the people that received (a) ITN only, (b) ITN and the new intervention, (c) no intervention and (d) the new intervention only, under increasing coverage with the new interventions spatial repellent and pushpull system. All simulations are with respect to the Kilombero setting with mid-range transmission intensity, 60% coverage with ITNs and random overlap between the two vector control interventions.
Across both settings, when increasing the coverage with the spatial repellent or the push-pull system, incidence decreased approximately linearly among people covered by the ITN intervention, irrespective of whether they personally received the new intervention or not (Figures 7(a) and (b) as well as 8(a) and (b)). This community effect is due to the disarming/killing effect of the transfluthrin-treated eave ribbons. The spatial repellent and the push-pull system do not show a direct personal effect for users who are already covered by an ITN since they provide some additional indoor protection but at the same time shift biting outdoors where people are not protected (see Figure 13 in the Supplementary information).
People not covered by the ITN intervention benefit most when given the spatial repellent or the push-pull system, with a sharp incidence reduction with increasing coverage of one of these new interventions to 20% and a flatter, approximately linear reduction when increasing coverage further (Figures 7(d) as well as 8(d)). However, for people without ITNs who do not personally receive a new intervention, incidence per person is unchanged (Ahero) or decreases slightly (Kilombero) when increasing respective coverage up to about 40% and rises considerably when increasing coverage further (Figures 7(c) as well as 8(c)). This unfavorable effect is due to the repelling effect of these new interventions which pushes mosquitoes towards unprotected hosts and increases their risk to be bitten in the simulation model. This effect is aggravated due to the declining numbers of people not protected by any intervention, as intervention coverage increases. Hence, for higher coverage levels, this increased biting offsets the community effect due to the killing/disarming effect which was apparent at low coverage levels. This suggests that the priority to ensure high coverage of ITNs should remain.
For Figures 7 and 8, a mid-range transmission intensity, 60% ITN coverage and a disarming effect of the interventions involving the transfluthrin-treated eave ribbon was assumed. Similar results were obtained under the killing assumption and for different assumptions on the transmission intensity and the ITN coverage. No analysis of personal vs community protection was performed for the Suna trap since the trap is not assigned to specific human hosts in the simulation model.
3 Discussion
The transfluthrin-treated eave ribbon (spatial repellent) and its combination with the odour-baited Suna trap (push-pull system) showed a potential to considerably reduce malaria incidence in the Kilombero setting, although its impact in the Ahero setting was smaller.
People not covered by ITNs especially benefit from a strong reduction in case incidence when receiving protection by the spatial repellent. This opens up promising intervention strategies with the spatial repellent in settings difficult to reach with ITNs and IRS such as migrant agricultural workers, especially given the low requirements on user compliance. However, since the spatial repellent provided only indoor protection, residual outdoor transmission is not addressed. The Ahero and Kilombero setting differed mainly in their entomological baseline characteristics, since the different baseline case management coverage levels made little difference to the relative intervention impact. In both settings, An. funestus was the dominant malaria vector, followed by An. arabiensis, with other species having a negligible contribution. For both abundant species, outdoor biting was more pronounced in Kilombero than in Ahero, yielding more residual transmission in Kilombero. However, both transmission intensity and malaria incidence at baseline were lower in Kilombero, since in Kilombero both species were much more anthropophilic and thus the killing effect of the ITN at baseline was more effective in Kilombero. The stronger impact of the interventions involving the transfluthrin-treated eave ribbon in the Kilombero setting compared to Ahero is explained by the strong impact these interventions showed to have on An. funestus and the high human blood index (HBI) for this species of 100% in Kilombero, compared to only 60% in Ahero. Hence, killing or disarming 100 An. funestus mosquitoes in Kilombero prevents 100 mosquitoes from seeking a human host (temporarily or definitively, respectively), while in Ahero this prevents only 60 mosquitoes, on average, from seeking a human host. For Kilombero, the entomological baseline parameters were taken from two different sources, so there may be issues of consistency. In particular, a human blood index (HBI) of 100% for An. funestus [Kaindoa et al., 2017] may not be realistic if the indoor biting proportion of this species was below 50% [Finda et al., 2018]. More importantly, the HBI for An. arabiensis in Ahero from the recent literature [Degefa et al., 2017] seems extremely low [Orsborne et al., 2018]. Comparison of the different settings showed that the relative impact of the interventions involving the eave ribbon was biggest in settings where An. funestus dominates malaria transmission and primarily uses human hosts, and where transmission intensity and ITN coverage are relatively low.
A clear antagonistic interaction of both the spatial repellent and the pushpull system with the ITN was shown, since the transfluthrin-treated eave ribbon pushes mosquitoes away from the house and thus outside of the reach of the killing effect of the ITN. Therefore, deploying the transfluthrin-treated eave ribbon, alone or within the push-pull system, at houses whose inhabitants use ITN showed almost no effect on malaria incidence in the total population, while a complementary intervention deployment strategy showed a strong reduction of malaria incidence. Therefore, a targeted intervention strategy for each setting taking into account the local ITN coverage and use is required when implementing the spatial repellent on a large scale. Interestingly, a modelling study indicated that passive transfluthrin emanators that are placed in the outdoor space further away from the house may push mosquitoes towards indoor host search and thus increase rather than decrease the number of mosquitoes killed by an ITN deployed inside the house [Hellewell et al., 2021]. We suggest field studies with experimental huts combined with measurements of outdoor mosquito densities, including investigating adjacent houses not covered by the spatial repellent, in order to improve the understanding of the interaction of the spatial repellent with both ITNs and IRS.
Our simulations indicated that people who are neither protected by the spatial repellent nor the ITN may experience an increased risk for clinical malaria, which constitutes an opposition to the health equity paradigm. However, this finding is not based on data on unprotected individuals, but only on field estimates of the repellency from protected houses in connection with the modelling assumption of perfect mixing between the mosquito and human populations. This modelling assumption implies that mosquitoes repelled from a protected host can readily encounter an unprotected host. However, this depends on the distance between the two hosts and the dynamics of mosquito foraging behaviour. Therefore, further research on mosquito movement in response to spatial repellents [Saddler et al., 2019, Malinga et al., 2019] is needed to elucidate this relationship and to refine our model of mosquito-human contact. In the meantime our modelling assumption provides a conservative estimate of the intervention impact and highlights the need to cover each human with at least one vector control intervention.
A previous semi-field study with 5 huts inside a single, long SFS indicated protection for inhabitants of an non-fitted house neighbouring a house fitted with the transfluthrin-treated eave ribbon [Mwanga et al., 2019]. This community protection was strongest in the outdoor space and relatively small with respect to indoor biting, so that a similar community effect seems unlikely for our field trial that showed no outdoor effect and on which the intervention parameterisation of the present study relies. Also, a semi-field setup with 5 well spaced houses but very limited air volume may not be a realistic setup to characterise the community protection provided by a spatial repellent. Interestingly, a crossover field study with transfluthrin coils found a strong negative community effect for unprotected neighbours at incomplete coverage (80%) with coils [Maia et al., 2016]. Similar effects were found for topical repellents [Moore et al., 2007, Maia et al., 2013]. Eventually, larger field trials with a different experimental design, preferably cluster-randomised, are needed to assess the impact of transfluthrintreated eave ribbons on inhabitants of unprotected houses under incomplete coverage. Findings from mosquito mark-release-recapture studies [Saddler et al., 2019], modelling studies on mosquito movement [Malinga et al., 2019, Multerer, 2020] may be useful to determine the required spatial extent of the clusters in such studies.
Climatic conditions (temperature and winds) leading to low concentrations of volatile transfluthrin in the outdoor space are thought to be responsible for the absence of an outdoor effect in the field studies that form the data basis of our modelling study, as discussed in [Fillinger et al., 2023]. The fact that the transfluthrin-treated eave ribbon increases rather than decreases outdoor biting, limits its effectiveness, leads to an antagonistic interaction with the ITN and is responsible for the stalling of its impact at high coverage levels. Upon availability of more data on the impact of climate and weather on the effect of the transfluthrin-treated eave ribbon, the climate characteristics of a given setting could be incorporated in our estimates. Other field studies [Mmbando et al., 2018], yet with less conclusive evidence, as well as semi-field studies [Mmbando et al., 2018, Mmbando et al., 2019, Mwanga et al., 2019] estimated both a strong indoor and outdoor protection by the transfluthrin-treated eave ribbon, and basing our study on these data sources would likely lead to much different results.
Estimates of the killing/disarming effect from semi-field experiments may be too high compared to a field setting since the constrained space may force mosquitoes to be exposed to the volatile transfluthrin for unrealistically long times. This may render the simulation experiments overly optimistic, especially for the killing assumption. On the other hand, mosquitoes attempting to enter a house in the field presumably get closer to the eave ribbons than the mosquitoes in the semi-field experiments, where a human was only present outdoors, which may compensate for the over estimation. Likewise, the estimates of the killing after biting effect from the semi-field experiments may be too low.
The semi-field data on An. arabiensis were used to estimate the killing/disarming effect of the transfluthrin-treated eave ribbon on An. funestus, adjusted for the repelling effect estimated from the field data for An. funestus, since semifield studies were not possible for An. funestus. However, this assumption is reasonable because we believe that the likelihood of coming into contact with transfluthrin, which is captured by the species-specific estimate of repellency, is more variable across species and more predictive for the intervention effect than the mosquitoes’ reaction after contact with transfluthrin. The speciesspecific repelling effect estimated from the field trial was not adjusted for the location-specific differences in indoor and outdoor feeding preferences of the given species.
Transfluthrin-coils have also been shown to induce prolonged feeding inhibition of exposed mosquitoes [Ogoma et al., 2014]. However, due to a lack of data for the delivery through eave ribbons, no feeding inhibition was assumed in our simulations, which may lead to an underestimation of the intervention impact. In the field trial there was always an ITN present indoors and thus the effect estimates for the new interventions always include a possible interaction with the ITN. Hence, the effectiveness of the interventions may be not be properly estimated when deployed alone.
The killing/disarming effect estimated from the semi-field data was adjusted to the field situation by scaling with the ratio of the repellency estimated from field data over the repellency estimated from the semi-field data. This is a conservative assumption since the repellency estimated from the field data is the weighted average over indoor and outdoor repellency and thus may be low (or even non-existent) even though many mosquitoes may have been affected by the volatile transfluthrin while trying to enter the house, and may have been killed or disarmed. More data would be needed to investigate how repelling, killing and disarming effects depend on the transfluthrin concentration. Both the spatial repellent and the push-pull system showed no effect on An. arabiensis in the field trial [Fillinger et al., 2023]. However, full posterior estimates of the effect on all species were used for the simulation, including possible draws with negative repellency for An. arabiensis, which led to large uncertainty in the estimates.
Large uncertainty is introduced by the open question of whether the spatial repellent primarily kills or disarms the mosquitoes which were shown to stop host-seeking due to the repellent in semi-field experiments [Denz et al., 2021]. In contrast to killing, disarming only has a limited impact on malaria transmission since disarmed mosquitoes are assumed to rest and are then subjected to low mortality. In our model, mortality varies over the stages of the feeding cycle but is not age-dependent; hence mosquitoes that were disarmed are potentially getting older, and in case they were already infected before getting disarmed, this may even increase the sporozoite rate in the population, thus offsetting the positive effects of disarming leading to reduced biting. Semi-field and field studies with experimental huts, or other means to recapture dead and resting mosquitoes, are needed to better distinguish between killing and disarming effects of the transfluthrin-treated eave ribbon. Also, larger-scale field studies on the effect of this intervention on the age structure and sporozoite rate of the mosquito populations are needed to better inform our model.
The simulation study estimated only a small impact of the Suna trap on malaria incidence in the given configuration, both as part of the push-pull system and when deployed alone. This is in contrast to the strong reduction in malaria prevalence during a mass-trapping campaign on Rusinga island, Western Kenya [Homan et al., 2016]. Also modelling studies suggested a strong transmission impact of odour-baited traps [Okumu et al., 2010]; however, these were based on assumptions on the trapping efficiency that were much higher than the values estimated from our field data. This discrepancy may be because in our field study, the traps were placed further from the houses to be outside of the area protected by the spatial repellent, while a positioning close to the house (as in the Rusinga study [Homan et al., 2016]) may increase the trapping efficiency [Fillinger et al., 2023]. However, it was estimated that the Suna trap slightly shifted indoor biting to outdoor biting in the field trial, and this effect was incorporated in the simulations, even though the trapping effect was modelled as being independent of the people given a Suna trap. Interestingly, the Suna trap showed a moderate reduction in malaria incidence in settings with high ITN coverage vs a marginal reduction in settings with lower ITN coverage, possibly suggesting some synergism between these two interventions.
Mosquito control by mass trapping still seems an auspicious strategy due to its insecticide-free mode of action with minimal risk of resistance building, but the search for the right tool and the right configuration needs to continue. Especially the placement of the trap relative to the house, as indicated by a semi-field study [Mmbando et al., 2019], a practical but sufficient supply of carbon dioxide [van Loon et al., 2015], and an improved trapping efficiency [Cribellier et al., 2018, Cribellier et al., 2020] may provide room for improvement. To investigate the impact of the studied interventions in near-elimination settings, where traps may serve as both intervention and surveillance tools, further modelling studies are required.
The impact of the transfluthrin-treated eave ribbon estimated in this study seems especially interesting in view of the expected low unit costs of the product, the possibility of local manufacturing and the minimal technical effort for installation. In contrast, the small impact of the trap in the given configuration do not justify the increased costs and technical requirements of their large-scale implementation. However, a detailed cost-effectiveness analysis of all considered candidate tools and for different settings should be carried out. Further studies on mosquito dispersion in the environment in response to large-scale implementation of the transfluthrin-treated eave ribbon are needed to fully characterise its effect. Based on this, spatial modelling may help optimise targeted implementation strategies. Finally, cluster-randomised controlled trials are needed to confirm the impact of the spatial repellent on clinical malaria and to rule out possible negative effects for inhabitants of unprotected houses.
4 Conclusion
This modelling study suggests that the transfluthrin-treated eave ribbon (spatial repellent) may substantially decrease clinical malaria in settings with lowtransmission or low ITN coverage, especially in regions where An. funestus dominates malaria transmission and primarily uses human hosts. People not covered by ITNs especially benefit from a strong reduction in case incidence when receiving protection by the spatial repellent, opening up promising intervention strategies for settings difficult to reach with ITNs and IRS. However, the spatial repellent reduced the killing effectiveness of the ITN and potentially increased risk for humans still not covered by either the spatial repellent or the ITN. The Suna trap had no significant impact on clinical malaria in the given setup and the push-pull system provided no significant advantage over the spatial repellent. These findings imply that although universal coverage with ITN needs to remain the major goal of malaria vector control, transfluthrin-treated eave ribbons can play an important role in reducing the malaria burden.
5 Methods
5.1 Simulation of clinical outcomes with OpenMalaria
OpenMalaria is an open source simulation platform for malaria [Smith et al., 2008] that combines a deterministic malaria in mosquito model [Chitnis et al., 2008, Chitnis et al., 2012] with stochastic, individual-based simulations of malaria in humans [Smith et al., 2006], including within-host-dynamics and many factors of malaria epidemiology such as demography, human-targeting interventions and health systems. OpenMalaria models the dynamics of malaria in mosquitoes for multiple species with different characteristics, taking into account seasonality, deployment as well as decay of vector control interventions, and infectiousness of humans to mosquitoes. The mosquito feeding cycle contains five stages allowing for inclusion of differential intervention effects, but does not distinguish between indoor and outdoor biting explicitly (for more details see [Denz et al., 2021]). Hence, this study assumes that indoor vs. outdoor biting is a daily behavioural choice of each single mosquito according to a certain species-specific ratio and that infectiousness of mosquitoes is homogeneous within each species. OpenMalaria was calibrated and tested against multiple data sets [Smith et al., 2012] and the source code is freely available [OpenMalariaDevelopmentTeam, 2020]. All simulations are run with a human population of 10.000 people.
5.2 Simulation setup
Figure 9 shows the simulation setup. Two epidemiological settings corresponding to two locations in East Africa are parameterised in OpenMalaria. For each location, a range of entomological and epidemiological parameters are varied for sensitivity analysis, resulting in 20 distinct baseline settings. For each of these settings, transfluthrin-treated eave ribbons (spatial repellent), the Suna trap and the push-pull system are deployed under a range of different coverage levels and different deployment scenarios, resulting in 72 distinct intervention scenarios plus a control scenario per baseline setting. For each coupling of a baseline setting with an intervention scenario, referred to as an ‘experiment’ here, 20 OpenMalaria simulations are run, each with a different random seed to initialise the simulator and a different parameter value drawn from the posteriors of the intervention parameters as estimated from the data.
Experimental setup. The seed determines both the seed to initialise the random number generator for the OpenMalaria simulation and the draw from the posterior of the intervention parameterisation. Note that the annual EIR specified here determines the transmission intensity before deployment of ITNs.
* An annual EIR of 50 is only simulated for Ahero.
As shown in Figure 10, all simulations start in the year 1916 with a warmup phase over one maximal human lifetime, which is used to ensure the entire human population has a level of natural immunity to malaria consistent with a lifetime of exposure. From 2006 onwards, ITNs are distributed in a three-year cycle through mass campaigns as a baseline for all scenarios, corresponding to the scale up of ITNs in the African region starting approximately at this time. In 2021, the interventions under consideration are deployed and maintained at constant effectiveness until 2024, when the simulations are terminated. For each simulation, a range of different epidemiological outcomes is reported monthly.
Timeline of OpenMalaria simulations
5.3 Locations / baseline settings
Two geographic settings are considered, Ahero in western Kenya and the Kilombero Valley in south-eastern Tanzania, differing in their mosquito bionomics, health systems and demographics, as summarised in Table 1. Ahero and Kilo-mbero are both considered high-transmission areas, with An. funestus dominating transmission with a contribution to the entomological inoculation rate (EIR) of about two thirds and three quarters, respectively, while the rest of the transmission is due to An. arabiensis. In Ahero An. funestus bites mainly indoors and An. arabiensis is opportunistic, while in Kilombero An. funestus is opportunistic and An. arabiensis bites mainly outdoors [Finda et al., 2018]. Other Anopheles species, including An. gambiae ss, were reported as not abundant or only marginally abundant (An. coustani) in the two studied locations [Degefa et al., 2017, Kaindoa et al., 2017] and thus were neglected. For further assumptions on mosquito bionomics and the seasonality with which the EIR varies over the year, see the Supplementary information. The effective coverage (E14) of case management (CM) with artemisinin-based combination therapy was set to 37.8% for Ahero and to 46.9% for the Kilombero valley, according to national estimates from [Galactionova et al., 2015]. The age structure of the human population was based on previous parameterisations of Western Kenya [Stuckey et al., 2012] and the Kilombero Valley [Briët and Chitnis, 2013].
5.4 Sensitivity Analysis per baseline setting
Transmission intensity
For each location, mid-range and high transmission intensities were simulated, corresponding to entomological inoculation rates (EIR) per person per year of 100 and 200, respectively, measured before implementation of ITNs, but in the presence of CM. For Ahero, a low transmission setting with an annual EIR of 50 (before ITNs) was also simulated, corresponding to highland settings in Western Kenya. The annual EIR levels are attributed to the different species according to the species composition per setting given in Table 1 and monthly EIR levels are computed according to the seasonality pattern by OpenMalaria.
Baseline vector-control interventions
Insecticide treated nets (ITNs) are distributed by mass campaigns in 3 year cycles from 2006 onwards. To cover variation in ITN coverage, settings with 60% and 80% coverage were simulated for both locations. ITNs only act on mosquitoes biting indoors and the ITN effectiveness was therefore modulated according to the indoor biting proportion for each species and each setting (see Table 1). For details on the ITN parameterisation in OpenMalaria see the Supplementary information.
Case-Management / health system
In addition to the estimates of the effective case management (CM) coverage for each setting (see Table 1), an increased effective coverage of 60% is simulated for both settings. The EIR value for a given setting corresponds to the state while CM is in place. Hence, two baseline settings with the same transmission intensity and different CM coverage levels must be seen as two distinct entomological settings with the same transmission intensity resulting from different CM coverage levels, and not as a single entomological setting with two CM strategies.
5.5 Intervention scenarios
On top of each baseline setting, transfluthrin-treated eave ribbons, the Suna trap and the push-pull system are deployed separately, with coverage levels of 20, 40, 60 and 80%. No combinations of the transfluthrin-treated eave ribbon, the Suna trap and/or the push-pull system are simulated, and the push-pull system is implemented as a separate intervention rather than a combination of its two components. In contrast to the ITN, no decay of the effectiveness of the transfluthrin-treated eave ribbon or the odour-baited suna trap was assumed, equivalent to assuming these tools are replaced as necessary. However, interventions vanish with the death of a simulated human to whom they were assigned and are not replaced.
For each coverage with the transfluthrin-treated eave ribbon or the push-pull system, three scenarios are considered for the distribution of the new intervention with respect to the baseline intervention of ITNs: ‘minimal overlap’ where the new intervention is first given to people without ITN; ‘random allocation’ where each person has the same probability of getting the new intervention regardless of whether he/she has an ITN and ‘maximal overlap’ where the new intervention is first distributed to people with ITNs. For the ‘random allocation’ scenario, incidence per cohort given none, one or both of the ITN and the new intervention is analysed. For each scenario, at least 400 individuals percohort were simulated in each simulation. For the transfluthrin-treated eave ribbon and the push-pull system, intervention scenarios are further divided by two potential modes of action: whether they kill mosquitoes or disarm them for two days, as described in detail in Section 6.10.
5.6 Data sources for intervention parameterisation
Both semi-field and field data, as summarised in Table 2, were used to parameterise the interventions.
5.7 Modelling the transfluthrin-treated eave ribbon
The transfluthrin-treated eave ribbon is assumed to act at three stages of the mosquito feeding cycle: disarming or killing before encountering a host, repellency, and killing after biting (post-prandial killing). For definitions of the effects see Table 3. The volatile transfluthrin may disarm, which is defined as stopping host-seeking for at least one day, or kill a mosquito before it encounters a protected host. Therefore, the killing and disarming effects are not attributed to single houses or hosts and cannot be estimated from the data of the field trial that measured mosquito densities per house for a small number of houses. However, from the semi-field data the combined disarming and killing effect can be estimated. Hence, the killing/disarming effect (for definition see Table 3) on An. arabiensis estimated from the semi-field data [Denz et al., 2021] was adjusted to the field conditions via the ratio of the deterrence effects estimated from these two data sources. For details on the estimation procedure see the Supplementary information. As both killing and disarming make mosquitoes unresponsive to host-seeking cues for the rest of the night, it is not possible to distinguish these two modes of action in semi-field experiments running over one night only [Denz et al., 2021]. Therefore, separate scenarios for disarming and killing were run as described in Section 6.10. Disarmed mosquitoes are modelled to rest for the duration of one feeding cycle (2 days), while being subject to the mortality rate of the resting state of the feeding cycle, and then restart host-seeking with no further impairments. Repellency partly prevents mosquitoes from encountering a protected human by reducing its availability to mosquitoes. Repellency is estimated from the field data as described in the Supplementary information. Exposure to the transfluthrin may kill mosquitoes up to 24 hours after feeding on a protected human [Denz et al., 2021]. The killing effect after biting effect (postprandial killing effect) (for definition see Table 3) was set to the estimate from the semi-field data on An. arabiensis [Denz et al., 2021], for both species. Figures of the posteriors and the drawn subsamples of all parameters are contained in the Supplementary information.
5.8 Modelling the Suna trap
Traps are modelled as dummy hosts, without the ability to contract or transmit malaria, which kill all mosquitoes encountering them. The Suna trap is deployed on a household level while assuming a household size of 5 humans and coverage by the Suna trap intervention is therefore given as the number of Suna traps divided by the number of protected households. Traps are parameterised in OpenMalaria by their relative availability to mosquitoes compared to an adult, unprotected human, with the trap availability comprising both the attraction to mosquitoes and the ability to catch them. The relative availability of the Suna trap with respect to the combined indoor and outdoor availability of an unprotected human could not be estimated directly from the field data as all humans were protected by a bed net indoors. Therefore, the relative availability of the Suna trap was estimated with respect to outdoor biting on an unprotected human only, for each species. The relative availability with respect to outdoor biting is then multiplied with the percentage of outdoor biting, for each species and each setting (1 − πi), which yields the relative availability with respect to the combined indoor and outdoor availability of an unprotected human. For details of the estimation procedure see the Supplementary information. It was estimated from the field data [Fillinger et al., 2023] that the Suna trap diverted indoor to outdoor host-search, which was accounted for by altering the πi value. Figures of the posteriors and the drawn subsamples of all parameters are contained in the Supplementary information.
5.9 Modelling the push-pull system
The repelling, killing/disarming and postprandial killing effects of the push-pull system were estimated and implemented analogously to the transfluthrin-treated eave ribbon from the respective data. In the semi-field experiments, used to estimate the disarming/killing effect, the Suna trap was baited with the human odour mimic MB5 and carbon dioxide produced by molasses fermentation, while in the field experiment, used to estimate the other intervention effects, the Suna trap was baited with MB5 and 2-butanone. The Suna trap of the pushpull system was implemented analogously to the Suna trap intervention, with relative availability estimated from the push-pull arm of the field data. Figures of the posteriors and the drawn subsamples of all parameters are contained in the Supplementary information.
5.10 Disarming vs. killing
Disarming and killing mosquitoes have very different impacts on transmission, but cannot be distinguished in the available semi-field or field data. Therefore, a scenario with only disarming and no killing as lower bound of the effect, and a scenario with only killing and no disarming as upper bound of the effect is run for each intervention scenario involving the transfluthrin-treated eave ribbon. Due to technical constraints in OpenMalaria, disarming is implemented as disarming for a duration equal to the length of the feeding cycle, which was set to 2 days. Disarming for 1 day and for 3 days showed very similar impact on relative reduction of vectorial capacity of An. arabiensis when relying on the semi-field parameterisation (no significant difference with respect to the uncertainty from the semi-field inference) [Denz et al., 2021].
5.11 Modelling the interaction of the transfluthrin-treated eave ribbon with the ITN
ITNs only affect mosquitoes biting indoors and their effectiveness therefore depends on the indoor mosquito density (see the Supplementary information for how the πi value is factored into the ITN parameterisation). Both the transfluthrin-treated eave ribbon and the Suna trap may prevent house entry of mosquitoes and therefore interact with the ITN by modulating the indoor-outdoor biting ratio (πi value). Hence, the proportional change of indoor biting (for a definition see Table 3) due to these interventions was estimated for each species from the field data (see the Supplementary information for details). For houses where no ITN is present, altering the indoor-outdoor biting ratio (πi value) has no effect on transmission in the present model. Both the repellency and the killing after biting effect by the ITN are multiplied with the corresponding effect by the spatial repellent or the push-pull system.
5.12 Uncertainty analysis for intervention parameterisation
All intervention parameters were estimated from semi-field and field data with a Bayesian framework yielding posteriors. The mean effect sizes corresponding to the estimate for an average house from the field analysis [Fillinger et al., 2023] was used for the intervention parameterisation, since OpenMalaria currently provides no means to introduce heterogeneity of the effect of the given interventions within a single simulation. For each parameter, a sample of 20 values was drawn from the corresponding posterior and each was coupled with one random seed used to run OpenMalaria simulations in a fixed order.
5.13 Analysis and visualisation of simulated data
Processing the input to OpenMalaria is performed with R [RCoreTeam, 2018]. Postprocessing of the simulated data and visualisation is performed with MATLAB [MATLAB, 2018].
6 Declarations
6.1 Competing interests
The authors declare no conflict of interest.
6.2 Funding
This work was supported and funded by the Innovative Vector Control Consortium (IVCC). NC and AD were supported by the Swiss National Science Foundation (SNF grant number 163473). NC acknowledges funding from the Bill and Melinda Gates Foundation under Grant #OPP1032350.
6.3 Authors’ contributions
AD, NC and TS developed the modelling methodology. AD performed the Bayesian inference. AD ran the simulations and created the figures. AD and NC interpreted the results. AD wrote the initial draft of the manuscript. AD and NC contributed to writing the manuscript.
Data Availability
All data produced in the present study are available upon reasonable request to the authors.
7 Supplementary information
7.1 Details on Methods
7.1.1 Estimation of intervention parameters from field and semi-field data
Note on differences between field and semi-field data
It is not possible to use the semi-field model [Denz et al., 2021] for the time-stratified HLC data for the field. The main reason is that the time-pattern of the mosquito HLC response in the field is driven by preferred biting times as well as environmental factors and not by mortality and successful landing (as opposed to semi-field where mosquitoes start host seeking readily because of starving and where environmental factors are well controlled). And even if the baseline time pattern of the mosquito HLC response in the field was known and the number of mosquitoes in the system could be assumed constant, no disarming or killing could be detected as the number of mosquitoes ‘in the system’ is much bigger than the number of mosquitoes affected by the intervention.
Analysis of field data
To estimate parameters from the field data, the Bayesian hierarchical models described in [Fillinger et al., 2023] were re-parameterised coupling indoor and outdoor biting so that all parameters are estimated with one single joint posterior distribution, allowing for sample-based computations of further parameter transformations. Hence, the following statistical model is equivalent to the one presented in the supplementary methods to [Fillinger et al., 2023] (see notation therein) and it was checked that the inference with stan [StanDevelopmentTeam, 2019] yields almost identical parameter estimates:
With the same second level model for both indoor and outdoor:
Consequently, the mean indoor and outdoor counts are given by:
and the mean indoor and outdoor effects are given by
To fully account for the experimental design, a version of the model where the intervention effects were allowed to vary by house and a version where they were allowed to vary by week were fitted separately to the data and then stacked with equal weights for the final parameter estimates.
The same model was used to analyse the Suna trap catch counts in relation to outdoor biting by replacing the indoor counts with the Suna trap catch counts.
Repellency
The repellency effect on total mosquito host encounters for each species, comprising both indoor and outdoor landing, is given as the relative reduction of total mosquito counts (protective efficacy) being the sum of the above mean indoor and outdoor counts, yielding a formula based on the sum over indoor and outdoor effect, weighted by the mean control indoor and outdoor counts, respectively:
The subscript I denotes the spatial repellent or the push-pull intervention and the subscript C denotes the control. The posteriors of the repellency estimates of the spatial repellent and the push-pull system by species, including the randomly drawn samples used for the simulations, are shown in Figure 11.
Posteriors of the repellency parameter for the spatial repellent (R) and the push-pull system (P) by species, and the corresponding randomly drawn samples used in the simulations.
Killing/disarming
The estimate of the killing/disarming effect of the spatial repellent and the push-pull intervention on An. arabiensis from the semifield data (κ) [Denz et al., 2021] was adjusted to the field conditions. Adjustment was made by multiplying with the ratio of the repellency (protective efficacy) estimated from the field data over the repellency estimated from the semi-field data. Note that the protective efficacy in the field trial [Fillinger et al., 2023] depends on repellency only, and is actually equal to repellency, since the killing/disarming effect induced by the interventions deployed to a couple of houses is negligible given the size of the local mosquito population. Scaling by the repellency is a conservative assumption since the repellency estimated from the field data is the weighted average over indoor and outdoor repellency and thus may be low (or even non-existent) even though repellency from indoor biting may be high. However, high indoor repellency indicates that many mosquitoes may have been affected by the volatile transfluthrin while trying to enter the house and thus may have been killed or disarmed, despite a low overall repellency. The killing/disarming effect was restricted to be positive in order to be biologically meaningful, and the estimate of the disarming effect on An. arabiensis was therefore truncated:
where κ stands for the killing/disarming parameter and π for the repellency parameter, as defined in [Denz et al., 2021]. However, the truncation came only into effect for An. arabiensis since the repellency estimated from the field data was very low for this species. The posteriors of the killing/disarming estimates of the spatial repellent and the push-pull system by species, including the randomly drawn samples used for the simulations, are shown in Figure 12.
Posteriors of the killing/disarming parameter for the spatial repellent (R) and the push-pull system (P) by species, and the corresponding randomly drawn samples used in the simulations.
Proportional change of indoor biting
The proportional change of the indoor biting (vs. outdoor) due to the spatial repellent, the trap or the pushpull system is given as the ratio of the indoor vs. outdoor biting proportion in the intervention arm (denoted with I) over the indoor vs. outdoor biting proportion in the control (denoted with C):
The corresponding posteriors, by species, and the randomly drawn samples used in the simulations are shown in Figure 13.
Posteriors of the parameter capturing the change of the indoor-outdoor biting ratio due to the spatial repellent (R), the trap (T) and the push-pull system (P) by species, and the corresponding randomly drawn samples used in the simulations.
The change of the indoor vs. outdoor biting proportion by the interventions is incorporated into the INT parameterisation via the indoor vs. total biting proportion (πi value). The indoor vs. total biting proportion under deployment of one of the interventions (πiI value) is obtained by:
where πiC denotes indoor vs. total biting proportion from the literature (see Table 1), depending on both the species and the location. The corresponding posteriors, by species and by location, and the randomly drawn samples used in the simulations are shown in Figure 14.
Posteriors of the indoor vs. total biting proportion (πi value) under deployment of the spatial repellent (R), the trap (T) and the push-pull system (P) by species and by location, and the corresponding randomly drawn samples used in the simulations.
Killing after biting
Killing after biting effect of the spatial repellent on An. arabiensis was estimated from the semi-field data as described in [Denz et al., 2021]. As no other data was available the same estimate was also used for the push-pull system, and for both species. The corresponding posterior and the randomly drawn samples used in the simulations are shown in Figure 15.
Posteriors of the killing after biting parameter for the spatial repellent (R) and the push-pull system (P) and the corresponding randomly drawn samples used in the simulations. Due to lack of data, the same estimate was used for all species and both the spatial repellent and the push-pull system.
Relative trap availability
The relative availability of the Suna trap, either deployed alone or as part of the push-pull system, as compared to an unprotected human outdoors is given by:
where I stand either for the trap or the push-pull intervention. The corresponding posteriors, by species, and the randomly drawn samples used in the simulations are shown in Figure 16.
Posteriors of the relative trap availability as compared to an unprotected human outdoors, for the trap (T) and the push-pull system (P), by species, and the corresponding randomly drawn samples used in the simulations.
This parameter is then multiplied with the outdoor vs total biting proportion of the given location to obtain the relative trap availability with respect to an unprotected human:
The corresponding posteriors, by species and by location, and the randomly drawn samples used in the simulations are shown in Figure 17.
Posteriors of the relative availability of the trap as compared to an unprotected average human for the trap (T) and the push-pull system (P) by species and by location, and the corresponding randomly drawn samples used in the simulations.
7.1.2 Implementation of intervention effects
Killing and disarming effect
The killing and disarming effects are implemented in OpenMalaria by adding a shadow host [Denz et al., 2021] for each protected human host whose availability rate is given by multiplying the mean availability rate of an unprotected human hosts, determined by the adult availability rate and the age-dependent availability function applied to the given demography, with κ. Mosquitoes that encounter a shadow host are either killed under the killing scenario, or restart host seeking 2 days later after completing a feeding cycle under the disarming scenario.
Repellency and killing after biting
The repellency and the killing after biting effect of the spatial repellent and the push-pull system are implemented in OpenMalaria through the ‘general vector control’ (GVI) intervention [Open-MalariaDevelopmentTeam, 2021].
Proportional change of the indoor vs. total biting proportion and interaction with ITNs
The interaction of the spatial repellent, the trap and the push-pull system with the ITN was implemented by altering the ITN effectiveness according to the indoor vs. total biting proportion under deployment of one of the interventions (πiI value), analogously to the exposition in Section 9.1.3 for πi.
Suna trap
The Suna trap was implemented in OpenMalaria via the ‘non-human host’ intervention [OpenMalariaDevelopmentTeam, 2021]. Since there was no functionality in OpenMalaria to set the number of non-human hosts of a given type at the time, the relative trap availability was multiplied with the given coverage level and divided by the house-hold size.
7.1.3 Baseline INT paramterisation
The deployment of ITNs-including decay of insecticide concentrations, formation of holes and attrition is implemented in OpenMalaria as explained in the Appendix to [Briët et al., 2012], and we used the parameterisation from the corresponding electronic supplementary information, which assumes uniform effects on indoor biting of all anopheles species. The insecticidal effect decays exponentially with a half time of 1.5 year, while the net itself decays in a smooth-compact manner over about 20 years.
The parameterisation of the ITN according to [Briët et al., 2012] is estimated from the outcomes of experimental hut studies. As suggested in [Briët and Schapira, 2017], we adjusted the ITN parameterisation for different indoor biting ratios (πi value) by weighting the effect of the INT on the experimental hut outcomes with πi.
7.2 Additional Results
7.2.1 Ahero
7.2.2 Kilombero
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Ahero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are randomly allocated with respect to ITN allocation. Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), with solid edges standing for the disarming and dashed edges for the killing assumption for the spatial repellent and push-pull interventions. Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the entomological inoculation rate (EIR) under different transmission settings in Ahero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Ahero with increased case management coverage at baseline, with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the number of severe malaria episodes per person per year (severe cases ppa) under different transmission settings in Ahero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are randomly allocated with respect to ITN allocation. Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Kilombero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are randomly allocated with respect to ITN allocation. Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), with solid edges standing for the disarming and dashed edges for the killing assumption for the spatial repellent and push-pull interventions. Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the entomological inoculation rate (EIR) under different transmission settings in Kilombero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the number of uncomplicated malaria episodes per person per year (cases ppa) under different transmission settings in Kilombero with increased case management coverage at baseline, with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are allocated to people regardless of their ITN ownership (random mixing). Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Estimates of the number of severe malaria episodes per person per year (severe cases ppa) under different transmission settings in Kilombero with either the spatial repellent (blue), Suna trap (red) or push-pull (purple) interventions with increasing coverage, under the assumption that the new interventions are randomly allocated with respect to ITN allocation. Lines denote the median estimate, with solid lines standing for the disarming and dashed lines for the killing assumption for the spatial repellent and push-pull interventions. The shaded areas show 50% credible intervals (equal-tailed intervals), combining the killing and disarming scenarios for the spatial repellent and push-pull interventions (lower 25% percentile of the killing scenario to upper 75% percentile of the disarming scenario). Note that the vertical axes are aligned per row, but not across all transmission settings, and that the vertical axes do not start at 0.
Acknowledgements
We thank Clara Champagne for helpful discussions and suggestions on the manuscript.
References
