Models of effectiveness of interventions against malaria transmitted by Anopheles albimanus

Anopheles albimanus is the main vector in malaria foci in Central America and the Caribbean, in particular in Haiti [1, 2], which accounts for most of the Plasmodium falciparum transmission in the Caribbean. Anopheles albimanus has very different characteristics from those of the Anopheles gambiae species complex and Anopheles funestus group, which comprise the main malaria vectors in Africa, and which have been most intensively studied. Anopheles albimanus has a higher intrinsic mortality rate (Additional file 1), is much more zoophilic and is much less likely to rest in houses [1, 3]. Also, importantly, it has a tendency towards crepuscular biting, although this may vary extensively depending on the location [1, 3]. Due to these differences, the impacts of different interventions on malaria transmitted by An. albimanus are likely to differ substantially from those estimated from trials in Africa, and the ranking of vector control interventions in terms of effectiveness may also be different. Therefore, the prioritization of vector control interventions in Central America and the Caribbean, in particular for eliminating P. falciparum from Haiti, should take local vector bionomic characteristics into account.

Malaria transmission in Central America and the Caribbean is low, with 4.6 reported cases (presumed and confirmed) per 1000 population at risk, compared to Africa with 189.6 cases per 1000 [4]. Effects of naturally acquired immunity on transmission are therefore likely to be negligible, and the effect of vector control interventions on transmission may be examined in terms of the effect on the vectorial capacity. Similarly, the effect of interventions against malaria infections in humans, such as chemo-prevention, might be studied in terms of the effects on the reproduction number in simple susceptible-infected-susceptible (SIS) models.

This study used data from field observations in Haiti and data extracted from published literature to parameterize mathematical models of malaria transmitted by An. albimanus, and the impact of interventions against it. As well, impacts were compared to that of these interventions against An. gambiae. The results are presented in terms of the effect of varying the coverage of different interventions.

The biting rhythm of An. albimanus, derived from the human landing catch data recently collected in Dame Marie (Fig. 2a) indicates that most of the exposure to bites by this species is around dusk and dawn. The data used for the human activity rhythm (Fig. 2c) indicate substantial variability in the time that people are in bed. When these data are combined with the An. albimanus biting rhythm data to calculate exposure by hour (Fig. 2e), most of the exposure is seen to occur in the early evening and late morning, irrespective of whether bed nets are used. Under the assumption that people are outdoors before bedtime (and after waking up) the proportion of mosquito bites that would be prevented by ITNs (π_i) for the Dame Marie location was 0.49, while for the extreme early biting rhythm observed by Taylor [14] in northern and central Haiti (see Fig. 1 in Additional file 1), the π_i was only 0.08. In the Laborde site, where Desenfant [15] observed the biting peak to occur at the time when most people are asleep (see Fig. 1 in Additional File 1), the π_i was 0.75, still below to the π_i of 0.95 observed for An gambiae in Tanzania (Fig. 2f), which bites during the middle of the night when most people are in bed (Fig. 2b and d). With the alternative assumption that people are indoors when not in bed, the π_i value was 0.57 for the Dame Marie location, 0.23 for northern and central Haiti and 0.79 in Laborde, while being 0.90 for Tanzania.

Due to these and other differences in the bionomics of An. albimanus and An. gambiae, the predicted impact of ITNs on the vectorial capacity was low in Haiti compared to Tanzania (Fig. 3 and Table 3), irrespective of the assumption about where people are when not in bed. If it was assumed that people are outdoors until they go to bed, a similar impact was predicted for IRS with DDT or bendiocarb as for ITNs (Figs. 4, 5 and Table 3). However, if it was assumed that people are indoors by the time that An. albimanus commences biting, the impact of IRS was much stronger than that of ITNs. The impact of each of these vector control interventions against An. albimanus was close to linear in coverage (Fig. 5), while for An gambiae more curvature was present. The vector control intervention having the highest impact on An. albimanus at a given coverage was LSM. The impact of LSM on the vectorial capacity was numerically equal to the coverage, as measured by the proportionate reduction in the emergence rate. House-screening, and hence spatial repellents, attained similar effects to LSM at the same coverage under the assumption that people are indoors by the time that An. albimanus bites, but the effect of house screening showed some saturation at high coverages.

At equal coverage, RTS,S mass vaccination (Fig. 6) was predicted to have a lower impact than ITNs (except with the very early biting rhythm observed in northern and central Haiti by Taylor [14]), but the impact relative to IRS depended on the insecticide, the biting rhythm, and whether people are indoors or outdoors when not in bed (Table 3). The impacts of chemoprevention and chemotherapy were predicted to be non-linear in coverage (Fig. 6). A key determinant of the effectiveness of chemo-prevention strategies is the proportion of the population that is missed [27], so effectiveness increased steeply as coverage approaches 100%. For case management, the slope of the coverage-effectiveness curve also increased with coverage.

The finding that interventions with deterrent effects were relatively effective in reducing the vectorial capacity suggested that their impact might be sensitive to the availability of alternative hosts. Since the ratio of humans to animals in the local environment is obviously highly variable, so are the potential effects of zoophagy on malaria transmission. In the absence of interventions, vectorial capacity was predicted to initially strongly increase with the proportion of mosquito feeds on humans, but to saturate at high HBI levels (Fig. 7). The implications of this were investigated by analysis of the sensitivity to variations in the HBI of the proportionate reduction in transmission achieved by house screening. However, the percentage impact of house-screening was independent of the HBI.

Calculating impacts on vectorial capacity provides clear general statements about impacts of entomological interventions, without the need to estimate prevailing levels of transmission or infectiousness. This was the original motivation of Garrett-Jones when he devised the vectorial capacity in 1964 [5]. The numerical equivalence of the vectorial capacity with the transmission parameter in the SIS model provides an easy way of comparing likely impacts of vector control with those of human-side interventions, because of the equivalence of effects on vectorial capacity and on the reproduction number (because a reduction in vectorial capacity is proportional to the ensuing reduction in reproduction number).

For An. albimanus, the slopes of the curves in Table 3 provide a concise way of comparing the potential impacts of the interventions, since most of them are close to straight lines below 80% coverage. However, these need to be combined with an analysis of operations and costs involved in reaching different levels of coverage if they are to be useful for elimination planning. These will have substantial effects on which intervention mixes are most attractive, since the constraints in achieving a high coverage are different for each intervention. For instance, in this model, the indication that LSM will be the most efficient intervention may be deceptive depending on the scenario. LSM is usually synonymous with larviciding and the World Health Organization recommends larviciding for malaria control in sub-Saharan Africa only in areas where the breeding sites are few, fixed, and findable [28]. In that context, sustained high coverage of larviciding is hard to achieve, entailing thorough searches and recurrent treatment of breeding sites. Indeed, Anopheles albimanus in Haiti breeds in diverse types of habitats [1, 2, 10, 29], which means that it is unlikely that breeding sites are small in number, fixed in location, or easy to find. However, in Haiti, where elimination is the goal, it may be possible to conduct a time-limited, cost-effective LSM intervention in a small, well-defined focus of malaria transmission as a supplement to other core interventions. This will contribute to focal suppression of mosquitoes and subsequent interruption of malaria. Potentially, the development of emerging technology such as drones might facilitate the identification and control of vector breeding sites at various geographical and operational scales in a cost-effective manner [30]. Measurement of coverage as the proportionate reduction in emergence is convenient for parameterizing the model, but estimation of this in practice is likely to be very difficult.

From the model, it is implied that house-screening in Haiti is likely to be a very effective intervention wherever there is malaria transmitted by An. albimanus. House-screening entails a one-off modification of houses, and coverage is relatively easy to quantify. However, there is little information available in literature on the coverage of house screening in Haiti. The very substantial predicted impact is consistent with diverse evidence from across the world [31], dating back to the original studies of house-screening in Italy [32]. Substantial impacts of house improvement against malaria transmitted by north American Anopheles were documented in the USA in the mid-20^th century [33]. Poor-quality housing is well-recognized as a factor contributing to the vulnerability of Haiti to natural disasters, and improving houses is generally popular with their inhabitants. Potentially, improved house construction and screened windows and doors could contribute to elimination of malaria from Haiti. However, the model parameterization is highly uncertain because the review by Tusting and colleagues [31] that it is based on, is a summary of studies of diverse vectors and house architectures, and the predictions are sensitive to assumptions about human behaviour. In Tanzania, there is little difference between these extreme assumptions for An. gambiae because the times when people are indoors in Tanzania overlap substantially with those when it is biting. For Haiti, the impact predicted for house-screening is highly dependent on the assumption that people are inside whenever mosquitoes are biting. When the alternative is assumed, i.e. that people stay outside when not in bed, the impact is lower, and the degree to which it is reduced depends on the biting rhythm. Local data are needed to assess which assumption is most realistic in any particular setting. It is quite possible that people increase time inside when houses are screened, or that the time spent inside is dependent on the availability of artificial light.

Anopheles albimanus contrasts with the dominant African malaria vectors in its greater tendency to bite in the evening or morning and to feed on animals. These differences lead to substantial discrepancies between estimates of the likely impacts of interventions against malaria transmitted by this vector and systematic reviews of intervention trials [34, 35], which are largely based on data from sites with anthropophilic vectors. The estimated impacts of ITNs and IRS against An. albimanus were substantially lower than those for An. gambiae (Table 3).

The two extreme assumptions about human location (inside or outside the house) when not in bed did not result in very large differences in impact of ITNs, because ITNs were assumed to only have a protective effect when people are in bed. The predicted impact of ITNs against transmission by An. albimanus is relatively poor compared with that of a systematic review of trials of ITNs, mostly against African vectors [34]. It is also consistent with the results of a recent case–control study in Haiti [36] which showed little personal protection, but which was not able to measure the community effect.

In contrast to ITNs, the predicted impact of IRS with DDT or bendiocarb was surprisingly large, but only with the assumption that people are inside (when not in bed) whenever mosquitoes are biting. This assumption may be unreasonable. Taylor [14] writes: “Over three quarters of the biting activity takes place between 17:30 and 21:00 h, the hours during which the majority of Haitians living in the provinces are outdoors.” In contrast to ITNs, IRS was assumed to protect against bites indoors when people were not bed. This may be reasonable, given the apparent effectiveness of DDT against An. albimanus in the 1960s, in particular in Haiti [2].

The effect of DDT against An. albimanus was purely modelled as a deterrent from house entry, without any killing effect, as the study this was based on Bangs [37] who did not detect a difference in mortality between the treatment and control arms, due to high mortality in the controls.

DDT has been shown to have spatial repellent effects on An. albimanus, in contrast to other insecticides [38], and these effects account for the greater impact of DDT in the models, supporting the general conclusion that that the most effective interventions against An. albimanus are those that divert the vector from humans to other hosts. Roberts and colleagues [39] also assessed that the dominant actions of DDT residues in reducing man-vector contact inside houses were effects of repellency and irritancy, and showed that these effects were larger with the American vector Anopheles darlingi than with An. gambiae. This is similar to the conclusions of analyses of zoophilic vectors in India [40]. Whereas IRS with DDT is probably not likely to be implemented in Haiti, spatial repellents such as transfluthrin [11] could well be practicable and affordable options. Spatial repellent emanators represent a less permanent solution than house screening, but they can be deployed adaptively in the times and places when the mosquitoes are biting, and even be effective when deployed near people spending evenings outdoors. On the other hand, permanent house improvements may be popular, even where malaria is not perceived to be a major priority. Similarly, when people deploy repellents, or go inside when mosquitoes are biting, they may be responding to nuisance mosquitoes, rather than to Anopheles, but equally they may also be protecting themselves against Aedes aegypti and the viruses it can transmit [41]. Such considerations make it hard to estimate the effective coverage that might be reached.

Because Haiti is aiming for elimination, malaria control paradigms from the African context may not apply. Rather, malaria interventions should be considered with an understanding that they will be implemented in a time-limited manner until malaria is eliminated from Haiti. While this analysis aims to guide decision-making in selecting interventions, it will be important for Haiti to weigh predicted impacts by programmatic factors of ease of implementation and expected time to elimination. However, translating these results into a practical algorithm for prioritization of interventions also needs to take into account the local feasibility, and availability of different resources.