A systematic, realist review of zooprophylaxis for malaria control

Integrated vector management (IVM) is recommended as a sustainable approach to malaria control. IVM consists of combining vector control methods based on scientific evidence to maximize efficacy and cost-effectiveness while minimizing negative impacts, such as insecticide resistance and environmental damage. Zooprophylaxis has been identified as a possible component of IVM as livestock may draw mosquitoes away from humans, decreasing human-vector contact and malaria transmission. It is possible, however, that livestock may actually draw mosquitoes to humans, increasing malaria transmission (zoopotentiation). The goal of this paper is to take a realist approach to a systematic review of peer-reviewed literature to understand the contexts under which zooprophylaxis or zoopotentiation occur. Three electronic databases were searched using the keywords ‘zooprophylaxis’ and ‘zoopotentiation’, and forward and backward citation tracking employed, to identify relevant articles. Only empirical, peer-reviewed articles were included. Critical appraisal was applied to articles retained for full review. Twenty empirical studies met inclusion criteria after critical appraisal. A range of experimental and observational study designs were reported. Outcome measures included human malaria infection and mosquito feeding behaviour. Two key factors were consistently associated with zooprophylaxis and zoopotentiation: the characteristics of the local mosquito vector, and the location of livestock relative to human sleeping quarters. These associations were modified by the use of bed nets and socio-economic factors. This review suggests that malaria risk is reduced (zooprophylaxis) in areas where predominant mosquito species do not prefer human hosts, where livestock are kept at a distance from human sleeping quarters at night, and where mosquito nets or other protective measures are used. Zoopotentiation occurs where livestock are housed within or near human sleeping quarters at night and where mosquito species prefer human hosts. The evidence suggests that zooprophylaxis could be part of an effective strategy to reduce malaria transmission under specific ecological and geographical conditions. The current scientific evidence base is inconclusive on understanding the role of socio-economic factors, optimal distance between livestock and human sleeping quarters, and the effect of animal species and number on zooprophylaxis.


Background
Despite renewed commitments and control efforts in recent years [1][2][3] malaria continues to be a major contributor to global health burden, with approximately 165 million cases in 2013 [4]. Integrated vector management (IVM) has been promoted as a sustainable approach to combat malaria [5,6] in the face of increasing insecticide resistance of malaria vectors, and environmental and health concerns [5,7]. This strategy involves combining chemical and non-chemical interventions targeted to specific ecological settings in a way that maximizes efficacy while minimizing cost and negative environmental impacts [5]. IVM makes use of environmental modification, environmental manipulation, chemical control methods, and biological methods [5] ( Table 1).
Strategic placement of livestock sheds or pens has also been proposed as a component of IVM to reduce contact between vectors and human hosts [8,9]. The World Health Organization (WHO) began recommending this type of intervention in 1982 as a method to divert mosquitoes from human populations [10]. This purposeful use of livestock (i.e. as dead-end hosts) to divert mosquitoes away from humans is described as active zooprophylaxis. Passive zooprophylaxis occurs where normal presence of livestock draws mosquitoes away from humans [11]. Insecticide zooprophylaxis, more commonly described in tsetse fly control, involves the use of insecticide-treated cattle and has also been investigated for the control of malaria vectors [12][13][14][15].
There remains considerable debate regarding the efficacy of zooprophylaxis [10,11,[16][17][18][19][20]. In addition to the literature supporting zooprophylaxis [21,22], there is evidence that supports zoopotentiation; livestock presence may actually increase malaria transmission by creating additional blood meal sources, which, in turn, can increase vector lifespan and population density [10,11,16]. Due to the divergent nature of the literature and the complexity of the relationship between livestock and malaria prevalence, there has been a reluctance to employ zooprophylaxis in control programmes [8,23,24].
The goal of this paper was to characterize and critically assess the potential for zooprophylaxis to reduce malaria transmission, with specific attention paid to the contexts under which it may be an effective component of IVM. The strategic framework for IVM calls for evidencebased decision-making in the selection of appropriate interventions that acknowledge the local context, including vector ecology, epidemiology and socio-economic factors [5].

Methods
A modified systematic review methodology, employing realist approaches [25,26] was applied to the self-identifying zooprophylaxis literature. This approach recognizes a priori that the scientific literature in this area is conflicting and in this case focuses on when, why, and in what contexts zooprophylaxis or zoopotentiation may occur. A meta-analysis, was not feasible due to the variety of study designs (including both observational and experimental designs) and outcome measures employed in this research area. ISI Web of knowledge, CAB Direct and PubMed databases were searched in December 2014 using the keywords 'zooprophylaxis' and 'zoopotentiation' . While this invariably excluded studies of malaria risk factors that consider the presence of animals, but did not self-identify using the terms 'zoopotentiation' or 'zooprophylaxis' , the search was limited to these explicit terms for two reasons: (1) to select a proxy sample of the key literature explicitly emphasizing and investigating zooprophylaxis, more likely to provide direct discussion, consideration of causal pathways of association and depth regarding the role of animals in malaria transmission; and, (2) to limit the number of results to a feasible and directly relevant sample for in-depth realist analysis. This search retrieved 75 documents after removal of duplicates. Only empirical, peer-reviewed articles that focused on either malaria infection in humans or mosquito behaviour associated with livestock presence were reviewed. Mathematical models of mosquito behaviour and review articles were excluded from the synthesis but their content was assessed to provide context for interpretation of results (Table 2). A total of 20 articles met final inclusion criteria and were retained for critical appraisal after full article review (Fig. 1). Forward and backward citation tracking were applied to the articles selected for critical appraisal with one additional relevant article identified.
Data extraction from each article included author, date of publication, study location, livestock exposure, malaria risk outcome measures, study design, and study limitations. Published results reporting significant associations at the 95% confidence level were classified as supporting a significant zooprophylaxis or zoopotentiation effect. Critical appraisal [27] resulted in the exclusion of 14 articles. Reasons for exclusion (Additional file 1) were related to data analysis such as a lack of evidence of statistical significance [28,29], and pooling of data preventing conclusions from being made on the effect of livestock on malaria risk [30]. Others were excluded based on study design issues, such as the absence of a comparison group [31] and small sample sizes [32]. The pertinent results and conclusions of each study were analysed with regard to the associations between livestock and malaria risk.

Study characteristics
Twenty articles met inclusion criteria; 15 were observational studies and five were experimental (Table 3). Of the observational designs, there were 12 cross-sectional, two case-control, and one cohort design. The majority (16) of studies were conducted in sub-Saharan Africa (SSA), nine from East Africa, five from West Africa and two from Southern Africa. The remaining four studies were carried out in Pakistan (2), Bolivia (1) and Lao PDR (1). Two articles reported on a single study conducted in The Gambia, although each article reports analysis of a

Outcome measures
Four studies measured parasitaemia as an outcome. Three articles defined parasitaemia by positive identification of the parasite by thick and thin blood smears and one used a positive result on a malaria rapid diagnostic test (RDT). One study used recurrent household-level malaria infection defined as two or more infections for two or more household members over nine screening events but did not report the screening method used. Four studies reported mosquito feeding behaviour as measured by human blood index, which is the proportion of blood meals taken on a human out of the total number of blood meals taken. Five studies used mosquito abundance or mosquito presence as their outcome measure and four studies measured host attraction either by human landing catches or human-baited traps. One study reported both human blood index and mosquito abundance as outcome measures and another used both human blood index and host attractiveness by human landing catch.

Key determinants of zooprophylaxis and zoopotentiation
Two main factors were consistently associated with zooprophylaxis and zoopotentiation: the predominant vector species present, and the location of livestock relative to humans, particularly during peak feeding times. Zooprophylaxis was considered to be dependent on the relative preference of mosquitoes for animal hosts (zoophily) in seven studies. Where the predominant mosquito species prefers human to animal hosts (anthropophily), and human hosts are available, keeping livestock nearby is unlikely to result in zooprophylaxis. Relative zoophily was reported as an important predictor of zooprophylaxis in five studies where multiple mosquito species were present. For example, Anopheles gambiae sensu stricto and Anopheles funestus were generally found to be anthropophilic compared to other species such as Anopheles pharoensis and Anopheles arabiensis, which were readily deferred from humans to feeding on livestock species [11,13,22,33]. In some cases, An. arabiensis were found to be opportunistic in their host choices, or were anthropophilic but exophagic (prefer to feed outdoors) and therefore would feed on animals if no humans were found outdoors [34]. Many of the entomological studies [11,12,33,34] collected only indoor resting mosquitoes for the assessment of blood meals, which may bias samples towards endophilic (indoor resting) and endophagic (indoor feeding) species, which often tend to be anthropophilic [24]. Mosquito species were not identified in the five studies that measured human malaria infection as the outcome.
Fourteen studies found that proximity or location of livestock relative to humans influenced malaria risk. When animals were housed inside at night, or in close proximity to sleeping rooms, malaria risk increased [11,16,23,24,33,35,36]. In contrast, when livestock were housed in separate shelters some distance away, malaria risk decreased [22,34,37]. However, some studies failed to find an association between location of livestock and zooprophylaxis or zoopotentiation. For example, in Lao PDR, owning a cow doubled the risk of mosquito house entry but keeping livestock near or underneath the house at night had no effect [38]. Similarly, a cohort study in The Gambia examined parasite prevalence in children sleeping within 20 m of the nearest cow compared to children sleeping at least 50 m from the nearest cow. No difference could be found in parasite prevalence between the groups when socio-economic factors were taken into account. It should be noted, however, that other livestock, such as goats, donkeys and horses, were commonly found in participating households but were not included in the analysis [10]. While no study specifically tested the impact of keeping livestock at varying distances on malaria risk, Maia and colleagues were unable to detect an effect of cattle at a distance of 20 m on human landing catches of mosquitoes [21].
Relative abundance of livestock to humans, or high cattle: human ratio may influence the success of zooprophylaxis [11,39]. Three studies carried out in the Rift Valley of southern Ethiopia, where An. arabiensis is the main malaria vector, examined the relationship between cattle: human ratio and malaria risk. Two of these studies found no association [22,34]. The third study did not account for the effect of humans sleeping on raised platforms in trees above cattle to avoid mosquito bites (with high cattle: human ratio) compared to the other two sites where humans slept in traditional dwellings (with lower cattle: human ratio) [12].

Influence of modifying variables
Two contextual factors were shown to modify the association between malaria risk and livestock: the use of bed nets and socio-economic status. The use of bed nets seems to be an effect modifying factor, preventing even highly anthropophilic species from feeding on humans, forcing them to feed on livestock as an alternative [33]. While two studies found that bed nets had no impact on malaria infection [40], or mosquito house entry [38] and another reported that pig ownership remained a significant risk factor for positive RDT when bed nets were accounted for [41], six studies reported a relationship between bed nets and zooprophylaxis [8,10,11,33,42,43]. In two of these studies, the effect of zooprophylaxis was diminished or became non-significant when bed net use was controlled for [10,43]. Iwashita et al. reported that bed nets dramatically reduced human blood feeding in the presence of livestock [33]. A study conducted within a rice irrigation scheme in Kenya suggested that the cause of lower prevalence of malaria in villages where irrigation took place (and where prevalence was expected to be high) was a result of preferential feeding on livestock [8]. Bed net use was not measured in this study. Other work in the same location has suggested that bed net usage is promoted heavily in irrigated areas where malaria risk is known to be high [42].
Socio-economic status, measured as wealth or asset ownership was considered in four studies [10,23,38,41]. One study identified a decrease in malaria prevalence with animal ownership, but controlling for wealth removed the effect of zooprophylaxis [10]. This study used a financial index based on livestock value to measure wealth and, therefore, collinearity might be expected between the presence of livestock and wealth. A second study noted that, in univariable analysis, sheep keeping was associated with decreased odds of infection with malaria while pig keeping was associated with increased odds of infection. When wealth was accounted for, the association with sheep ownership was no longer statistically significant while the relationship with pig ownership persisted [41]. Ghebreyesus et al. included household radio ownership in multivariable analysis and found that livestock sleeping inside the house increased incidence of infection in children [23]. Hiscox et al. did not find that household television ownership was significantly associated with mosquito house entry in univariable analysis, and it was therefore not included in multivariable analysis [38]. Yamamoto et al. controlled for maternal education level, a robust and commonly used measure of socio-economic status [44], and found that the protective effect of donkeys, rabbits and pigs was removed when level of education and bed net use were controlled for [43]. These studies and others [8] emphasized the strong association between measures of socio-economic status and malaria risk. This important association can confound the relationship between animal ownership and malaria prevalence given that animal ownership is a reflection of social standing. Socioeconomic status is likely an important unmeasured confounder affecting zooprophylaxis in the scientific evidence base.

Discussion
This systematic realist review points to three key findings regarding the context under which zooprophylaxis may be utilized as a component of IVM. First, zooprophylaxis is most likely to be effective when the mosquito species present do not have a strong preference for human hosts. Second, in order to take advantage of mosquito preference for animals, animals must be kept out of human sleeping quarters at night. There is evidence that even in the context of mosquito species with preference for animal hosts, close proximity to humans at night may result in zoopotentiation. Third, where bed nets are used, mosquitoes are more likely to feed on animal hosts as an alternative.
Proximity of livestock to humans at night has been identified as an important factor in zooprophylaxis [45]. What remains unclear is the appropriate distance at which livestock should be kept in order to promote zooprophylaxis or prevent zoopotentiation. It is also unknown whether this distance differs between regions, species and contexts. Incidence rates of Plasmodium vivax were reduced in Sri Lankan households where cattle sheds were located within 70 m of the home when wealth, bed nets and other protective measures were considered; however, this effect was weak (RR 0.70, 95% CI 0.47-1.03) [46]. Current evidence supports the exclusion of animals from human dwellings at night, particularly where mosquito species are zoophilic. Improved estimation and precision around appropriate livestock proximities would benefit from the inclusion of livestock species, their number and location, and use of bed nets or other malaria prophylaxis in future studies.
Mosquito species characteristics were also identified as a key predictor of zooprophylaxis and zoopotentiation. Highly anthropophilic species were generally unaffected by the presence or absence of livestock whereas zoophilic and opportunistic species may be deterred from humans in the presence of alternative hosts. This is consistent with a model by Saul predicting that for vectors with a low human biting index, an increase in animal host density can significantly decrease disease transmission, while the same did not hold for weakly zoophilic species [18]. Similarly, Franco et al. predict that in the presence of moderately zoophilic vectors, such as An. arabiensis, the introduction of livestock would increase malaria transmission except in two cases: (1) where vector carrying capacity has already been reached in the system and the addition of livestock hosts does not increase vector density; and, (2) where livestock density and availability are so great as to counteract the effect of increased vector density associated with the introduction of livestock [47].
With regard to the impact of bed nets, since the rate of disease transmission is dependent upon host species interaction, any intervention that decreases contact between host and vector will decrease the risk of infection [5]. This has been corroborated by mathematical transmission models which find that while increased cattle density can decrease malaria transmission when sufficient animals are present and are housed separately, the most successful reduction transmission occurs when personal protective measures are also employed [18,39]. Where accessibility of humans relative to animals is decreased, it is predicted that malaria prevalence and number of bites will decrease [18,47]. Time of biting and human behaviour may also have an impact on the effectiveness of bed nets. If people are outdoors during peak biting times, bed nets will not provide protection against mosquito bites [21].
Socio-economic factors may be important unmeasured confounders in studies of zooprophylaxis. Risk factors for malaria are related to poverty through limited access to preventative measures such as bed nets, screened windows, closed roofing, and adequate health care [48]. Livestock ownership is also associated with increased socio-economic status, especially among the rural poor [49][50][51]. It has been suggested that in addition to zooprophylactic effects, livestock may be a confounder for reduced malaria risk as those who own livestock may also be able to afford preventative and treatment measures [10] or have better overall health and nutritional status [51,52]. Households keeping animals indoors at night may represent those who are financially restricted from providing alternative livestock shelters, further complicating the inter-relationships between wealth and animal ownership in malaria transmission.

Conclusions
There is scientific evidence to support zooprophylaxis where the dominant vector is highly zoophilic and livestock are kept away from human sleeping quarters during peak vector activity. The use of protection such as bed nets may be complementary, and would be expected to reduce the measured effect of zooprophylaxis in empirical studies. Where vector preference is mixed, varied or unknown, or where the appropriate distance of livestock from sleeping quarters is in debate, there is insufficient evidence to support the use of zooprophylaxis, and some evidence to suggest the possibility of zoopotentiation. Research in three priority areas is required for clearer evidence of contexts to maximize the likelihood of zooprophylaxis and minimize the likelihood of zoopotentiation: (1) estimation of the distance threshold and conditions whereby processes of zoopotentiation transition to zooprophylaxis for specific livestock host and mosquito vector species combinations; (2) consideration of the preference of species to feed indoors versus outdoors in entomologic studies in order to accurately assess mosquito host preferences; and, (3) inclusion of socio-economic factors and the use of other prophylactic measures as key covariates in empirical research assessing zooprophylaxis and zoopotentiation. These research priorities may aid in the development of guidelines for the use of zooprophylaxis as a malaria control intervention for agricultural extension agencies who may wish to make livestock management recommendations, such as the optimal placement of livestock shelters with respect to human sleeping quarters. Zooprophylaxis has the potential to contribute to IVM strategies due to its non-chemical nature, optimal combination with bed nets, potential social desirability, and minimal financial requirements. It will require interdisciplinary collaboration between agricultural extension officers, veterinarians and health care professionals with ongoing monitoring of efficacy.