The entomological inoculation rate (EIR), expressed as the number of infectious mosquito bites per person per unit time, is a direct measure of malaria transmission intensity and, hence, an important metric for malaria surveillance and control programme evaluation [1,2]. Calculating EIR requires trapping host-seeking Anopheles vectors to determine the human biting rate (HBR) and prevalence of sporozoite infection [1,3,4]. Methods for operational mosquito sampling currently lack standardization, however, making it difficult to quantitatively compare EIRs between different locations or even within one location over time [2]. The gold standard in research studies for estimating mosquito-human contact has traditionally been the human landing catch (HLC), which can be employed either indoors or outdoors to capture mosquitoes as they land to feed on a human host [5,6]. Unfortunately, HLC is a labour-intensive procedure requiring highly trained collectors and extensive supervision, and results can be biased due to differences in the skill of collectors or their attractiveness to mosquitoes [2,7,8]. Furthermore, use of HLC has declined in recent years due to ethical concerns about potential exposure of collectors to mosquito-borne pathogens [5,9]. For all these reasons, HLC is unsustainable for large-scale operational sampling of malaria vectors.

Several exposure-free trapping methods have been evaluated as substitutes for HLC, but their reliability for estimating HBR and sporozoite prevalence have varied widely depending on characteristics of the study location or vector population [9,10]. Centers for Disease Control and Prevention miniature light traps (CDC-LT) are commonly used and are most efficient when hung next to a human host that is protected under a bed net [11,12]. Several studies have demonstrated close correlation between the numbers of Anopheles mosquitoes caught by CDC-LT compared to HLC [10-15]. Others researchers, however, have reported low catch rates [16,17] or inconsistent results using CDC-LT [18]. Also of concern, CDC-LTs have been found to catch higher numbers of Plasmodium-infected females compared to HLC [16,19], potentially resulting in overestimation of EIR.

Window exit traps (WETs) and resting traps are easy to deploy and do not require a dedicated collector to act as human bait [6,20]. Neither of these trap types, however, specifically target host-seeking females. WETs capture endophagic or endophilic vectors as they leave the house after feeding and/or resting. Resting traps are particularly useful for studying mosquito host choice behaviour because they collect large proportions of fed females immediately after a blood meal [6]. The efficacy of WETs and resting traps are strongly affected by environmental variables, such as the number of alternative exits or resting sites available for mosquitoes [6,17,21]. The C-design of the Ifakara tent trap (ITT) was recently developed as an alternative exposure-free tool to collect host-seeking malaria vectors [22]. While this tent trap has demonstrated high capture rates for Anopheles in both rural and urban study sites in Tanzania [22,23], its efficacy is known to vary depending on local vector density [24] and only one evaluation outside of Tanzania has been conducted in neighbouring Zambia [15].

The aim of the present study was to compare mosquito catch rates from CDC-LT, ITT, WET, pot resting trap (PRT), and box resting trap (BRT) against HLC in four different districts in western Kenya to identify the most useful tools for operational surveillance in this region of declining vector density. By employing a study design similar to recent trap evaluations in Tanzania [17] and Zambia [15], the secondary goal was to provide information enabling comparisons of trap efficacy between countries, an important step toward standardizing vector collection techniques across multiple sites.

All available mosquito collection methods suffer from some kind of bias or shortcoming [6,40] and there is no consensus regarding which techniques should be applied in which settings [2]. As a result, lack of consistent methodology greatly limits the ability to quantitatively compare EIRs between populations, or even within a population over time [2,40]. Numerous researchers have attempted to calibrate mosquito trapping techniques against the gold standard of HLC [10,18,40-42], but these studies employ varying designs and typically focus on site-specific results rather than facilitating comparisons of trap efficacy between sites [2,15]. With the aim of contributing to a standardized evaluation of mosquito trapping techniques across multiple countries, relative Anopheles catch rates from six collection methods were compared using a study design that has also been tested in Tanzania [17] and Zambia [15].

For the purpose of estimating Anopheles biting rates in western Kenya, trapping methods specifically targeting host-seeking mosquitoes (i e, CDC-LT and ITT) were the most effective. A high catch rate for CDC-LT is consistent with numerous reports from regions throughout Africa [10-14,24,43], including the recent trap evaluation in Zambia employing a similar Latin Square design [15]. In contrast, the parallel trap evaluation in Tanzania demonstrated low capture rates for CDC-LT in urban Dar es Salaam, an area with many competing light sources [17]. Other researchers have cautioned that light traps may lack sensitivity in areas where mosquito or human populations are particularly sparse [16,18]. Thus, while CDC-LT efficacy is generally expected to be quite high, caution should be used when extrapolating vector biting rates in certain areas that are very well-lit or have unusually low densities of humans or mosquitoes.

Consistent with results from the recent Tanzanian [17] and Zambian [15] evaluations, ITT was highly effective for collecting malaria vectors in western Kenya. A concern was raised, however, that nearly 18% of females caught by ITT had recently fed, a considerably higher percentage than those caught by CDC-LT (which prevents exposure to bites), but a similar percentage to those caught by HLC (which necessitates exposure to mosquito bites). One explanation could be that the ITT C-design did not entirely protect the collector from mosquito bites. In contrast, an ITT evaluation in Tanzania showed that when unfed mosquitoes were released into a semi-field enclosure, all females caught by the ITT C-design remained unfed (i e, females were unable to feed on the collector) [22]. Thus, another explanation could be that already engorged females were trapped by the ITT while attempting to rest or feed again. Further investigation, including analysis of blood meal sources, will be required to address this question definitively.

Although WETs did not specifically target host-seeking females, these traps performed moderately well in western Kenya with pooled relative catch rates of 52% for An. gambiae s.l and 49% for An. funestus. In contrast, both the Tanzanian and Zambian evaluations reported very low WET catch rates compared with HLCs (1% for An. gambiae s.l in Tanzania [17] and 3% for An. funestus in Zambia [15]), which the authors attributed to houses having open eaves and numerous points for entry and exit. Houses in western Kenya similarly had open eaves, leaving no clear explanation for the discrepancy between results from the present study and those from Tanzania and Zambia. It is plausible that the houses in western Kenya had narrower gaps between the roof and the walls or fewer windows to serve as alternate exits (only one WET was used per house). By fitting experimental huts in Tanzania with net baffles to allow entry but prevent escape of mosquitoes via eaves, Okumu et al.[44] were able to collect similar numbers of mosquitoes in WETs compared to CDC-LTs, suggesting that WETs can be very effective if household exits are limited. Another possibility is that different lighting or wind conditions may affect exit behaviour and WET capture rates. More detailed evaluations of WETs in different types of houses and environmental settings will be necessary for understanding when and where use of WET is reliable.

Resting traps, which primarily capture fed females seeking a secluded place to digest, yielded very few malaria vectors in western Kenya, a pattern also observed in Tanzania [17] and Zambia [15]. A potential explanation is that competing resting sites were plentiful enough indoors (e g, walls, clothing, furniture, etc.) and outdoors (e g, roof, vegetation, etc.) that only small numbers of mosquitoes entered the resting traps. While the absolute numbers of females caught in resting traps were low, the odds ratios for catching fed females in outdoor PRTs and BRTs were six to seven times greater compared to HLC. Thus, resting traps present a trade-off; they are practical for sampling engorged females to study feeding behaviour, host selection, etc. [20], but the numbers caught do not represent quantitative estimates of vector biting density. Even if resting traps accounted for a larger proportion of available resting sites, these traps would likely catch substantially fewer adult Anopheles compared to techniques that actively lure host-seeking females [45]. In areas dominated by exophilic vectors, however, one exception might be a novel barrier screen that could be erected between hosts and resting sites (if flight paths can be identified) to collect fed females on one side and host-seeking females on the other side [46].

Wide variation in the relative catch rate of each trapping method was observed between the four districts. Variability in trap performance was likely driven by district-specific differences in ecology, vector species composition, host availability, and/or history of malaria interventions. For example, ITT efficacy is known to decline with increasing vector density [47] and the lowest relative capture rate for ITT was observed in Rarieda, the district with the highest mosquito density. Interestingly, even the relative catch rates for indoor vs outdoor HLC differed between districts. Significantly more An. gambiae s.l were caught by indoor HLC compared to outdoor HLC in Rarieda, but this pattern was reversed in Rachuonyo, suggesting that outdoor biting was more prevalent in Rachuonyo. This disparity could be due to recent IRS activity (in addition to high bed net coverage) in Rachuonyo selecting for more outdoor feeding compared to Rarieda, where IRS was not conducted. Widespread use of indoor insecticide-based interventions has been correlated with increased exophilic and exophagic behaviour in residual mosquito populations in Tanzania [48], the Solomon Islands [49], and Equatorial Guinea [50]. Detailed analyses of biting location and biting time will be carried out to address whether vector control interventions in western Kenya have promoted changes in mosquito feeding behaviour, as well as the consequences for malaria transmission.

An obvious limitation of this study was that not all mosquito collection methods potentially suitable for operational vector surveillance could be tested. Knockdown pyrethrum spray catches (PSCs) are effective for collecting fed females resting indoors [6] and are commonly conducted throughout Africa [51,52]. Due to issues of insecticide persistence inside houses, however, PSCs were not incorporated into the Latin Square study design. Also promising are traps that employ novel synthetic human odorant blends to attract host-seeking Anopheles[53]. By utilizing a standardized a mixture of synthetic odours instead of human bait, such traps may circumvent problems due to differential skill or attractiveness of collectors. The relative attractiveness of odour-baited traps appears to vary with proximity to humans [53], however, and further field testing is required to identify settings appropriate for using these traps.

Despite high overall catch rates in this study for certain trap types (e g, CDC-LT, ITT, and to some extent WET), variability at the district level made it impossible to calculate any simple ‘calibration factor’ to convert trap catches into HLC equivalents consistently across western Kenya. Such variation in trapping efficacy is unavoidable as no method is an exact substitute for HLC [6,10,18,40,42]. Even the catch rate for HLC, the gold standard itself, is not constant; HLC results are influenced by the skill of collectors and their innate attractiveness to mosquitoes [2,7,8]. For the practical purposes of large-scale vector surveillance and operational estimation of EIRs, high overall capture rate and scalability allowing for intensive sampling are likely more important than perfect precision with regard to HLC. Vector populations exhibit marked spatial and temporal heterogeneity, sometimes spanning several orders of magnitude [54-57], such that estimating local population abundance is an inexact science. The challenge of monitoring malaria transmission intensity entomologically is further compounded as vector density and sporozoite infection prevalence decrease in response to successful malaria interventions; mosquito population distribution becomes even more sparse and patchy so that remaining infected mosquitoes are harder to find [54]. This point is well illustrated in western Kenya, where Anopheline populations have declined dramatically since the late 1990’s due to scale-up of ITNs [33]. As a result of sparse mosquito populations, collectors in the present study were unable to capture any infected females in Kisumu West, Nyando, or Rachuonyo (leading to EIR estimates of zero in those districts), despite timing collections to correspond with peak seasonal mosquito density. Malaria transmission was on-going in all four districts, indicating that these study results underestimated the true risk of infection in at least three districts.

Comparison of trap efficacy results (relative to HLC) from western Kenya to those from Tanzania and Zambia highlights the difficulty in estimating calibration factors for even the most widely-used, well-standardized trapping methods, such as the CDC-LT. One interpretation of these findings is that calibration factors must be validated before use in new geographic areas. Variability at very fine spatial and temporal scales, however, suggests that calibration factors may be inherently unreliable. Inter-district variation was observed for nearly all trap types tested in this small region of western Kenya. Furthermore, even a series of three exercises to calibrate CDC-LT against HLC within a single Tanzanian village yielded quite different results over the span of two and a half years [24,58]. Thus, rather than attempting to convert all results to HLC equivalents, vector biologists may be better served by employing consistent trapping methods over time and detailing the specific method they used to obtain EIR values. Concerns have been voiced regarding whether EIRs calculated directly from alternative trap methods are accurate with respect to ‘true’ EIRs from HLCs. To achieve sufficient spatial and temporal resolution, large-scale entomological surveillance programmes require frequent sampling [59] with traps that are inexpensive and easy to use with little or no expert supervision [15,17]. Because non-HLC trap methods afford greatly increased sampling effort, these cost-effective tools may in fact provide better indications of local malaria transmission. Recent evaluation of a routine vector surveillance programme in Tanzania using the ITT revealed that, despite low ITT catch rates per night of trapping compared to HLC, intensive and geographically extensive sampling with ITT was far more accurate for predicting risk of malaria parasite infection than the limited number of quality-controlled HLC collections conducted over the same period [47].

With respect to western Kenya, CDC-LT and ITT are the most effective trapping alternatives to HLC for large-scale operational vector sampling. If either method is impractical, due to bulkiness of the trap (ITT) or lack of electricity for recharging batteries (CDC-LT), WET may be an adequate substitute in this context. More broadly, these data suggest that traps actively targeting host-seeking females at night (e g, CDC-LT and ITT) will generally be the most useful for estimating vector biting rates in sub-Saharan Africa, particularly in areas with declining mosquito densities. Novel collection tools such as barrier screens or odour-baited traps, however, may be more appropriate in other regions where malaria is transmitted predominantly outdoors or early in the evening. Identifying which vector sampling methods are effective and practical for large-scale application under different epidemiological settings will require further standardized evaluations of mosquito collection techniques that permit cross-site comparisons.

The authors declare that they have no competing interests.

JW performed the statistical analysis and wrote the manuscript. NB and GO assisted in the design of the study and supervised the implementation of the study. GFK participated in the design of the study and helped to draft the manuscript. MJH conceived of the study, participated in its design and coordination, and helped to draft the manuscript. JMV participated in the design of the study and assisted with implementation and supervision of the study. JEG conceived of the study, participated in its design and coordination, advised with analysis, and helped to draft the manuscript. All authors read and approved the final manuscript.