A community-based vector surveillance system was implemented in the Asembo (-0.18139; 34.38552) and Uyoma (-0.31667; 34.3167) communities of Rarieda sub-county, Siaya County in western Kenya (Fig. 1). The study site is part of the Kenya Medical Research Institute (KEMRI) and Centers for Disease Control and Prevention (CDC) Health and Demographic Surveillance System, where the prevalence of microscopy-confirmed malaria among children < 5 years of age is 39% [17]. Malaria vector species in this region are An. funestus, An. arabiensis and An. gambiae s.s. [18, 19]. The region has a bimodal rainfall pattern, with ‘long’ rains between April and June and ‘short’ rains in October and November, which are associated with increased malaria transmission.
Study design. Eighteen clusters were designated in the study area, with each cluster ≈ 4km in diameter and centred on the house of the collector. In each cluster, 60 houses were randomly selected for mosquito collection (Fig. 1). Each of the 60 houses was sampled monthly by indoor and outdoor light traps and indoor Prokopack aspiration. Each CBC began their collections from a randomly selected reference house in each cluster and sampled a set of 3 neighbouring houses each day for 20 days. Each primary collection house was assigned two replacement houses for mosquito collection if the primary houses became unavailable.
Selection and training of community-based collectors. A mixed approach was used in the identification and recruitment of the CBCs with the following inclusion criteria: (i) a resident of the community, (ii) resident of a house with a tin roof for installation of solar panels; (iii) ownership of a personal means of transport, preferably a bicycle or a motorbike, (iv) prior experience as either a community health volunteer (CHV) or mosquito collector, and (v) ability to operate a mobile device for data collection and transmission. The CBCs were identified through local health facilities if community health volunteers (CHVs) were recruited or through the local administrative authorities where the CHVs were unavailable. Of the eighteen collectors, six had prior entomological experience from previous studies in the region while the rest did not.
The CBCs were trained in mosquito collection techniques using CDC light traps and Prokopack aspiration. Additional training included basic mosquito identification using morphological features to differentiate between anopheline and culicine mosquitoes and between female and male mosquitoes, and to classify the physiological status of mosquitoes as either fed, unfed, gravid, or half-gravid. The collectors were also trained on the capture and transmission of entomological and household data using Open Data Kit (ODK) software on an Android mobile device. Other training included operating a solar charging system for charging light trap batteries and tablets, administering a questionnaire and the consenting process. All training included practical demonstrations and field practice and was undertaken over five days before the start of the study. After the initiation of mosquito collection, support training was provided to the CBCs as needed.
Building and installation of the solar charging system. Eighteen solar charging units were assembled by a local engineer within Kisumu city (0.0917° S, 34.7680° E). Each unit was comprised of four solar panels attached to a lockable metallic frame. Three solar panels were connected to charge controllers (SolarTech®), with each charge controller connected to a 12 V rechargeable battery. The fourth solar panel was connected to a Universal Serial Bus (USB) cable for charging the tablet (Fig. 2).
During installation, the set of solar panels on a metallic frame was attached on top of a tin-roofed house belonging to the CBC. The frame was attached to the roof with screws from inside the house, while the screws holding the individual panels on the frame were blocked with a metallic plate to prevent theft. Cables from the solar panels were passed under the iron-sheet roof to connect to the charging station within the house.
Equipment and material. Each CBC was issued with a solar charging unit, three 12 V rechargeable batteries with terminals modified to connect two light traps at the same time, six CDC light traps, three with 5 m long connecting cables for outdoor installation and three with 2 m cables for indoor trapping, a Prokopack aspirator and three collection cups for indoor mosquito collection, an adult mosquito cage, a mouth aspirator, three paper cups, a pair of forceps, a magnifying lens for mosquito identification and Petri dishes for transferring and holding collected mosquitoes. Other equipment included a tablet (Nexus 7, ASUSTek Computer Inc., Taipei, Taiwan) for the collection and transmission of data, data forms and consent forms. Additionally, a set of 20mL scintillation vials with 70% ethanol for the preservation of collected mosquitoes was provided to the collectors bi-weekly and the tablets were loaded with data bundles for internet connectivity monthly.
Consent. Before initial collections, under the supervision of project staff, the CBCs obtained written consent from the heads of all 60 randomly selected households in each cluster. After the initial written consent, verbal consent was sought from the household during each subsequent mosquito collection. Consent was only sought from the backup households when an original household withdrew from the study, and a replacement household was recruited for sampling.
Mosquito collection. Mosquito collections by indoor/outdoor CDC light trap and indoor Prokopack aspiration were conducted in three houses per night. CBCs undertook collections over five consecutive nights each week, sampling each of the 60 consented houses monthly. Indoor CDC light traps were set in the sleeping area next to an occupied bed net at about 1.5m from the floor. Outdoor traps were placed within 5m from the house, suspended at 1.5m from the ground on either a tree, pole or under the roof. Both traps were run from a single 12V battery. The traps were operated from 18:00 h to 07:00 h the following morning. After removing the light traps in the morning, the Prokopak aspirator was used to collect any mosquitoes resting in the house.
During the mosquito collection period, the collector administered a brief questionnaire to collect information on household characteristics, including roof type, wall type, presence of eaves, presence of bed nets and net use, presence of cattle and number of people that slept in the house at every collection. The location of each house was recorded using a Global Positioning System (GPS) at each visit.
Mosquito processing. The CBCs processed the mosquitoes in their homes using a magnifying lens to assist in identification. The CBCs sorted mosquitoes by subfamily (anopheline or culicine), by sex, and by abdominal status (fed, unfed, gravid, or half-gravid). The number of mosquitoes in each of these categories was recorded on a paper form, and the data were subsequently entered into the tablet and transmitted to the cloud server. All mosquitoes were preserved together in 70% ethanol in a scintillation vial. Each vial was labelled with the collection date and method, and house code. The collectors were instructed to record and preserve any insect which they thought to be a mosquito.
The preserved mosquitoes and completed paper forms were collected from the field every two weeks and preserved mosquitoes underwent further processing at the KEMRI laboratories in Kisumu. Trained entomology technicians who were blinded to the results of the CBC identifications repeated the classification process performed by the CBCs. In addition, all mosquitoes of the Anopheles genus were further identified by the trained technicians to species/complex level using morphological features [20].
Monitoring of light trap battery charging cycles. The light trap batteries were charged daily, and the charge status was recorded at the beginning and end of every charging session. CBCs scanned the battery barcode label and recorded whether the battery was fully charged, half-charged or completely discharged using light indicators on the charge controllers. In addition, the CBCs submitted daily status reports on battery charge levels, any faults in the solar charging unit, lost/broken items, and needed supplies.
Supervision of CBCs. Initial trainings on light trap collections and mosquito processing were provided within the first month of mosquito collection. Subsequent collections, sample processing and identification, were not supervised routinely. Monitor of the collections was planned to be conducted remotely by daily reviews of submitted data, but because of budget limitations, data quality monitoring was rarely done. Cases of delayed or failed data submission were followed up directly with the individual CBCs, and where necessary, field visits were conducted to assist with specific challenges. Every CBC was visited fortnightly to pick up samples and provide new tubes with 70% ethanol. Specialised technical assistance with the solar charging system was provided routinely by a field-based technician within the first year of study. After the initial training at the beginning of the study, no additional structured training sessions were offered for the collectors, but technical support was provided as needed during regular supervisory field visits.
Quality-assured collections by trained entomology technicians. Parallel collections by trained entomology technicians were conducted in eight of the eighteen clusters sampled by the CBCs for fifteen months (May 2017 to July 2018) as a quality assurance check on the CBCs. The eight clusters were selected based on mosquito densities from the CBCs: three clusters with higher densities, three with the lowest and two with median mosquito numbers. The collections were conducted in the same houses as the CBCs within two weeks following the CBCs’ visit, without the CBCs’ knowledge. If houses sampled by the CBCs were unavailable, the quality assurance (QA) team selected a neighbouring house for sampling. The quality-assured collections were conducted for one week each month by indoor CDC light trap and indoor Prokopack aspiration in ten houses out of the 60 that were visited by CBCs each month in each of the eight clusters that were monitored by the QA team. The CDC light traps were deployed from 18:00 h to 07:00 h the following morning in the sleeping area next to an occupied bed. After the removal of light traps in the morning, indoor resting collections were performed by indoor aspirators in the same houses.
Cost-effective analysis. The cost of each sampling scheme was estimated for indoor CDC light trap collections. The cost per sampling effort was approximated based on the total expenditures, including procurement and installation of equipment, personnel and transportation costs.
Laboratory analysis. All mosquitoes were transported to the laboratory and identified to species level morphologically [21, 22], and the abdominal status was scored as either fed, unfed, gravid or half gravid. Female mosquitoes were divided into three parts; the head and thorax were used for determination of sporozoite infection rate by enzyme-linked immunosorbent assay (ELISA) [23], the abdomens of blood-fed females were kept for blood-meal host determination and the remainder of the specimen was used in polymerase chain reaction (PCR) analysis to identify species within the An. gambiae s.l. and the An. funestus group [24] and for future molecular genetic analysis. Approximately 30% of the mosquitoes morphologically identified as An. gambiae s.l. were randomly selected and identified to species by PCR. For the An. funestus group, only 20% were identified by PCR [25], as previous studies in this area had found that the only member of the Funestus group present in adult collections was An. funestus s.s.
Data management and analysis. Data collection was undertaken using ODK Collect, designed with an interface to limit entry errors. The house code was unique for every house sampled, and each collection effort was uniquely identified by a combination of house code, collection method, and collection date. At the end of each collection, each collection cup, paper cup or light trap bag containing samples was labelled with a combination of variables to distinguish between different collections. The combination of date, collection method, and house code was used to track the samples through laboratory processing.
During morphological identification of the mosquitoes, a unique barcode was given to individual mosquitoes and used to link the various laboratory procedures to the individual mosquito, including species identification by PCR, analysis of sporozoite infection by ELISA procedure and blood meal analysis.
Data analysis was performed using R version 3.4.1. Comparisons of the number of mosquitoes collected for 15 months were made between the CBCs based on identifications by the trained technicians and separate collections by the QA teams from the same houses as the CBCs, approximately 2 weeks apart. Houses that were visited only by CBCs or QAs were excluded from the analysis. Mean abundance of An. gambiae complex, An. funestus and An. coustani were calculated per trapping night for indoor CDC light trap and Prokopack aspiration, and a Generalised Linear Mixed Model (GLMM) was fitted to the data to measure the differences between the number collected by the different teams (CBCs vs QA teams). Since the data were over-dispersed, a negative binomial model was fitted using the glmmTMB package in R. Collection team, and collection method were included in the model as fixed effects, whereas house was treated as a random effect. Subsequent analyses were done by negative binomial regression after aggregating the total number of mosquitoes collected per month and cluster. A test of association between CBC and QA collection was performed by Pearson’s correlation and Spearman’s rank correlation.
Although paired identifications of individual mosquitoes by CBCs and trained technicians were unavailable, comparing aggregated numbers of Anopheles mosquitoes from the same collections was done to assess the accuracy of identification by the CBCs. To assess the accuracy of identification by the CBCs, mosquitoes classified as Anopheles by each CBC were aggregated by collection date, house of collection, method of collection and the relative numbers identified in each collection (total Anopheles by CBCs/total Anopheles by lab technicians) was calculated.