Study area

The study was conducted in Ulanga District, Morogoro region, Tanzania, in the villages of Minepa, Mavimba, Igumbiro and Lupiro, along the Kilombero river valley (Fig. 2). The climate is hot and humid, with an annual rainfall between 1200 and 1800 mm and mean annual temperature of 20–32 °C [19]. All the villages have moderate perennial malaria transmission [20], as climatic conditions and rice farming (both irrigated and non-irrigated) create ideal conditions for high densities of mosquitoes [21]. Most community members are subsistence farmers, cultivating mostly rice, but also maize and other crops such as sweet potatoes and beans. The study included both permanent household residences in the villages, and in the distant semi-open improvised farm houses, commonly referred to in Kiswahili language as Shamba huts (Farm huts), where many adults spend significant periods of time tending to their crops. According to data from the Ifakara Health Institute Health and Demographic Surveillance System (HDSS), the local houses in the main villages have walls mostly made up of mud (56%), or of baked mud bricks [19]. The roofs are mostly thatched (70%) or of corrugated iron sheets [19] (Fig. 1a). The temporary structures (hereinafter referred to as Shamba huts) that are used by the migratory farmers in the fields are made from bamboo stems; sometimes have thatched grass/palm tree leaves for walls or just mud. Some are raised on stilts for protection from water and wild animals [22], and to give the farmers vantage when watching over their crops (Fig. 1b). The evaluation of the prototype mosquito-proof hut was done inside semi-field systems (SFMs) at Ifakara Health Institute, Kining’ina campus (8.11417°S, 36.67484°E). Details of the design and use of this SFS have been provided previously [23, 24].

Entomological assessments of human-biting mosquito densities inside and around the houses used by residents while in the main villages, and Shamba huts used while in the rice fields

First, enumeration of all the active Shamba huts in areas surrounding the four villages, Minepa, Mavimba, Igumbiro and Lupiro villages of Ulanga district, at the beginning of the study period. A full listing of main houses in the same villages was also obtained from Ifakara Health Institute HDSS. From the master list of Shamba huts, five Shamba huts were randomly selected in each village, so that there were 20 selected Shamba huts, located at the edges of the 4 different villages. To match the twenty Shamba huts used when the farmers are out in their farms, a set of 20 main houses regularly used by families were selected in the same four villages. The Shamba huts were matched village wise to the main houses, such that the Shamba huts were located in the adjacent rice fields near each of the villages. This way, in each village, a set of five main houses was paired with a set of five Shamba huts. These surveys were initially done in July and August 2013 and then repeated between July and September 2014. To quantify actual biting exposure in the Shamba huts relative to biting exposure within the main villages, mosquito collections were conducted in the selected main houses and also in the Shamba huts located at the edge of each of these respective villages.

Indoor collections were done using Centre for Disease Control (CDC) light traps^® set next to occupied bed with a person under a bed net [25, 26], from 1830 to 0700 hours each night, while outdoor collections were done using a newly-designed exposure-free system for conducting human-baited catches, where an adult male volunteer sits inside a two-chambered netting cage and catches mosquitoes before they actually reach the volunteer [27]. In this system, also called the M-Trap and earlier described by Mwangungulu et al. [27], the volunteer can sit during the night protected from mosquito bites, and mosquitoes attempting to bite him are trapped within the second compartment also having netting walls. Mosquitoes enter the system through three envelope-shaped entry points on the sides. Five such outdoor collection stations, each with an adult male volunteer (18–35 years old) were set up near the same five Shamba huts and another five M-traps set up near the matching main houses in the main villages. During these mosquito collections, continuous observations of temperatures and humidity were also done on hourly basis, inside both the Shamba huts and the main houses using portable indoor climate Tinytag Plus^® data loggers (Omni Instruments, London, UK).

Design, construction and testing of a prototype mosquito-proof hut for use by the migratory rice farmers while away in their distant rice fields

Design

The main aim was to create a portable mosquito-proof hut prototype with the following essential characteristics: (1) easy to transport, (2) large enough to accommodate a migratory family of two adults and one child, (3) easy for one person to set up while in the field on his/her own, (4) robust and durable for long-term field use, (6) highly ventilated and (7) can be mounted on basic pedestals already being used by farmers in the study area (Fig. 1b, d). A tentative hut design to meet these features was proposed (Fig. 3), upon which the structural engineers at the partnering company (Elastic Product Manufacturing Company Limited, Tanzania), worked to create the final prototype. Construction was done based primarily on this original design, while also considering preferences suggested by the farmers during our interviews and focus-group discussions, as well as additional modifications from the expert engineers. The final prototype design, also called Swai hut is shown in Fig. 4.

The basic structure consists of a 10ft × 10ft × 8ft steel frame supporting an 8ft × 8ft × 8ft square housing structure made of durable canvas and UV-resistant shade netting. It has large windows on the sides, with foldable canvas window flaps that can be rolled up or down to close the windows, and/or the entire side walls of the huts. It has wide screen viewing windows, which also improve ventilation and air flow. The large windows and open netting structure ensures utmost ventilation in the huts. The inside surface has a separating canvas wall that can be rolled up or down depending on need. The floor of the hut is made of thick poly-vinyl chloride (PVC) canvas, which is water proof, and extends upwards on the sidewalls forming a water-proof skirting for added protection. The roofing is designed to slightly slant backwards so that whenever it rains, all the rain water easily flow backwards, without seeping into the huts. This roofing material is foldable and made of high density polyethylene material. To enhance protection from biting insects, the huts have a double-panel door to prevent insects. The hut is fitted with hooks on the sides attached to the steel beams so that it can be tightly secured onto the ground, or mounted on top of a pre-fabricated sub-structure, as is common practice in rural-south eastern Tanzania (Fig. 1b, d). All the doors are secured using high-strength zippers, while the roll-down canvases, over the windows have laces so that they can be tightly fastened. This initial prototype was made at a total cost of US$ 1460.38 inclusive of construction labour and value added tax.

Semi-field and field testing of the portable huts to assess protection from host-seeking disease-transmitting mosquitoes

Controlled semi-field and field experiments were conducted to demonstrate that the portable mosquito-proof house can reduce mosquito house entry and bites. The semi-field experiments were conducted inside two chambers of the SFS. Each of the semi-field chambers used measured 9.6 m × 9.6 m, inside which there was growing vegetation, thus mimicking real-life mosquito ecosystems and villages [23].

The portable mosquito proof prototype was assembled in one of the chambers and a locally-made Shamba hut replica (of similar characteristics to those described and seen in the rice fields, but with dimensions similar to the prototype) was constructed in a different chamber of same size, so that there was a treatment and control chamber. A pair of consenting male volunteers were recruited to sleep inside each of the houses under bed nets as basic protection. Each night, 500 hungry 6–8 days old laboratory-reared female Anopheles arabiensis mosquitoes that had not previously taken any blood meals were released into the semi-field chambers, 1 h before start time of the experiments, which was 1900 hours. In the first round of experiments, the volunteers were provided with intact new Olyset^® nets, while in the second round they were provided with bed nets having 20 holes measuring 2 cm × 2 cm to mimic torn nets. The test was done for two rounds, each lasting 10 days. The different hut types were rotated between the two chambers, in a 2 × 2 cross-over design while the volunteers and hut positions remained fixed. Mosquito collections in both Swai hut prototype and the Shamba hut replica was done throughout the night using CDC light traps^® set next to the volunteer-occupied bed net inside the huts [25, 26]. Each morning, any mosquitoes left resting or dead on the walls, floor and other surfaces of the two huts were also collected by the volunteers, in this case using mouth aspirators.

Full field experiments were conducted in 100 m × 100 m open field sites in each of the four study villages in Ulanga district, south eastern Tanzania. In each of the villages, the portable mosquito proof hut and a replica Shamba hut (similar to the one used in semi field experiments) were placed 50 m away from each other and compared directly. A pair of consenting adult male volunteers was recruited to sleep inside each of the huts under Olyset^® nets each night. This was done for 16 days in each of the four villages, with the two hut types rotating positions on the ninth day, to account for any positional bias. The volunteers however did not change their positions, and in this way, the volunteers and position were taken as a single source of experimental variation, as the hut types were rotated. Mosquito collections inside both the Swai hut prototype and the Shamba hut replica were done throughout the night using CDC light traps^® set next to the occupied bed net [25, 26]. Each morning, any mosquitoes resting or dead on the walls, floor and other surfaces of the two huts were also collected by the volunteers using mouth aspirators. These binary 16-night comparative tests were repeated in each of the four villages, working with a different pair of volunteers per village.

After the field controlled trials, the Swai hut design was tested when in use with actual rice farming families as compared to normal Shamba huts that are used in the rice fields. This was done by rotating the Swai hut between four rice farming families in a 4 × 4 Latin square after every 10 days. The end of this final experiment coincided with the end of harvest season, when rice farmers were leaving the rice farms, back to the main villages.

Mosquito identification

All the mosquitoes collected during the field experiments were sorted by taxa and blood feeding status (i.e. as blood fed, gravid or non-blood fed). The sorting was done on fresh samples each morning, without letting the mosquitoes dry. A sub-sample of Anopheles gambiae s.l and Anopheles funestus group mosquitoes was stored in small micro-centrifuge tubes (Eppendorf^®), containing silica gel. These samples were further identified into sibling species through polymerase chain reaction (PCR) [28, 29]. Enzyme-linked immunosorbent assays (ELISA) were also conducted to determine Plasmodium falciparum sporozoite infection rates in the mosquitoes [30]. All the laboratory analysis were conducted at Ifakara Health Institute, Tanzania.

Assessments of views, behaviours and experiences of the migratory rice farmers regarding malaria transmission and its control

A qualitative survey was conducted in the same four villages, Minepa, Mavimba, Igumbiro and Lupiro, where entomological surveys were done. This involved a stage-wise approach where three different complementary behavioural science methods for data collection were used, that is: (a) semi structured interviews (SSI) with household heads, (b) timed participant observations (PO) of activities conducted by members of households, and (c) focus group discussions (FGDs) with a selection of the community members who had participated in the SSI and PO assessments. All of these were implemented using study guides prepared and piloted in advance of the study.

A cross section of migratory rice farmers was identified using the non-probability sampling technique of snowballing among target populations in the study villages. This way the migratory farming households helped nominate others who were also migratory. Initially, the study team identified and planned to visit a total of 138 households (35–36 households per village), but this was reduced by half to 64 households (16 households per village), after the pilot study suggested a high level of homogeneity among the migratory farming households, who were giving highly similar answers indicating the data would be quickly saturated (i.e. answers from participants starting to be repetitive). During the SSIs, the researcher asked and gently probed for participants’ opinions on issues, such as: (a) whether they were aware of differences in risk of mosquito bites while in the rice fields compared to main villages, (b) whether they had any experiences with mosquito-borne diseases, including malaria, (c) what control or protective measures they were using while away in their farms, and (d) how they cope with bites and malaria infection whenever they are in the rice fields.

After, half of the interview candidates in each village (eight households per village) were then selected to participate in the timed participant observations to identify the main activities in which the migratory farmers and their family members were usually involved in at different times of the night, and which could expose them to mosquito bites. Selection of candidates for the participant observations was based on willingness to participate, as well as the presence of at least one member of the household who is able to read and write, so that he or she could conduct the actual observations after being trained. All activities carried out from 1800 to 0700 hours were catalogued in the observational checklist given to the trained family members in each participating household. This was done for three nights in each household, resulting in a total of 24 household-level observations in each of the four villages. The reason for relying on trained community members was the needed to minimize the observer bias, at times also referred to as the Hawthorne effect, where study subjects might change or modify their behaviours in response to being observed [31]. Every hour, the observers noted down by ticking a pre-defined check box whether any of the family members was participating in any of the stated outdoor activities. In case there was an activity being conducted, that had not been pre-included in the observation list, the observer wrote this down as well at the end of the observation sheet. This procedure allowed us to catalogue all outdoor human activities occurring in the peri-domestic space and to specify on hourly basis when each of these activities was most frequently done.

After the semi-structured interviews and timed-participant observations, a group of participants was recruited from each of these same villages to participate in FGDs on the observed outdoor behaviours and associated risks experienced in the rice fields and also the main villages. The FGD consisted of groups of 6–8 adults from the migratory farming communities. During these sessions, how the participants reacted to and interacted with the newly created Swai huts for protecting the migratory farmers was also assessed. These interactions with the Swai hut were also video-taped after group consent. Two FGDs were conducted in each of the four villages, males and females separately but with mixed ages ranging from 21 to 68 year olds. At the start of the first sessions of each FGD, the participants with help from the research team assembled the Swai hut prototype. The rest of the discussions were then conducted around the hut, while the participants handled the device, creating an opportunity for them to make direct suggestions on specific features that could or should be improved. A total of eight FGD’s were completed, during which a group of 6–8 adults participated in setting up the prototype hut, while discussing its potential benefits and limitations, focusing particularly on the mosquito-proof features, portable nature and ease-of-use. Each discussion lasted about 35–40 min excluding the assembly of the Swai hut prototype. These were conducted at school grounds in each of the villages. The other themes for the FGDs included key concerns and proposed coping strategies currently being used by migratory rice farmers while in the fields, considerations of housing as a protective measure against infections, and specific views on the portable mosquito-proof hut prototype i.e. the Swai hut.

Data analysis

All quantitative data was entered and verified in Microsoft Excel 2010, after which analysis of the mosquito catches was performed using the open source R statistical software [32]. Relationships between the indoor mosquito densities and the different hut types were i.e. main houses, the Swai huts or the Shamba huts, were examined using generalized linear mixed effects models (GLMMs), with lme4 package [33]. Mosquito densities were modelled as a function of fixed factors including, house type and village, treating volunteer pairs and date of collection as random factors. To address the over-dispersion observed in the field data, a negative binomial family of models with log-link function was used.

The qualitative data on the other hand was analysed as follows: All audio formats of the SSI and FGD’s were transcribed and then translated from Kiswahili (the language in which the data had been collected) to English. The translated transcripts were then imported to Atlas.ti software and analysed as per the following themes: challenges in the distant farms, malaria prevention in the farms, effectiveness of traditional huts in preventing mosquito entrance and views regarding the newly designed portable mosquito-proof huts. A code book to allow easy identification of the different themes of interest from the translated transcripts was created. The observational data was entered into Epi Data^® software version 3.1 and then imported to STATA statistical analysis software package 9 (Stata Corp). All the different activities performed were tabulated with respect to time of night, and then the final histograms produced in Microsoft Excel.