The Cascades Region of Burkina Faso is 18,406 km2, with a dry season generally stretching from the end of October to May and a rainy season from June to September. The main human income generation activity is agriculture, particularly cereal crop and some cotton production. The area has a high burden of malaria infection with prevalence in all ages, reaching 60% [20,21] to > 80% [22]. The time communities in Niakore and Toma spent indoors overnight, when Anopheles mosquitoes that can transmit malaria are most active, is assessed in the analysis. Both human behavioural data and entomological data are then combined to estimate the number of bites received per person per night, and the proportion of mosquito bites received outdoors [23,24] for sociodemographic groups defined by age (under 10-years, 11–20 years, 21–50 years and over 50 years), gender (male and female), and month (April-May; July; October-November). In the analysis, the age groups were identified to provide broadly similar sample sizes in the respective cohorts. These covariates are then used to explain the variation observed in the estimated metrics using regression analyses. Temperature estimates were recorded in situ using two devices: Elitech USB Temperature Data logger, Elitech UK, and Tinytags, Gemini UK.
Human behaviour
Data on human behaviour were collected within a broader ethnographic approach. Between March 2017 and August 2018, the ethnographer conducted 14 months of participant observation in the region. The ethnographer lived in the communities long-term, and partook in the daily activities of the research participants, including farming, attending social and religious events, and participating in family activities [25,26]. Participant observation was used to explore the intersection between local life ways and malaria preventive strategies, and with the additional purpose of limiting response bias, expected in respondent-dependent methods (e.g. surveys, diaries, and other self-reporting approaches) [27]. Additionally, qualitative data on mosquito net use indoors were collected through semi-structured interviews, informal conversations and focus group discussions (Supplementary Material S1). These methods were adopted to investigate how compliance with mosquito net use was understood and performed by the hosting communities in everyday life, focusing on informal mosquito net procurement, intra-household mosquito net allocation, and how participants negotiated treatment-seeking practices for ill family members with their social duties and financial means.
This paper draws on three rounds of structured observations of night-time activities (October-November 2017; April-May, and July 2018). The observations were conducted in two study communities, Niakore and Toma, relying on convenience sampling. Recruited participants resided within 30 metres from the ethnographer’s house and were selected regardless of their gender or age to provide a representation as broad as possible of the nocturnal activities of community members within the confines of haphazard sampling. Adopting a convenience sample ensured consistency in the timing of data collection and research participants’ pool, limiting the structured observations to those participants who lived and stably resided, throughout the weeks of observations, within the established area surrounding the ethnographer’s residence. In Toma, this resulted in 47 individuals (33 females, 14 males) belonging to six different households for observations conducted in October 2017 and April 2018. In July 2018, internal migration of students and farmers increased the pool of participants to 71 (43 females, 28 males), requiring the help of a research assistant. In Niakore, the sample consisted of 24 individuals (14 females, 10 males) from a single household, but this behaviour reflected that of the community more broadly. Age ranges were uneven (e.g. there were only three children aged between 10 and 15 in the sample from Niakore). Across the 7 sampling nights and 3 rounds of data collection, there were a total of 327 and 1070 nightly estimates of the time spent indoors by an individual in Niakore and Toma respectively. Table 1 summarizes sampling night data in each village.
Night-time observations
Structured observations, organized in rounds of one-week, recorded the presence of individuals outdoors throughout the night. The initial observations were conducted after the ethnographer lived in each village for a minimum of 6 weeks to minimize reactivity (the reaction of research participants to the awareness of being observed) [28]. In Toma, observations were conducted in October 2017, April and July 2018; in Niakore, in November 2017 and May 2018. The third round of structured observations in Niakore, scheduled for August 2018, had to be cancelled due to logistical circumstances. Bias in observations was minimized because the ethnographer was embedded within the community, familiar with the participants involved, and able to understand individual nightly patterns of exposure.
Each observation, conducted at intervals of 30 minutes between 18:00 and 06:00, recorded the time individuals went indoors or exited houses throughout the night. The results do not include shorter instances of time outdoors (e.g. exiting to visit the toilet). A 30-minute interval allowed the ethnographer to safely complete the tour of the compounds and confirm the identity of the residents without interfering with any activity they were conducting. At the same time, such an interval ensured that the observations could be standardized, so that people were concluded to be in- or outdoors in a binary fashion, allowing the quantification of human behaviour and the estimation of biting risk. To achieve standardization, this approach scored as ‘outside’ for any given half-hour interval a person who was outdoors at the moment of the observation and engaged in activities classified as ‘labour’ or ‘leisure’. Labour included household chores, farming, the harvesting of caterpillars; leisure referred to resting, sleeping, socializing, participating in social or religious activities. Shorter periods outdoors falling outside this interval, and which may still result in exposure to bites, have been missed and remain a limitation of this approach.
Mosquito biting behaviour
Mosquito feeding attempts were measured using human landing catches (HLC) [29] across the Cascades region and conducted monthly from 1st October 2016 to 29th December 2019, between the hours of 19:00 to 06:00 [19]. In each village, the collection was carried out twice a month, at two different households each time. Each month collections were made both inside houses and within the peri-domestic areas. Collectors were between 19 and 30 years of age and randomly assigned to households by pair (2 collectors per household). To avoid bias due to differences in individual attractiveness to mosquitoes, each member of the pair rotated between the indoor or outdoor position every hour. On each collection day, a minimum distance of 30 metres was observed between houses and 8 metres between the indoor and outdoor collection points of the same household were maintained to avoid biases linked to household location and indoor/outdoor collection points. Mosquitoes were actively collected for 45 minutes, followed by a 15-minute break each hour, as they attempted to feed on the exposed legs of a volunteer.
In Niakore (sampled between 25th October 2016 and 10th November 2017, Supplementary Data), the principal malaria vector mosquito complex present was Anopheles gambiae sensu lato (s.l.) (373 mosquitoes indoors and 317 outdoors. Approximately 82% of those molecularly analysed (n = 212 mosquitoes) were An. gambiae and 18% were Anopheles coluzzii (Supplementary Data). Very few other Anopheles species were recorded (2 Anopheles pharoensis indoors and 4 outdoors; 4 Anopheles nili indoors and 1 Anopheles funestus s.l. indoors). In Toma, mosquitoes were collected in the dry season (17th, 19th, and 22nd April 2018) and in the wet season (5th, 7th, and 10th September 2018, Supplementary Data). The principal species complex was again An. gambiae s.l. (566 mosquitoes indoors and 563 outdoors) but molecular distinction of the species complex was not performed. There were 36 An. pharoensis (50% indoors), 2 An. nili (1 indoors), and a single Anopheles coustani and An. funestus, both located outdoors. This analysis focused only on An. gambiae s.l. (Fig. 1a) given that other species were in meagre numbers. Insecticide resistance was tested in Niakore using susceptibility bioassays performed on adult mosquitoes reared from larvae. A total of 58%, of 91 An. gambiae s.l. tested survived exposure to the discriminatory dose of pyrethroid deltamethrin according to WHO guidelines [30]. In Toma, insecticide susceptibility data were not collected; one limitation is the assumption that the villages sampled in Sanou et al. [19] are representative. The raw mosquito data for Niakore and Toma, included in the broader analysis, are provided in Supplementary Data.
There is minimal difference in the number of mosquitoes observed indoors and outdoors in Niakore (373 vs 317, respectively). However, it cannot be established whether this is the result of opportunistic mosquitoes seeking blood meals on exposed volunteers conducting the experiments outside, who would otherwise be indoors (and potentially protected by mosquito nets). The simplifying assumption is that, in the absence of indoor protection, bites received indoors and outdoors are broadly equivalent. This is supported by related work showing relatively equal biting was observed across the Cascades Region [19].
The mosquito densities in Cascades Region, Burkina Faso, are predictably seasonal and consistent between sample sites, as observed from human landing catch data for An. gambiae s.l. (n > 40,000 vector mosquitoes) seeking blood meals outdoors across 12 villages, 324 sampling nights, in 2016 and 2017 (Fig. S1a; figure adapted from [19]). The pattern of biting activity, i.e. the number of mosquitoes recorded seeking to blood-feed outdoors at each hour of the night, for each week of the year when human activity was observed, was also relatively consistent across months (Fig. S1b). For each person, for each hour of human activity observations, the time spent outdoors was multiplied by the estimated number of mosquitoes biting at the matched hour and given the predicted seasonal densities corresponding to the week of the year when the human observations were completed. For each night, these hourly estimates of mosquito bites received were summed to estimate the per night per person exposure to Anopheles bites.
The number of mosquitoes caught during an hourly period (Fig. 1a) is assumed to represent the number of mosquitoes attempting to feed on humans for the same period. In the absence of data, no bites are assumed to occur during the hours for which mosquito bites were not sampled (06:00-19:00). Raw data were converted into the proportion of all mosquito bites received over 24 hours, taken indoors (denoted λI(t)) or outside (denoted λO(t)) at hour (t) using:
where subscript h indicates whether bites are taken indoors (h = 1) or outdoors (h = 0) (31).
For different sociodemographic groups (age, gender), months and villages, the proportion of An. gambiae s.l. bites received outdoors was assessed to understand the potential of indoor interventions and the protection gap remaining for outdoor control. The observational data on people movement are summarized by gender, age and month to estimate the varying proportion of An. gambiae s.l. bites received outdoors (ϕO) following [23,31]:
where pI(t) is the proportion of people inside at hour (t), λI(t) is the biting rate indoors at hour (t), and λO(t) is the biting rate outdoors at hour (t). The analysis of variance results using the Cascades data to estimate the proportion of mosquito bites received indoors and outdoors is provided (Table S1) and histograms of the Niakore specific data are presented in Fig. 1 (Table 2).
Statistical analysis
At the individual level, the association between age, sex, and month was investigated in relation to: i) the number of hours spent indoors per person per night (yA), or; ii) the number of bites received per person per night (yB) using two generalized linear mixed-effects models (GLMM) that took the structure:
where the linear predictor η (transformed by the inverse link function f and assuming a Gaussian distribution D) was fitted to the data with age, month (distinct between villages) and gender included as explanatory variables in matrix X, and repeated measures for each individual i included as random effects in Z. The parameter is the standard deviation. Parameters β and u are coefficients at the population level and group level (for random effects) respectively. Given the observations made on distinct individual behaviours between months, interactions between age and month (which was later dropped as insignificant), and sex and month were included in the model exploring the number of bites received per person per night. Both models were fitted in a Stan computational framework (http://mc-stan.org/) accessed with the ‘brms’ package [32]. All data are provided in Supplementary Data.
Finally, differences in the predicted number of bites received per person across weeknights were tested independently for each month to understand how social activity might be driving differences in exposure risk. A general linear model (‘stats’ package, R [33]) was fitted to the log-transformed number of bites received per person per night (yC) with weeknight (Monday to Sunday) and gender included as explanatory variables (Supplementary Table S2, Fig. S3).