Study Area
This study took place in six Shehia (wards) on Unguja Island, the main island of Zanzibar, an archipelago located off the coast of mainland Tanzania (Figure 1). Sites were selected in partnership with the Zanzibar Malaria Elimination Program (ZAMEP) on the basis of high malaria incidence, defined as annual parasite incidence (API) of 5/1000 or higher, and receipt of IRS in 2016.
Study Design
This study included household surveys, structured observations of nighttime human activity and sleeping patterns, and indoor and outdoor mosquito collections. Human behavioral data were collected in the dry season in December 2016 and rainy season from April through May 2017, while entomological data were collected from December 2016 to December 2017. The same households were used for human behavioral and entomological data collection and across data collection time points.
Sample Size
The sample size was generated to answer the primary research question of whether there was a difference in number of malaria vectors biting indoors compared to outdoors each night. The method developed by Cohen for power calculations in behavioral sciences was used [11]. Based on previous estimates from entomological monitoring, an average of 7 An. arabiensis caught outdoors and 4 caught indoors was assumed, which translates to medium effect sizes. However, considering potential for heterogeneity in vector biting densities across households, sites, and seasons, effect sizes (measured as R2/1-R2) as low as 0.02-0.15 were assumed on a generalized linear model regressing mosquito counts as a function of position (indoors/outdoors as a fixed effect and day and location as random effects to account for heterogeneity in the data). The study was then designed to achieve 80% power at 95% confidence intervals, which returned a requirement for 200 nights of mosquito collection per site.
Based on previous research using direct observation of night time human behavior, it was determined that 20-25 households per site would be needed to capture variation in human behavior across households and sites [12]. Therefore, to achieve 200 nights of mosquito collection per site from the same households where human behavioral data was collected, eight nights of indoor and outdoor mosquito collection were carried out in each household. Collection nights were evenly distributed across seasons
A random number generator was used to select 30 households for each of the six shehia (wards) using household listings provided by the Sheha (local leader) for each site. Of the 180 households selected, 143 were home, and therefore approached, during community entry. A total of 135 households consented to participate as follows: Bwejuu (n=20), Charawe (n=23), Donge Mchangani (n=24), Mbaleni (n=23), Miwani (n=25) and Tunduni (n = 20). Of the households approached, four households declined to participate, two household heads were not available to provide consent, and two indicated that they would be traveling throughout the data collection period.
Data Collection
Human Behavior
Study team members administered a survey to respective heads of household prior to beginning night-time observations. The survey included questions on household members, housing characteristics, and bed net ownership. For each person living in the household, information was collected on relationship to head of household, age, sex, and pregnancy status, if known. Net ownership and characteristics were recorded using a standard net roster [13].
Study team members made structured observations half-hourly of each individual household members’ activities and sleeping patterns from 6:00pm to 7:00am. This included a) whether each household member was indoors, outdoors, or away from home, b) whether each household member was awake or asleep and c) if sleeping, whether they were using an ITN.
Data from surveys and household observations were recorded electronically using tablets configured with programmed questionnaires and observation forms. These data collection tools were first translated into Swahili and then programmed using the open-source platform, Open Data Kit (ODK) [14]. Data collectors were trained on how to operate tablets, complete the forms, and upload the data. The data was uploaded daily to a secure server configured with Secure Socket Layer (SSL) with encryption. Appropriate logical constraints were implemented on every question to ensure data quality. In addition, for the household observations, time stamps were fixed to block entry of missed observations. Supervisors reviewed data for quality on a daily basis and provided feedback to the data collection team.
Vector Behavior
Indoor and outdoor mosquito collections were conducted hourly from 6:00pm to 7:00am in the selected households in each shehia using human-baited miniaturized double net traps (DN-mini) (Figure 2), an exposure-free method developed at Ifakara Health Institute [15]. This trap was developed based on the design previously used by WHO [16] and modified by Tangena et al [17]. Observations of indoor and outdoor proportions and hourly biting patterns of Anopheles in Tanzania with DN-mini match those of the gold standard estimate of human exposure to mosquito bites the human landing catch [15]. Mosquitoes were collected hourly using a mouth aspirator and put in a paper cup, with a separate cup labeled for each hour of collection. The collectors sampled mosquitoes for 45 minutes each hour and rested for 15 minutes. Collectors worked in two sets with each set doing collections indoors and outdoors for six to seven hours each night of collection.
Mosquitoes were sorted by taxa, sex, and physiological status (fed, unfed or gravid), and then stored individually or in batches for laboratory analysis. These samples were stored in microcentrifuge tubes containing cotton wool and silica gel, and were later analyzed by Polymerase Chain Reaction (PCR) to distinguish between members of An. gambiae s.l., and by enzyme-linked immunosorbent assays (ELISA) to determine proportions carrying Plasmodium falciparum sporozoites in their salivary glands [18]. The field data and laboratory results were recorded electronically using tablets, linked, cleaned, and stored in a secure web-based database application, the Ifakara Entomology Bionformatics System (IEBS) [19].
Data Analysis
Human behavior
Descriptive analysis of household survey data and observation data were completed using STATA 14 [20] and graphs were generated in Microsoft Excel [21]. ITN access was calculated using the approach originally described by Kilian et al and recommended by the Roll Back Malaria Monitoring and Evaluation Reference Group [13, 22]. Potential ITN users were calculated by multiplying the number of ITNs in each household by two (assuming a maximum of two users per ITN). If the potential users exceeded the number of people in the household, the number of ITN users was set to the number of household members. ITN access was then calculated by dividing potential ITN users by the total number of study participants [18]. The use to access ratio (UAR) was calculated by dividing the proportion of the study population observed to be using an ITN by the proportion of study population with access to an ITN.
Vector Behavior
Mosquito biting patterns were assessed based on hourly catches each night for dry and rainy seasons separately. Collection nights were evenly distributed across seasons. No mosquitoes were infected with Plasmodium, and therefore no calculation was done for the sporozoite rate. The probability of a mosquito biting indoors or outdoors was estimated from a Generalized Linear Mixed Effects Regression (GLMER) with a Poisson distribution with a log link, using household ID and round of collection as random effects and location (in versus out) as a fixed effect. Analysis was done using R statistical package version 3.6.1[23].
Human-vector interaction
Human exposure to malaria vectors was calculated based on data from household observations carried out in the peri-domestic setting and indoor and outdoor mosquito collections in the same households. Exposure patterns were calculated only for An. gambiae s.l. as densities of other Anopheles complexes were too low to explore patterns of exposure.
Analysis included calculation of the following indicators of human-vector interaction, described by Monroe et al. (currently under review at Malaria Journal):
1. Percentage of vector bites occurring indoors for an unprotected individual () [24-28]: Calculated as the sum of the measured indoor vector biting rates (BI) for each one-hour time period (t) over a 24-hour period weighted by the estimated proportion of humans indoors (I) at that time, divided by total location weighted exposure, i.e. itself plus the sum of the outdoor biting rates weighted by the proportion of humans outdoors (O, where O=1-I) at each time over the same 24-hour period: (see Formula 1 in the Supplementary Files)
2. Percentage of vector bites occurring while asleep indoors for an unprotected individual () [24-26, 28, 29]: Calculated as, the sum of the indoor vector biting rates (BI) for each one-hour time period (t) over a 24-hour period weighted by the estimated proportion of humans sleeping (S) indoors at that time, divided by total location weighted exposure i.e. the sum of the indoor and outdoor biting rates respectively weighted by the proportions of humans indoors and outdoors at each time over the same 24-hour period : (see Formula 2 in the Supplementary Files)
3. Percentage of all vector bites prevented by using an ITN ()[27, 30-32]: Calculated as the product of the proportion of exposure occurring while asleep and the personal protection against bites (feeding inhibition) provided by an ITN while in use (ρ). ITNs were assumed to prevent 97% of vector bites when in use. This estimate for ρ was based on reference estimates from experimental hut trials of 7 brands of ITNs in Tanzania [33]. (see Formula 3 in the Supplementary Files)
4. Percentage of remaining exposure occurring indoors for a protected user of an ITN () [26-28]: Calculated by adjusting the estimate of πI,u to allow for the indoor personal protection provided by using an ITN: (see Formula 4 in the Supplementary Files)
5. Population-wide mean personal protection against biting exposure provided by community-level coverage of humans (C) with ITNs : Calculated as the product of the coverage of the human population with ITNs, estimated as the proportion of humans using an ITN at each hour during the night and the overall personal protection provided by an ITN while it is in use, and accounting for the attenuating effects of exposure occurring when the user is active outside the net. (see Formula 5 in the Supplementary Files)
Ethical Approval
This study received ethical approval from the Johns Hopkins Bloomberg School of Public Health (IRB# 7390), Ifakara Health Institute (IHI/IRB/No: 035 - 2016), and the Zanzibar Medical Research and Ethics Committee (Protocol #: ZAMREC/0005/OCT/016). Only consenting mosquito collection volunteers participated. Volunteers received appropriate training and were provided with medical supervision, chemoprophylaxis, and access to diagnosis and treatment on a regular basis. The heads of household provided separate written consent for household observations and mosquito collection respectively. Community entry activities were conducted prior to beginning data collection. This included a one-day information session for Sheha (local leaders) and assistant Sheha in the selected sites and district-level representatives. During site visits study team members explained the purpose of the study to community members and obtained informed consent from selected heads of household.