Context
The study was conducted in four districts: Gurue, Alto-Molocue, Sussundenga, and Machaze. The districts of Gurue and Alto-Molocue are located in the province of Zambezia and have estimated populations of 403,558 and 375,504 inhabitants in 2015, respectively [10]. The districts of Sussundenga and Machaze are located in the province of Manica and have estimated populations of 165,616 and 134,515 inhabitants in 2015, respectively [10]. All four districts are rural type, with hardship health services access, and low social and economic conditions. In 2015 malaria prevalence in Zambezia and Manica was 67.9 and 25.5%, respectively [8].
Study design
A before-after design with control group was carried out 6 months after LLINs distribution, i.e., between June and July 2016. Two groups were considered: intervention (districts of Gurue and Sussundenga) and control (districts of Alto-Molocue and Machaze). These districts were selected based on the following matching criteria: (i) population size similarities; (ii) geographical area; (iii) similarity in the number of LLINs allocated for distribution; and (iv) having rural characteristics [9].
All the localities of these districts were selected for the survey. Within each locality household sample size was calculated by dividing the total sample size of the district by the number of existing localities. After determining the number of households in each locality, households were selected using systematic probabilistic sampling method. For both intervention and control districts the following household definition was assumed: includes all the people who live together or sleep in the same house/yard/plot and share the same food at meal times. When a man has more than one wife or woman, each of them is considered as a separate household.
Study sample size
For each group, sample size was computed in order to detect a significant difference of 10% between the intervention and control: p1 (intervention) = 80%; p2 (control) = 70%; alpha = 0.05; power = 0.9; Cp,power = 10.5. Therefore, the sample size for each group was 776 households, i.e., each district had 388 households as sample size.
$$n = \frac{{[p_{1} (1 - p_{1} ) + p_{2} (1 - p_{2} )]}}{{(p_{1} - p_{2} )^{2} }} \times c_{{{\text{p}},\,{\text{power}}}}.$$
Sampling strategy
A systematic random sampling was used in which every Nth member of the target population is selected to be included in the study. The sampling unit is the household.
Selection of households
In each locality, the households were selected based on the following strategy: first, households list (population frame) was identified and a number assigned to each household; then, the sample interval (number of households divided by sample size) was computed and a random number was chosen to start with; finally, from this first random number, households were systematically selected until the sample size was complete.
Data collection
A semi-structured questionnaire with open and closed questions was used. Before the beginning of the study a pilot study took place by applying the questionnaire to 20 households located in districts that were not part of the study. Some adjustments were made to improve the original version of the questionnaire. In order to avoid information bias, interviewers were not informed about the expected outcomes of the study, or if the district was from an intervention or a control group.
Additionally, the interviewers used observation techniques to support and validate some the responses given by the households, namely those related to the effective use of LLIN. Interviewers explained the purpose of the study and obtained authorization and written informed consent; if the household member refused to participate, the questionnaire was applied to the nearest house.
Variables
The questionnaire had questions related to the following quantitative and qualitative variables: (i) number of de facto people living in the household (people living in the same household for at least 6 months); (ii) presence or absence of campaign LLINs; (iii) number of campaign LLINs; (iv) use of campaign LLIN in the previous night and in the last four nights prior to the survey. All other existing LLINs (e.g., acquired from prenatal care or from campaigns prior to 2015, or from other source) were excluded from data collection during the interview and were considered as households without LLINs. The same approach was applied for those households that had campaign LLINs but slept under LLINs from another source; in this case was considered as owning campaign LLIN, but not sleeping under campaign LLIN. This was important to avoid information bias and effectively evaluate only the outcomes from the pilot.
Households inclusion criteria
The following inclusion criteria were additionally used to select the households to be surveyed: (i) households from the selected districts, (ii) households living in the district since July 2015 (period of the beginning of the campaign preparations), (iii) interviewee with at least 18 years of age, regardless of gender.
Outcomes of interest
The main outcomes are: (i) percentage of households with at least one LLIN in the intervention and control districts; (ii) percentage of population that slept under an LLIN the previous night (among the interviewees); (iii) percentage of LLIN owners that slept under an LLIN in the last four nights (among the interviewees); and (iv) percentage of households achieving universal coverage targets (one LLIN for every two persons).
Statistical analysis
All data were introduced and analysed using SPSS version 23.0. Univariate and bivariate statistical analysis was performed. For quantitative variables descriptive statistics such as mean, median, and standard deviation [SD] were used, while absolute frequencies and percentage were calculated for qualitative variables. For universal coverage estimation, the number of LLINs available in each household was divided by the number of de facto members from the respective household. Values greater than or equal to 0.5 (meaning that one LLIN is for two persons) were considered as universal coverage target achievement. Subsequently the percentage of households that reached universal coverage was calculated.
In order to analyse associations between the district categories (intervention and control) and the main outcomes of the study (LLIN ownership, use, and universal coverage achievement), odds ratio (OR) and 95% confidence intervals (CI) were calculated. For all statistical procedures, a 0.05 significance level was adopted for rejecting the null hypothesis.