Effects of community-level bed net coverage on malaria morbidity in Lilongwe, Malawi

The malaria transmission intensity (MTI) study was an annual cross-sectional survey of P. falciparum prevalence, conducted between 2011 and 2013, in support of a multi-site Phase III trial (GSK MAL 055) of the candidate vaccine RTS,S/AS01 [20]. The catchment area comprised Traditional Authorities (TAs) surrounding the Malawi Ministry of Health’s Area 18 health centre. The survey was implemented during the peak malaria season (February–June). Each year, 400 households were randomly selected from a geographic information system (GIS) database containing global positioning system (GPS) coordinates for all catchment area households [21]. Households with a child under 5 years old were eligible to participate. A child under 5 years old was recruited from each household. A second participant over 5 years old was also recruited from each household. Recruitment of older participants included individuals aged 5–19 years, and adults over 20 years, and was alternated by household. A coin toss was used for selection in instances where multiple people in a household of the same age group were eligible to participate. A total of 800 individuals (400 <5 years, 200 5–19 years, 200 >20 years) were recruited annually. Subjects in the RTS,S vaccine trial were excluded from participation.

Field data collection was conducted in teams of three, which included two community health workers, and one nurse. Community health workers were trained to administer questionnaires and collect GPS coordinates of participating households. Field nurses were trained to administer a rapid diagnostic test (RDT) using SD Bioline HRP-2 Kits (Gyeonggi-do, Republic of Korea). The questionnaire, administered by the community health workers, captured demographic, health and behavioural information, including individual-level bed net use the previous night, household mosquito repellent use during the previous 7 days, and household indoor residual spraying during the previous 12 months. Participants were also asked to report the number of household members using a bed net while sleeping. Household characteristics, including wall, floor, roof, window material, and electricity, were also recorded. Household GPS coordinates were linked to each participant.

During the interview, a field nurse conducted a finger prick and collected blood from each participant for a rapid diagnostic test (RDT) and microscopy. The RDT was performed to test participants for the presence of malaria parasitaemia in the field. A field nurse assessed the RDT and clinical status of a participant and provided immediate care and treatment in accordance with national guidelines, and referred the participant to the nearest health facility for treatment. Thick and thin blood smears were prepared. Slides were made in duplicate, and transported to the Area 18 health centre laboratory. For each slide, the presence of parasitaemia was determined independently by two trained microscopists. In the case of non-concordance, parasitaemia was determined by a third microscopist.

ITNs were distributed to all participating households after completion of each annual survey.

The primary outcome for this study was P. falciparum infection represented as a binary variable defined as the presence of any parasitaemia by light microscopy. An index variable representing household wealth was constructed using principal components analysis, and is a composite score of household characteristics including wall, floor, roof, window material, and electricity [22]. The composite wealth index was categorized as low, medium, or high [22].

Using the ‘buffer’ tool in ArcGIS 10.0, 400-m and 1-km radii were placed around each participant household. The intersect tool was used to identify households located within the 400-m and 1-km buffers, from which the total population and the total number of individuals sleeping under a bed net were calculated. The percentage of the study population sleeping under a bed net within each radius, excluding the individual participant was then calculated. Bed net use by close proximity neighbours can reduce individual risk of P. falciparum infection [17], and community coverage effects have been identified as far as 1.5-km [12]. Therefore, a 400-m radius was selected because it was the minimum distance required to capture at least one neighbour for each household in the study area. A 1-km radius was selected to measure the effect of community bed net coverage at a larger spatial scale. A 1-km radius captured a large proportion of neighbouring households while maintaining neighbourhood variability between participant households.

Community-level vaccine coverage was also measured to account for potential protective herd effects [23, 24]. Data were obtained from the RTS,S/AS01 candidate malaria vaccine trial to identify the percentage of the population vaccinated in the study area [20]. Using the buffer and intersect tools, community-level vaccine coverage variables were constructed for 400-m and 1-km areas surrounding participant households.

Euclidean distance (i.e., straight line distance) between household and the nearest body of standing water was also measured to capture potential exposure to mosquito breeding areas. Community bed net coverage, vaccine coverage, and distance variables were all constructed using ArcGIS version 10.0 software (Environmental Systems Research Institute, Redlands, USA).

Sample population characteristics were compared across the three survey periods using Pearson’s Chi square test for categorical variables and the Kruskal–Wallis test for ordinal variables.

Spatiotemporal patterns of malaria prevalence and bed net use were examined using kernel density estimation. A roving window or kernel passes across the study area and computes the density of events within the spatial window. Density estimates are calculated at the window centroid and are based on the weighted values of all events in the window. Events are weighted according to distance from the centroid, with greater weight assigned to events near the centre of the window [25]. Kernel density estimation was selected to visually examine a smoothed study surface while maintaining participant confidentiality. Kernel density estimates were computed and mapped using ArcGIS version 10.0 software (Environmental Systems Research Institute, Redlands, USA).

Community-level bed net coverage (400-m and 1-km) between individuals with malaria and those without malaria was compared using a two-sample t test.

The relationship between community-level bed net coverage and P. falciparum infection was examined using mixed effects logistic regression to account for clustering of individuals within each survey period. Models were stratified by age group (<5 years; 5–19 years; ≥20 years). Percentage bed net use among study participants within 400-m and 1-km radii were examined separately. Variables selected a priori included individual ITN use, age, and household wealth. The percentage of vaccinated persons living within 400-m and 1-km of participants were considered at a statistical significance level of p < 0.10. Additional variables considered at a statistical significance level of p < 0.10 included living within 1-km of a river or a large standing body of water, household commercial mosquito repellent use, and indoor residual spraying. Predictive margins were also plotted to examine the probability of P. falciparum infection given the percentage bed net coverage within 400-m and 1-km of each household.

This study was approved by the Institutional Review Boards at the University of North Carolina at Chapel Hill (Chapel Hill, North Carolina, USA) and the National Health Sciences Research Committee, Lilongwe, Malawi. Study information was provided in both oral and written form to participants. Written consent was obtained from all participants and legal guardians.