Wealth, mother's education and physical access as determinants of retail sector net use in rural Kenya

The Kenyan Ministry of Health (MoH) developed an ITN strategy paper in 2000, which spelt out the government's intentions for scaling up the use of ITN nationwide [8]. This was incorporated into the Kenyan National Malaria Strategy (NMS) as one of the four principal pillars of malaria prevention and control [9]. In 2001, Population Services International (PSI) launched a nationwide ITN social marketing programme with funding from the British Department for International Development (DFID). The aim of the project was to establish a "net culture" in the country through extensive marketing and education. Information campaigns were undertaken to increase basic knowledge on the causes of malaria and means of prevention, emphasizing the vulnerability of pregnant women and children under five years of age [10]. A rural kiosk-based distribution system was established to increase subsidised net availability run in parallel with joint programs to sell nets through other NGOs.

The study was carried out in four districts purposively sampled by the MoH to provide detailed longitudinal information on coverage of interventions proposed as part of the NMS [11]. The study districts have been described elsewhere in detail [12, 13] and represent the broad categories of malaria transmission across Kenya [14]. They include Kwale district on the coast with seasonal, high intensity malaria transmission; Bondo on the shores of Lake Victoria with high intensity perennial transmission; Kisii Central and Gucha districts (Greater Kisii) representing the low seasonal transmission conditions of the Western highlands; and Makueni district, a semi-arid area with acutely seasonal, low malaria transmission.

A geographic information system (GIS) was developed for each district reflecting high spatial resolution enumeration area (EA) population data, roads and footpaths, rivers, forests, gazetted areas and a 100 m Digital Elevation Model (DEM). In addition all market centres were mapped using Global Positioning Systems (GPS) [13, 15]. An EA was defined by the Central Bureau of Statistics (CBS) during the 1999 national census as an area of about 500 people or 100 homesteads [16]. The GIS data were all stored in ArcView GIS 3.2 (ESRI Inc., New York, USA) raster format.

A larger randomly sampled community survey undertaken in 2001, reported elsewhere [12, 17] provided the sampling frame for the present study. EA sampling was done to ensure that there was enough power and precision to detect changes in net use over time among rural communities. The study was based on a cross-sectional survey with the unit of analysis being a child under the age of 5 years residing in a homestead within a randomly selected rural EA cluster. In each district a total of 18 rural EAs were randomly selected. All homesteads within the sampled EAs were surveyed. Questionnaires were translated into the primary language of each district and back-translated into English by independent assessors fluent in each language to identify and correct ambiguities in translation and meaning. Field staff were recruited at each district and trained over three days in the survey instruments, how to approach homesteads and check data in the field. Fieldwork began during November 2004 to coincide with the short rainy season in each district. Field staff used GIS generated district and EA maps to locate each homestead within selected EA. All members of the homestead were enumerated and assigned a unique identification number. Homestead heads or responsible members of the homestead were interviewed as part of the survey related to homestead socio-demographic characteristics, assets and other wealth indicators (Table 5) and malaria prevention practices. Also captured was information on homestead head's education and main source of income. Mothers or guardians of all children under the age of five were asked specific questions related to the child's use of nets, whether the nets had been treated with insecticides and net sources. Details of the mother's age, marital status, educational attainment and source of income were recorded.

Data entry and storage was undertaken using MS Access (Microsoft, Redmond, USA), through customised data entry screens with in-built range and consistency checks. Descriptive summaries of all information were generated using STATA version 8.2 (Statacorp 2003, College Station, USA) and MS Excel 2002 (Microsoft, Redmond, USA). All information specific to the EA, homestead, mother or guardian were linked to the relevant child through the use of a primary homestead identifier that was consistent across all data sets.

Principal component analysis (PCA) was performed, using STATA, to construct a homestead wealth assets index from information on the broad range of indicators collected during the survey. PCA is a data reduction technique that provides a method of identifying, from a large set of variables, those that contain most of the information common to all. The first principal component often represents the linear index of these important variables with most information [6, 18]. Quintile distributions were derived for each district separately based on this assets index. Quintile descriptions based on district-specific homestead wealth assets index were used to allow for aggregated wealth assets to be defined at the district level rather than across districts. The district effects were then adjusted for during the regression analysis.

GPS coordinates for homesteads and market centres were used to develop physical accessibility models. Most studies of physical access to health interventions use straight-line (Euclidean) distances between population and service locations [19, 20]. Previous work has shown that this approach does not accurately reflect actual distances travelled and overestimates the extent of physical access to interventions especially in rural settings [21]. A travel time model was developed from the high-resolution spatial raster data on market centres, transport network, rivers, permanent water bodies, topography and land cover for four the study districts. Using a GIS algorithm based on the specifications of the Naismith-Langmuir rule for pedestrian movement [21, 22] the model simulated the ease or difficulty of physical access to markets centres based on the presence of a road or footpath, changes in slope along the path and restriction caused by the presence of barriers such as rivers, forests or gazetted areas. This rule states that a walker can maintain a speed of 5 km/h on flat road; 2.5 km/h on level ground off-road; 1 hour added for every 600 m of ascent; 10 minutes subtracted for each 300 m moderate descents (-5° to -12°); and 10 minutes added for each 300 m steep descent (steeper than -12°). The algorithm used an iterative region-growing approach in which each pixel containing a market centre was taken as a 'seed' pixel around which regions of travel time pixels were grown. Only the fastest route to a given pixel was used to calculate travel time to the market centre. Only pedestrian motion was modelled as majority of people in rural homesteads of the study districts walked to market centres. Travel time to the nearest market centre was assigned to each homestead.

The study set out to look at predictors of use of nets by children <5 years of age obtained from the retail sector as these represent the principal sources of nets during the periods leading up to the time of the survey. Therefore, children who used nets purchased from the retail sector were compared to those who did not use nets. Children who used nets from sources other than the retail sector were excluded from the regression analysis. First, a univariate regression analysis was performed to identify which of the predictor variables were significant to the outcome measure, i.e. use of nets purchased from retail sector by children. These predictors defined homestead, mother and child characteristics. In the univariate analysis the odds ratio (OR), p-value and 95% confidence interval (CI) for each factor's association with nets purchased from the retail sector were computed. Any factor with a p-value < 0.15 was considered to be a potentially important covariate of retail-net use. The factors examined were: homestead wealth assets index; gender of homestead head; homestead demographics; travel time to the nearest market centre; homestead use of insecticide residual spraying (IRS); whether any homestead member attended public malaria awareness meeting (baraza) or owned printed education materials on malaria; whether mother could read; mother's education level (no education, primary incomplete, primary complete and secondary and above); mother's main source of income; mother's marital status; whether mother was pregnant; age of the child; sex of the child; and child's ownership of a health card.

Seventy two rural EAs were sampled in the four districts, covering a total of 2,807 homesteads. Three homesteads refused to participate in the study while 104 had homestead heads or other responsible members away on visits during the survey (and after re-visits). Overall 2,700 homesteads were surveyed (a response rate of 96.2%) but completed questionnaires for five of them were lost, reducing the number of rural homesteads in the database to 2,695 with a total population of 24,020, including 3,755 children under the age of five years belonging to 2,603 mothers/guardians in 1,731 homesteads. A total of 562 (15.0%) children slept under a net the night before the interview. 327 (58.2%) slept under nets purchased from the retail sector of which 224 (68.5%) were socially marketed by PSI (Supanet^®). 235 slept under nets obtained from a variety of other sources, including subsidised nets through antenatal clinics (ANC) (Table 1).

Data on 46 asset indicators for the 2,657/2,965 homesteads were used in the PCA to construct homestead level wealth assets index (Table 5). There were differences in the set of indicators accounting for the biggest overall effect on the homestead wealth assets index between the districts. For example in Bondo, houses with stone/concrete walls increased the homestead wealth assets index the most (1.60) while in Greater Kisii occupants paying rent in Greater Kisii contributed the most to the index (3.09). The index assigned 15.4% of the children in the analysis to homesteads in the least poor quintile while 20.1% to those in the most poor quintile. Overall, use of nets purchased from the retail sector increased with homestead wealth and was significantly different between the top and lowest quintiles (least poor, 40.1%; most poor, 5.5%; χ² = 137.1, p < 0.0001). The univariate regression of use of nets purchased from the retail sector against homestead wealth assets index showed that the least poor children were more than five times more likely to use retail sector nets than the most poor, even after this was adjusted for the variation between districts (OR = 5.12, 95% CI = 3.38–7.97; adjusted OR = 5.30, 95% CI = 3.42–9.20) (Table 2).

Children's use of nets purchased from the retail sector was shown to be closely correlated with mother's education, with only 14.4% of children of uneducated mothers using nets compared with 32.7% of those whose mother's had education up to secondary level and above (χ² = 37.91, p < 0.0001). The univariate logistic regression of use of nets purchased from the retail sector with mother's education level (with and without adjusting for district effects) showed that children whose mothers were educated to secondary level or above were up to three times more likely to use nets purchased from the retail sector than were those whose mothers were not educated (OR = 2.42, 95% CI = 1.64–3.57; adjusted OR = 2.92, a95% CI = 1.93–4.41) (Table 2).

Modelled pedestrian travel time to the nearest market centre was used as an indicator of access to sources of retail sector nets. The mean travel time to market centres for users of nets purchased from the retail sector was 24 minutes compared to 35 minutes for those who were without these nets (Table 2). Most of the children, (54.7% of those who did not use nets purchased from the retail sector and 67.3% of those who did), were within 24 minutes of the nearest market centre. There was a significant difference in use of nets purchased from the retail sector with travel time between children within 24 minutes of the nearest market centre and those 48 minutes or more (χ² = 37.85, p < 0.0001). The univariate logistic regression showed that each log unit increase in travel time resulted in a 75–82% decreased probability of using a net purchased from the retail sector (OR = 0.25, 95% CI = 0.13–0.44; adjusted OR = 0.18, 95% CI = 0.09–0.38) (Table 2).

The 15 variables studied as potential predictors of net use were structured at different levels: community; homestead; mother; and child. Of these, three had univariate p values of ≥ 0.15 and were dropped from the multivariable analysis (Table 3). Two other variables (mother's main source of income and homestead use of Indoor Residual Spraying (IRS) were also dropped because the former was highly correlated with homestead wealth assets index and mother's education level while over 95% of the children lived in homesteads which did not use IRS. In the descriptive analysis, all the three key variables (homestead wealth assets index, mother's education and travel time to market centres) showed a significant association with use of nets purchased from the retail sector (Table 2). Mother's education, however, was found to be highly correlated with homestead wealth asset index and was excluded from the final model (Table 3). When the model was fitted with either mother's education level or homestead wealth asset index only, both predictors showed significant association with use of nets purchased from the retail sector (results not shown).

Multivariable analysis of the remaining 9 variables showed a variety of associations with use of nets purchased from the retail sector (Table 4). All predictors were adjusted for the effect of the district in the multivariable regression. Homestead wealth assets index showed the strongest significant association with use of nets purchased from the retail sector. Children in the least poor quintile were 10 times more likely to use nets from this sector than those in the poorest quintile. Those in the intermediate quintiles were also between two to five times more likely to use nets purchased from the retail sector than children living in the poorest homesteads. Travel time to the nearest market centre (transformed using natural logarithms) remained strongly associated with use of nets purchased from the retail sector use even in the multivariable model, with every log unit increase in time resulting in 49% less likelihood of using these nets. The gender of the homestead head was not associated with use of nets purchased from the retail sector. The presence of older children of primary school going age (5–14 years) appeared to reduce the likelihood of use nets from the retail sector among younger children (adjusted OR = 0.68, 95% CI = 0.52–0.90). Surprisingly, attendance at public meetings or ownership of printed materials related to raising malaria awareness appeared to have a negative association with use of nets from the retail sector (adjusted OR = 0.70, 95% CI = 0.46–1.05), although this was only of borderline statistical significance (p = 0.087). Children of mothers who were married or unmarried but living with partners were significantly more likely to use nets purchased from the retail sector than those whose mothers were either single or widowed (adjusted OR = 2.74, 95% CI = 1.83–4.10). Children older than one year and those belonging to mothers who were pregnant at the time of the survey were 29% and 41% less likely to use nets purchased from the retail sector respectively. Children who owned an immunisation card were twice as likely to use nets purchased from the retail sector as those who did not have a card.