Study design and sample size
A two-stage, cluster-sample, cross-sectional household survey was carried out in seven of the eight Central region districts chosen for the Central 2 region of the 2009 MIS (Figure 1); one district was excluded as it had recently received an NGO-supported universal coverage campaign. In the first sampling stage, 30 enumeration areas (EAs) located in the study districts were selected from a list of EAs used in the 2002 Uganda population census using probability proportionate to size sampling. Within each sampled EA, all households were listed and 20 were randomly sampled from the listing for inclusion in the survey. In each household, an adult household member aged ≥18 years, if possible the head of household, was asked to respond to the household questionnaire (see below) and all women aged 15-49 in selected households were asked to participate in the woman’s questionnaire.
Allowing for a 10% non-response rate and a design effect of 1.7, a sample size of 600 households was calculated to estimate net use within 5% and to assess changes in LLIN ownership and usage of 25% or greater relative to 2009 MIS Central region measures, translating to a prevalence ratio of 1.25 or greater, with 80% power.
Study questionnaires and variables
Two questionnaires were used in the survey: a household questionnaire and a women’s questionnaire for all women aged 15-49 years in selected households. Both instruments were based on the model MIS questionnaires developed by the Roll Back Malaria Partnership Monitoring and Evaluation Reference Group, as well as other questionnaires from previous surveys conducted in Uganda, including the 2006 Uganda Demographic and Health Survey (UDHS) and the 2009 Uganda MIS.
The questionnaires were translated into Luganda, the major local language commonly spoken in central Uganda. The questionnaires were pre-tested prior to the main data collection to assess the appropriateness of the wording of the questions and to verify the translations and skip patterns.
The household questionnaire was used to list all usual members and visitors in the selected households, inquire about bed net ownership and use, and gather information on household assets and characteristics. The women’s questionnaire was used to collect information from all women aged 15-49 years on background characteristics (age, education, literacy, employment), recent reproductive history, pre-natal care and preventive malaria treatment received during pregnancy for the most recent birth, treatment of fever among children under five and knowledge about malaria (causes, ways to avoid, types of medicines).
Data collection
Training for survey teams lasted ten days and included didactic sessions on survey objectives, methodology and questionnaires, classroom role-play and mock interviews, field testing, and feedback to trainees. A total of three survey teams, each composed of three interviewers and one supervisor, carried out the survey over three weeks, between 17 January and 7 February, 2011. Survey teams spent up to two full days in each EA and made at least three attempts to interview each sampled household. The teams used hand-held computer tablets programmed using QDS software (Nova Research, Bethesda, Maryland, USA).
Data analysis
Data were weighted using survey weights that accounted for the probability of selection at each stage as well as non-response at the household level. Analysis of key bed net ownership and use variables were adjusted for the complex survey design using the svy commands in Stata 11.0 (Statcorp, College Station, Texas, USA), which account for both the survey weights and clustering at EA level, using Taylor series linearization. Key variables from the 2011 data on bed net ownership and use were compared to the 2009 MIS data for the Central region to assess trends after the targeted distribution.
Principal components analysis (PCA) was used to create wealth scores for each household based on ownership of selected assets, dwelling characteristics, source of drinking water, sanitation facilities, and other characteristics related to a household’s socio-economic status[11]. Specific variables included in the PCA were: source of water, source of toilet, source of fuel, flooring material, wall material, roof material, electricity, and household ownership of a radio, cassette player, television, mobile telephone, refrigerator, table, chair, sofa, bed, cupboard, clock, watch, truck, bank account, and scooter. The PCA accounted for 18.6% of the variability of the first component, within the range found in other studies using a PCA approach to characterize wealth[12]. To assess the effectiveness of the campaign targeted to households with pregnant women and children under five years of age, households were defined as eligible if they had a child under six years of age at the time of the survey, approximately eight months after the campaign, which would also capture households with pregnant women at the time of the campaign. Since there was no data on eligibility of each household during the campaign, this variable is only a proxy for eligibility at the time of the bed net campaign.
For assessing predictors of LLIN use among children, bivariate and multivariate prevalence ratios from Poisson regression (with robust standard errors) were used, which have been shown to more accurately reflect risk than odds ratios when the outcome is common[13]. Predictors of bed net use that were significant at p < 0.10 in bivariate analyses were included in a multivariate model.
Ethical approval
The protocol was reviewed and approved by the Centers for Disease Control and Prevention in Atlanta, GA and by the Committee for Human Research of the University of California San Francisco, USA. Ethical approval for the survey was also obtained from the Makerere University College of Health Sciences Research and Ethics Committee and the Uganda National Council of Science and Technology.