Study sites
The survey was conducted in Oromia and SNNPR regions in January 2007. Oromia is located in south-central and western Ethiopia and has an estimated population of 26,553,000; whereas SNNPR Region lies in the south-west and has an estimated population of 14,901,990 (Figure 1) [8]. Since 2001, The Carter Center-assisted onchocerciasis control program has been implementing CDTI in these two regions where 54% of the country's population live. The regions are divided into administrative units which include, in descending order: zone, woreda, kebele and stateteam. The stateteam, also called Got/Garre in some areas, is the lowest administrative unit and comprises an average of 50 households or about 250 people with an elected representative.
The two regions (Oromia and SNNPR) were the main domains for the survey. Two other overlapping domains within the regions were defined based on the presence of onchocerciasis and thus coverage by the Carter Center CDTI program: CDTI areas and non-CDTI areas, with the aim that baseline net utilization would be estimated in these areas for future comparison. Our hypothesis is that the involvement of CDDs in LLIN promotion will improve LLIN utilization in CDTI areas. Only TCC-assisted onchocerciasis areas were intensively assessed; two other zones where CDTI is managed by another organization (East and West Wellega) were included in the survey but not over-sampled. Kebeles known to be non-malarious were excluded from the sampling frame.
Sample size estimation
The sample was estimated to determine prevalence of malaria and mosquito net coverage and use within: 1) each of the two regions; 2) the CDTI areas of the two regions combined; and 3) non-CDTI areas of the two regions combined. We calculated our sample size for malaria parasite prevalence testing assuming an expected prevalence of 8.0%, 2.0% margin of error for 95% confidence interval (CI), 5% level of significance, design effect of 1.2, and up to 20% allowance for non-response. A sample of 1,000 people (~200 households) was estimated as the minimum number required for malaria parasite prevalence testing per domain.
For estimating sample size for net use, we assumed 25% of children under-five years old slept under a net the previous night, with 5% margin of error for 95% CI, 5% level of significance, and design effect of 1.2, meaning we would need 381 children under-five years in each domain, which we would expect to find in ~400 households (16 clusters of 25 households). We decided to have 16 clusters (400 households) in our smallest domain (CDTI areas) to assess mosquito net coverage, but to take blood slides from alternate households (200 households per domain).
For logistical reasons, each region was stratified into four quadrants, in each of which eight clusters were selected. The CDTI areas comprised 2 quadrants (one in each region with a total of 16 clusters) while the remaining six quadrants (48 clusters) formed the non-CDTI domain. This gave 32 clusters per region and a total of 64 clusters. Comparison of the results between TCC-assisted CDTI areas and non-CDTI areas will be reported separately.
Sample selection
To select 200 households in each quadrant, we used a multi-stage cluster random sampling design. In each quadrant, eight kebeles (clusters) were selected using probability proportional to population size. Within the kebele, five stateteams (which are all approximately similar in size) were selected by lottery, literally drawing the names out of a hat at the kebele office. In the final stage, five households were selected from the state team using the random walk method [9]. All members of the selected households were included in the sample.
Household questionnaire
The survey questionnaire was based on the Malaria Indicator Survey Household Questionnaire, modified for the local conditions [10]. The questionnaire was translated and printed in Amharic and Oromiffa languages and field-tested in non-survey kebeles to determine the validity of the pre-coded answers. Interviews were conducted with the head of household, or another adult if the head of household was absent or unable to respond for any reason. If interviews were conducted with someone other than the head of household then the respondent was requested to answer as though he or she were the head of household. The data collection form had two parts: household questionnaire and malaria parasite prevalence.
In the household questionnaire, respondents were asked about: proxy indicators of wealth (electricity in the household, possession of a functioning radio set, and possession of a functioning television set); room construction materials; indoor residual spraying; presence and type of mosquito net (verified by observation); demographic information on residents; and where people slept. Interviewers asked to see each net by room in the house, determined whether it was a LLIN or not, and asked who slept under it the previous night. The questionnaire was also designed to determine whether any household residents slept outside, and if so whether they used a net (ITN, LLIN or untreated) or LLIN.
Malaria parasite prevalence
Consenting residents of even-numbered households were recruited for the malaria parasite prevalence survey. Participants had both a rapid diagnostic test (RDT), which gave an on-the-spot diagnosis, and provided thick and thin blood films for microscopy. The rapid diagnostic test used was ParaScreen (Zephyr Biomedical Systems, http://www.tulipgroup.com) which detects both P. falciparum and other plasmodia species (most likely P. vivax in the Ethiopian context). The test uses approximately 5 μl of blood and is readable after 15 minutes. Participants with positive rapid tests were offered treatment according to national guidelines: CoArtem® for P. falciparum infection, chloroquine for other malaria infection, and clinic-based quinine therapy for self-reported pregnant women [11].
While the RDTs were used to provide rapid treatment for infected persons in the field, the parasite prevalence survey was based on microscopic examination of stained blood films. Two blood slides, each composed of thick and thin films, were taken from each participant by a medical laboratory technician according to standard WHO-approved protocol [12]. Slides were labelled and air-dried horizontally in a slide tray in the field, and thin films were fixed with methanol immediately after drying. Slides were stained with 3% Giemsa for 30 minutes at the nearest health facility when the team returned from the field. Usually, field teams returned to the clinic each evening but when working in more remote areas, they were sometimes obliged to sleep in the field and stain the slides the following day. To ensure maximum participation, households with absentees were revisited on the same day to recruit those missing at the first visit. Repeat visits to recruit absentees on other days were not logistically possible.
Blood slides were read at a reference laboratory in Addis Ababa and classified qualitatively as either negative, P. falciparum positive, P. vivax positive, or mixed infection. One hundred high power fields of the thick film were examined at a magnification of 1000×, before identifying a slide as negative or positive. If positive, the thin film was read to determine the species. Parasite density was not quantified. To ensure accuracy, all positive slides and a random sample of 5% of the negative slides were re-examined by a separate microscopist, who was blinded to the diagnosis of the first slide-reader. The overall agreement between the two microscopists was 99.4%. The second slide from each participant was used if the first was broken or unreadable. The identity of survey participants who had positive blood slides but had negative RDT results was sent back to the field teams for follow-up and appropriate treatment, where necessary. Comparison of the results obtained with RDT and slides has been reported elsewhere [13].
Quality control, data entry and analysis
Forms were checked by the supervisor in the field and inconsistencies verified with the respondents. Data were double entered by different entry clerks and compared for consistency using Census and Survey Processing System (U.S. Census Bureau Washington DC, USA). Statistical analysis was conducted using Stata™ 9.2 (Stata Corporation, College Station, Texas, USA). Sampling probabilities were calculated for kebeles, and sampling weights derived as the inverse of the product of sampling probabilities at the kebele level. Descriptive statistics were used to examine the characteristics of the sample. Differences in proportions between the regions were compared using chi-square test while t-test was used to compare means. Point estimates and confidence intervals were derived using the SURVEY (SVY) routine in Stata which controlled for clustering in the sample design as well as weighting for sampling probability [14].
Routine surveillance data
Three-year data on reported monthly clinical cases of malaria were obtained from the integrated disease surveillance (IDS) system from July 2004 to June 2007 (Ethiopia malaria statistics are reported by the MOH fiscal year July-June). In order to assess whether the survey was done in a representative year for malaria transmission (i.e. not epidemic or abnormally low transmission) we extracted the total annual numbers of cases from the IDS data for Oromia and SNNPR regions and expressed these as incidence/1000/year.
Ethical considerations
The protocol received ethical approval from the Emory University Institutional Review Board (IRB 1816) and the Oromia and SNNPR Regional Health Bureaus. Informed consent was sought in accordance with the tenets of the declaration of Helsinki. Verbal informed consent to participate in interviews was sought from the heads of the household. Signed informed consent for blood testing was sought from each individual and parents of children aged 17 years and younger. Personal identifiers were removed from the data set before analyses were undertaken.