The MIS, a cross-sectional, nationally representative survey following RBM Monitoring and Evaluation Reference Group (MERG) guidelines, was conducted in the main malaria transmission season of 2007, from October–December. The principal objectives of the survey were to measure the coverage and use of malaria interventions, including ITNs, indoor residual spraying (IRS) and antimalarial drugs, as well as estimate the prevalence of fever, malaria parasitemia, and anemia. All enumeration areas (EAs) in the country with a median altitude below 2,500 m were stratified into below 1,500 m and between 1,500 m and 2,500 m altitude categories. EAs located between 1,500 m and 2,500 m were further stratified by urban and rural categories.

The details of the survey have been described previously in full [6], [7]. In brief, 319 EAs were selected based on probability proportional to size and then 25 households per each EA randomly selected. Personal digital assistants (PDAs) were used for the random sampling of households and recording of household interviews, malaria rapid diagnostic test (RDT) and hemoglobin results[30].

The questionnaires included a household and a women's sections, which were adapted to the local context from the MIS questionnaires developed by RBM MERG [5], [6]. The household questionnaire was administered to the head of the household to assess socio-economic characteristics and household coverage of malaria interventions. If women between the ages of 15–49 years were identified, the women's questionnaire which included their background characteristics, reproduction history, knowledge, attitude, and practices (KAP) around malaria, and fever history in children U5 was completed. Although the standard RBM MERG women's questionnaire does not include an assessment of knowledge, several countries have adapted the questionnaire to include this section [31], [32]. Women's knowledge of malaria was assessed by asking unprompted questions about the cause of malaria, signs and symptoms of malaria infection, signs and symptoms to suggest severe malaria, and ways to prevent malaria. The differences in answering unprompted questions (recall) compared to prompted (recognition) are well recognized and in general, response levels for prompted questions are higher, but less accurate [33], [34], [35]. Unprompted questions were used in this setting to measure recall as a more stringent and accurate measure of knowledge [33]. A woman's exposure to malaria educational information and, specifically, the setting and type of messaging heard were also assessed.

PDA data were downloaded into Microsoft Access 2003 (Microsoft, Redmond, WA, USA) and imported into SAS 9.2 (SAS Institute Inc., Cary, NC, USA) for data management and statistical analyses. ITNs were defined as any LLIN or a conventional net treated with insecticides in the past 12 months. ITN use was defined as reported sleeping under an ITN the night prior to the survey. Household wealth quintiles were derived from a principal components analysis of ownership of durable assets, housing materials (e.g. roof, wall, and floor), and type of access to water and sanitation. The first principal component with its associated variance was used to rank the households.

The knowledge section included one question each assessing the cause, symptoms, danger signs, and preventive measures of malaria. Each unprompted question had a differing number of potential correct answers and respondents were allowed to give more than one answer. Varying methods of calculating composite scores and categorizing them from quantitative knowledge data have increasingly been used [19], [33],[36],[37]. In the present study, a composite malaria knowledge score was calculated for each woman where every correct answer received a single point. This score was then dichotomized to women with no correct knowledge of malaria (score = 0) and those with any correct knowledge of malaria (score ≥1). Although a ranking of malaria knowledge derived from the principal components analysis showed a Pearson's correlation coefficient of 0.9 with this composite knowledge score (data not shown), the simpler to comprehend and calculate composite knowledge score was used to summarize women's knowledge.

Survey procedures within SAS adjusted for multi-stage sampling, clustering, and weighting for selection probabilities. Households with more than one eligible woman or mothers with more than one child U5 were taken into account in the overall weighting.

Descriptive statistics were used to explore the characteristics of all women and a subset of women who were mothers of children U5. Significant differences between the two groups were tested using chi-square and t-test for ordinal and continuous data, respectively. Univariate analyses of specific areas of malaria knowledge and receiving malaria information were performed, but not included in the multivariate model due to its correlation with the overall malaria knowledge variable. Multivariate logistic regression analyses were performed to assess women's malaria knowledge as it predicted household ownership of any ITNs and her own use of an ITN the night prior to the survey, with variables eliminated from the full model in order of least significance. ITN use and fever treatment in children U5 were also explored in terms of their mother's knowledge of malaria. All analyses of ITN use were restricted to households that owned at least one ITN.

Along with the explanatory variables, the following potential two-way interaction terms were included in the full model: malaria knowledge with household wealth quintiles, malaria knowledge with women's history of school attendance, women's school attendance with household wealth quintiles, and EA category (urban vs. rural) with school attendance. For terms that suggested the presence of an interaction (p<0.10), stratified analyses to assess the role of such effect modifiers were performed in order to interpret these associations in programmatically relevant terms.Results of the interaction terms are presented as p-values because of the difficulty of interpreting the odds ratio (OR) of an interaction term [38].

The MIS protocol received ethical clearance from the Emory University Institutional Review Board (IRB#6389), the CDC Ethical Review (IRB#990132), the Program for Appropriate Technology in Health (PATH) Ethical Committee, and the Ethiopian Science and Technology Agency. Verbal informed consent, which was approved by the above ethical boards and recorded in the PDA, was obtained from the heads of the household to participate in the household questionnaire, each eligible woman to participate in the women's questionnaire, and again from every individual prior to blood sample collection. Additional verbal assent was obtained from children 6 to 18 years of age.

Table 1 shows the comparative household and individual characteristics in the survey for all women and a subset of those women who were mothers of children U5. Data from a total of 6,657 women (5,949 households) which included 4,714 mothers of children U5 (3,447 households) were analyzed. Fewer of the latter lived in urban EAs (11.7%) compared to all women (17.2%). Mothers of children U5 also lived in larger households (5.5 mean number of household members) compared to all women (4.9 mean number of household members). Fewer (15.0%) mothers of children U5 lived in households of the highest wealth quintiles compared to all women (20.1%), while more (57.4%) mothers of children U5 lived in households that owned at least one ITN.

More women in general attended any school (23.7%) and had some level of literacy (15.7%) compared to women who were mothers of children U5 (16.9%; 12.5%, respectively). Although, 5,023/6,657 (73.8%) women had heard of malaria, only 2,953 (43.6%) stated that fever was a symptom, 2,233 (34.6%) mentioned mosquitoes as the cause, 2,296 (31.5%) reported that nets or ITNs could prevent malaria, and 2,731 (39.7%) named at least one danger sign of malaria. Overall 4,573 (67.6%) women and 3,193 (67.2%) mothers of children U5 answered at least one malaria knowledge question correctly, which was not significantly different between the two groups (Table 1).

Of the 1,800 women from 1,716 households who could not answer any malaria knowledge question correctly, fewer lived in urban EAs (8.4% versus 17.2% of all women, Table 1), belonged to highest wealth quintiles (13.0% versus 20.1%), or attended any school (13.1% versus 23.7%).

Of all women, only 25.4% (n = 1946) had seen or heard messages about malaria and even fewer 4.3% (n = 410) had received malaria information in their homes (Table 2). Friends and family (33.7%) and government clinics or hospitals (28.9%) were the most common sources of malaria information. Print media e.g. posters, billboards, and newspapers were cited least often. The importance of sleeping under a net or ITN (58.3%) and environmental sanitation (41.0%) were the most often heard messages.

Of the 5,949 households with eligible women, 3,463 (55.1%) lived in a household with at least one ITN (Table 1). Univariate analyses of specific knowledge areas showed that knowing mosquitoes as the cause (OR = 1.3 [95% confidence interval (CI) 1.0–1.7]), and nets/ITNs as a preventive measure of malaria (OR = 2.3 [95% CI 1.7–3.0]) were significantly associated with ITN ownership (Table 3). Having seen or heard malaria messages was not significantly associated with ITN ownership (OR = 1.4 [95% CI 1.0–1.9]) while receiving malaria messages at home was significant (OR = 3.1 [95% CI 1.9–5.3]).

Multivariate analysis showed that ITN ownership was significantly associated with any malaria knowledge (adjusted OR [aOR] 2.1 [95% CI 1.6–2.7]), EA altitude (aOR 0.3 [95% CI 0.2–0.5]), household sprayed with insecticide (aOR 4.6 [95% CI 2.6–8.2]), and an interaction term between EA category and school attendance (p = 0.03) (Table 4). Residence in an urban or rural EA category predicting ITN ownership depends on whether the woman attended any school.

Of the 6,657 women surveyed, 3,463 (33.7%) slept under an ITN the night prior to the survey (Table 1). Of the women who had slept under an ITN the night prior to the survey, 49.7% (1,084/2,359) slept under the same ITN as their child U5. Restricting all analyses to households with at least one ITN, knowing nets/ITNs could prevent malaria (OR = 1.5 [95% CI 1.1–1.9]) as well as receiving malaria messages at home (OR = 2.5 [95% CI 1.5–4.2]) were significantly associated with women's ITN use (Table 3).

The multivariate analysis which was also restricted to households with at least one ITN showed that women's use of ITNs were significantly associated with malaria knowledge (aOR 1.8 [95% CI 1.3–2.5]), having a child U5 (aOR 1.9 [95% CI 1.5–2.4]), woman's school attendance (aOR 2.0 [95% CI 1.1–3.9]), residence in an urban EA category (aOR 1.7 [95% CI 1.1–2.5]), household size (aOR 0.8 [95% CI 0.8–0.9]), household sprayed with insecticide in the past 12 months (aOR 1.8 [95% CI 1.3–2.5]), number of ITNs in the household (aOR 1.7 [95% CI 1.4–2.0]), and interaction terms between correct malaria knowledge and school attendance (p = 0.04) and between EA category and school attendance (p = 0.01) (Table 5). In assessing the interaction terms (Table 5), both malaria knowledge and EA category were significant predictors of woman's ITN use only if the woman had not attended any school.

1,876/5,225 (33.1%) of children U5 slept under an ITN the night prior to the survey [6], [7]. Similar to woman's use of ITNs, knowing nets could prevent malaria (OR = 1.5 [95% CI 1.0–2.1]) as well as receiving malaria messages at home (OR = 2.3 [95% CI 1.1–4.9]) were significantly associated with her child's ITN use (Table 3).

Restricting the multivariate analysis to households with at least one ITN, it showed that ITN use by children U5 was associated with their mother's malaria knowledge (aOR = 1.6; 95% CI 1.1–2.4), their mother having attended school (aOR = 4.4; 95% CI 1.6–12.1), living in an urban EA category (aOR = 1.8; 95% CI 1.1–2.8), living in a household sprayed with insecticide in the past 12 months (aOR = 1.5; 95% CI 1.1–2.2), increasing number of household ITNs (aOR = 1.6; 95% CI 1.3–2.1), and decreasing household size (aOR = 0.9; 95% CI 0.8–0.9) (Table 6). The restricted model also included a significant interaction term between any correct malaria knowledge and the mother having attended school (p = 0.04). Malaria knowledge predicting a child U5′s ITN use depends on whether his/her mother attended school. If the mother attended school, malaria knowledge was not a significant factor of U5 child's ITN use (p = 0.8); however, if the mother did not attend school, any correct knowledge of malaria was a significant factor associated with the child U5′s ITN use (p = 0.009).

Of the 22.3% (1,034/4,384) of children U5 with fever in the two weeks prior to the survey [6], [7], only 342/1,009 (36.5%) sought any type of treatment, which included public and private sector health facilities or hospitals, health extension workers, pharmacies, shops, and traditional healers. Knowing nets could prevent malaria (OR = 1.5 [95% CI 1.0–2.3]) and mosquitoes as the cause (OR = 2.3 [95% CI 1.5–3.5]) were significantly associated with seeking treatment, but knowing clinical features such as fever as a symptom or any danger signs were not significant.

In multivariate analysis, a mother of a child U5 having any correct malaria knowledge (aOR = 0.9; 95% CI 0.3–2.2) was not significantly associated with the child having sought treatment for a fever episode. However, increased treatment seeking was significantly associated with his/her mother having attended school (aOR = 6.5; 95% CI 1.9–22.9) and residence in an urban EA category (aOR = 4.0; 95% CI 1.9–8.3), while decreased treatment seeking was significantly associated with residence in an EA with altitude 2000–2500 m (aOR = 0.5; 95% CI 0.3–0.8) and the household owning at least one ITN (aOR = 0.5; 95% CI 0.3–0.9) (Table 7). The restricted model also included significant interaction terms between malaria knowledge and wealth index (p = 0.04) and between malaria knowledge and school attendance (p = 0.05). Again, malaria knowledge was a significant factor in predicting fever treatment seeking only amongst mothers who did not attend school. Albeit non-significant, the direction of association between malaria knowledge and fever treatment varied depending on the household wealth quintile.