Study design, setting, and period
We used cross-sectional data from Ethiopia Demographic and Health Survey (EDHS 2016) for this study. Ethiopia is a sub-Saharan African country with 1.1 million Sq. km coverage and the second-most populous country in Africa with an estimated population of 114,963,588 in 2021 . Administratively, Ethiopia is federally decentralized into two city administrations and nine regions . The datasets are publicly available from the DHS website www.dhsprogram.com . The surveys are nationally representative of the country and population-based with large sample sizes .
The source population was all preschool age children (aged 12–59 months) preceding five years of the survey period in Ethiopia whereas, the study population was preschool age children preceding five years of the survey period in the selected primary sampling unit (PSU). Mothers who had more than one child within the two years preceding the survey were asked questions about the most recent child .
Based on DHS recode manual, recent birth children who were died were excluded from the study. However, missing values and “don’t know” responses on whether the child took drugs for intestinal parasites in the last six months preceding the interview are included in the study but considered as not dewormed .
Weighted values were used to restore the representativeness of the sample data and calculated from children’s records or kid’s records (KR) EDHS 2016 datasets. Finally, a total weighted sample of 8146 children in the age category of 12–59 months was included in this study.
Using the 2007 Population and Housing Census (PHC) as a sampling frame, the EDHS used a stratified two-stage cluster sampling technique. Stratification was achieved by separating every eleven regions of Ethiopia into rural and urban areas. In total, 21 sampling strata have been created (except the Addis Abeba region which is only urban). Therefore, in the first stage, 645 Enumeration Areas (EAs) (202 in the urban area and 443 in the rural) were selected with probability selection proportional to the size of EA. In selected EAs, households (HHs) comprise the second stage of sampling. In the second stage, after listing the households, on average, 28 households have been selected using equal probability systematic sampling in the selected EAs . The detailed sampling procedure was available in each DHS report from the Measure DHS website .
The outcome variable of this study was taking deworming medication by preschool aged children. During the survey, their mother was asked questions about their under five years children who take drugs for intestinal parasites in the last six months preceding the interview .
Individual and community-level independent variables have been studied. The individual-level factors include socio-demographic characteristics such as; the age of the mother, mother employment, marital status, family size, maternal education, media exposure, and household wealth status were included. Child-related factors such as the age of the child, sex of the child, the plurality of birth, and birth order are all taken into account. Health service utilization-related factors such as place of delivery, pregnancy wontedness, and ANC visit were also considered. The community-level factors include; distance from health facilities, community media exposure, community poverty level, community women education, place of residence, and region were considered.
Media exposure was created from three variables; listening to the radio, watching TV, and reading newspapers. If a woman has at least one type of media exposure, she was considered exposed to media . Whereas, community-level media exposure was assessed using the proportion of women who had at least been exposed to one media; television, radio, or newspaper. It was coded as “0” for low (communities in which < 50% women had media exposure at least for one media), “1” for high community-level media exposure (communities in which ≥50% women had at least for one media [28, 29]. Community level poverty was also determined using the proportion of women in the poorer and poorest quintiles obtained from the wealth index results. It was coded as “0” for low (communities in which < 50% women had poor and poorest wealth quintiles), “1” for high (communities in which ≥50% women had poorest and poorer wealth quintiles) poverty communities [28, 29]. Community-level women’s education was also assessed by the proportion of women who had at least primary education. It was coded as “0” for low (communities in which < 50% women had at least primary education), “1” for high community-level women education (communities in which ≥50% women had at least primary education (at cluster level) [28, 29].
Based on the development status and the need for governmental support, the 11 regions of Ethiopia are categorized into three groups; large central (Tigray, Amhara, Oromia, SNNPR), “small peripherals” (Afar, Benishangul Gumuz, Gambelia and Somali), and ‘three Metropolis’ (Addis Ababa, Harari, and Diredewa) .
Data collection tools and quality control
Demographic and Health Survey (DHS) surveys collect data through different types of questionaries using interviewer administer questionnaire techniques. The missing values in the outcome variables were clearly defined by the DHS guideline . But variables that have a missing value greater than 5% in explanatory variables were dropped from further analysis since complete case analysis is a better missing data management in a crossectional study. The data extractions were performed by public health experts who have experience with DHS data to ensure the quality.
Data processing and analysis
This study was performed based on the DHS data obtained from the official DHS measure website www.measuredhs.comafter permission has been obtained via an online request by specifying the objectives. The standard DHS dataset was downloaded in STATA format then cleaned, integrate, transformed, and append to produce favorable variables for the analysis. Microsoft Excel and STATA 16 software were used to generate both descriptive and analytic statistics to describe variables in the study using statistical measurements.
Model building for multi-level analysis
Since the DHS data has hierarchical nature, children were nested within a cluster which violates the standard logistic regression model assumptions such as the independence and equal variance assumptions, a multilevel binary logistic regression model was fitted. Four models were fitted for multi-level analysis. The first was the null model (Model 1) which contained only the outcome variables. It is used to check the variability of deworming utilization across the cluster. The second (Model 2) and the third (Model 3) multilevel models contain individual-level variables and community-level variables respectively. In the fourth model (Model 4), both individual and community level variables were fitted simultaneously with the prevalence of deworming utilization. Model comparisons were done with the standard logistics regression model using the Log-likelihood and deviance test and the model with the highest log-likelihood and lowest deviance was selected as the best-fitted model. The variance inflation factor (VIF) was used to detect multicollinearity, and a variable that has a VIF result of 10 and above is regarded as indicating having multicollinearity . But in this study, all variables had VIF values less than five and the mean VIF value of the final model was 1.50. In the fixed effect measure of association, the variable which has significant association in Adjusted Odds Ratio (AOR) ratios was declared using a p-value of < 0.05 with 95% confidence intervals. The random effect used to measure the variation was estimated using the median odds ratio (MOR), Intra Class Correlation Coefficient (ICC), and Proportional Change in Variance (PCV) [29, 31, 32].
Global Moran’s I statistic spatial autocorrelation measure was used to assess the spatial distribution of deworming among preschool age children in Ethiopia . Getis-Ord Gi* statistic hot spot analysis was used to show significant cold spot area for deworming among 12–59 months of children. The proportion of children taking deworming medication among 12–59 month old children in each cluster was taken as an input for cold spot analysis. To predict deworming utilization among preschool children in Ethiopia for unsampled areas based on sampled clusters, the Inverse Distance Weighted (IDW) type spatial interpolation technique was used. Bernoulli based model spatial scan statistics were employed to determine the geographical locations of statistically significant clusters for not dewormed preschool aged children using Kuldorff’s SaTScan version 9.6 software . The scanning window that moves across the study area in which children who had not taken deworming medication were taken as cases and those children who had taken deworming medication were taken as controls to fit the Bernoulli model.