Study design and setting
A repeated cross-sectional study design was conducted in Ethiopia. Ethiopia is located in the Horn of Africa and bordered by Eritrea to the north, Djibouti, and Somalia to the east, Sudan and South Sudan to the west, and Kenya to the South. Ethiopia is home to about 13 million children under 5 years of age, approximately 16% of the total population . Ethiopia has 9 Regional states with two administrative cities. These are subdivided into different administrative Woredas and further divided into the smallest administrative units in the country called Kebele.
For this study, the 2019 Ethiopian Mini Demographic and Health Survey (EMDHS) dataset was used, the second EMDHS and the fifth DHS implemented in Ethiopia. The survey was conducted by Ethiopian Public Health Institute (EPHI) in collaboration with the Central Statistical Agency (CSA). The 2019 EMDHS generates data for measuring the progress of the health sector goals set under the Growth and Transformation Plan (GTP), which is closely aligned with the Sustainable Development Goals (SDG) . The survey was conducted from March 21, 2019, to June 28, 2019. Shapefiles were downloaded from the Africa open data website (https://www.africaopendata.org).
Sampling procedures and populations
The 2019 EMDHS was conducted by the CSA, and a complete list of the 149,093 enumeration areas (EAs), covering an average of 131 households, was created for the 2019 Ethiopia Population and Housing Census (EPHC). A two-stage stratified cluster sampling was used. Each region was stratified into urban and rural areas, yielding 21 sampling strata.
At the 1st stage of selection, a total of 305 EAs (93 in urban, 212 in rural) were selected independently with a probability proportion to each EAs. At 2nd stage of selection, a fixed number of 30 households/cluster were selected with an equal probability systematic selection from the newly created household listing . The detailed sampling procedures were presented in 2019 EMDHS report from the measure DHS website (https://www.dhsprogram.com). In this study, all living children aged 6–23 months were the source population, and all sampled living children aged 6–23 months living with their mother were the study population. Zero coordinates and clusters which had no a proportions of vitamin A rich foods intake were considered as an exclusion criteria.
Study variables and their measurements
Vitamin A rich foods intake among children aged 6–23 months is the dependent variable of the study which was determined by respondents’ reports and assessment of vitamin A rich foods [29, 30]. Vitamin A rich foods were measured by the seven food items such as 1. Eggs, 2. Meat (beef, pork, lamb, chicken), 3. pumpkin, carrots, and squash (yellow or orange inside), 4. fish or shellfish, 5. Any dark green leafy vegetables, 6. Liver, heart, and other organs 7. Mangoes, papayas, other Vitamin A fruits. Accordingly, if the respondent reported that the child had took at least one of those vitamin A rich foods item was considered as "Yes", otherwise "No" .
Potential predictor variables such as Sex of children, child Age (Month), baby postnatal checkups, Educational status of mother, Mother’s Age (Year), Religion of mother, Current Marital and Pregnancy status of mother, ANC visit and Place of delivery, and Wealth status, Mother exposure to media, Sex of household head were individual level independent variables. Whereas, Place of residency, and Region of mothers were taken as community level predictor variables for this study.
Recently, infant and child feeding practice is related to media (radio, Television) spots, and access to media may help to hear nutritional information, or messages . Therefore, mother who access to media offer a diversified diet to their children, and so considered mother had media exposure. Otherwise, mothers had not media exposure.
In this study, if the child’s mother had visited the health facility at least four times for ANC service during their pregnancy was considered as children’s mother had adequate ANC visit. Otherwise, inadequate ANC visit .
Data management and processing
Data cleaning, labeling, and processing was done using STATA version 15 software and Microsoft Office Excel. To yield accurate parameters estimation, and to handle representativeness of the survey, sampling weight was done. The descriptive analysis results were presented in table and text narrations.
Spatial data analysis
Global spatial autocorrelation and hot spot analysis
ArcMap version 10.7 software was used for spatial autocorrelation and detection of hot spot areas analysis. Global spatial autocorrelation (Global Moran’s I) statistic measure was used to assess whether vitamin A-rich foods intake among children was dispersed, clustered, or randomly distributed in Ethiopia . Moran’s I values close to minus one (-1), close to plus one (+ 1), and if it is zero (0) indicate a dispersed, clustered pattern and random distribution vitamin A rich foods intake among children aged from 6–23 months respectively [34, 35]. A statistically significant Moran’s 1 value (P value, 0.05) had a chance to reject null hypothesis which indicate the presence of spatial autocorrelation. Vitamin A rich foods intake among children with either hot spot or cold spot values for the spatial clusters are determined by the z scores and significant p-values of hot spot analysis [36, 37].
Vitamin A rich foods intake among children aged 6–23 months in the unsampled areas of the country were predicted by using the spatial interpolation technique. To predict Vitamin A rich foods intake among children aged 6–23 months in the unsampled areas, current vitamin A rich foods intake among children aged 6–23 months on sampled areas was used as an input. To minimize prediction uncertainty and filter out measurement errors, Ordinary Kriging Gaussian interpolation technique was employed. Based on the input data at each locations, semi variogram model was constructed, and used to define the weight that furtherly determine the prediction of new values at unsampled areas. As a result, a new simulated semi variogram model was generated [38, 39].
Spatial scan statistics
The Sat Scan version 9.5 software was used for the local cluster detection analysis . We employed purely spatial Bernoulli-based model scan statistics to determine the geographical locations of statistically significant clusters with high rate of vitamin A-rich foods intake among children . Those children who intake vitamin A rich foods were taken as cases and those who didn’t intake foods rich in vitamin A were taken as controls to fit the Bernoulli model for the scanning window that moves across the study area. The scanning window that moves outside the study area was clipped. The default maximum spatial cluster size of < 50% of the population was used as an upper limit, allowing both small and large clusters to be detected and ignored clusters that contained more than the maximum limit with the circular shape of the window. For each potential cluster, a log-likelihood ratio test statistic (LLR) was used to determine if the number of observed cases within the potential cluster was significantly higher than expected or not. The circle with the maximum likelihood ratio test statistic was defined as the primary cluster, then compared with the overall distribution of maximum values. The significant clusters were identified according to their p values and ranked based on their likelihood ratio (LLR) test based on the 999 Monte Carlo replications .
Multilevel logistic regression analysis
The assumption of independence among observations was violated due to the hierarchical and clustering nature of EDHS data. This implies a need to consider variability between-cluster by using advanced models since there are concerns which could not be addressed by basic logistic regression models. Four models were considered in the multilevel logistic regression: Model A = empty without out explanatory variable which examines vitamin A rich food intake without the explanatory variables that specified only the random intercept and the overall variance of vitamin A rich food intake among clusters, Model B = individual level variable, Model C = community level variables, Model D = both individual and community level factors. Measurement of variation and correlation also were determined using variance and interclass correlation (ICC). As a result, 28.4% of variance and 30.2% of ICC’s values confirmed that there were significant variations and correlations on vitamin A rich food intake among children aged 6–23 months in the country. Hence, multilevel mixed effect logistic regression analysis was fitted to assess both individual and community level variables. Finally, the model fitted was selected based on Akaike’s Information Criteria (AIC) and Log Likelihood Ratio (LLR). Variables having p value up to 0.2 in the bi-variable analysis were selected to fit the model in the multi variable analysis. Finally, p-value less than 0.05 in the multivariable model of mixed-effects logistic regression was used to select variables which had statistically significant association with vitamin A rich food intake.
Ethical approval and consents from study participant were not necessary for this study because the study was based on secondary data source which is publicly available from the Measure DHS program website (https://www.dhsprogram.com). After a clear working plan, and description on how to use the DHS data was written, request was sent to the Measure DHS program to download and used for this study. As a result, we obtained permission to access the EMDHS 2019 data through (https://www.dhsprogram.com/Date/terms-of-use.cfm) for statistical analysis and reporting. There are no attributes that uniquely identify individuals or household addresses in the data files. The geographic coordinate files are randomly displaced within a large geographic area, and it is only for EAs as a whole. As a result, specific ERs, individuals and households cannot be identified uniquely.