Study settings and data source
Ethiopia is a country located in the Northeastern part of Africa (horn of Africa). It is the second most populous country in Africa, next to Nigeria. The country consists of 9 regions and 2 city administrations at the time of the original survey. Demographic and Health Survey (DHS) is a nationally representative household survey usually conducted in every five-year time interval. For this analysis, we used the 2019 Ethiopian Mini Demographic and Health Survey (EMDHS) data.
Study design and sampling procedures
A community-based cross-sectional study design was conducted in Ethiopia between March 21, 2019 and June 28, 2019 among 9,012 eligible women identified for individual interviews. However, interviews were done with 8,885 women, giving a response rate of 99%. The EMDHS uses a multistage stratified cluster sampling design to collect data from study participants. Initially, country is first divided into geographic regions, which are then further divided into urban and rural sectors. The required number of enumeration areas (EA) within each region were picked at random using the sampling frame designed for the 2019 Ethiopian population and housing census. The frame includes information on the EA’s location, type of dwelling (urban or rural), and the projected number of residential households. To make the sample comparable across regions, 25 EAs were picked from eight regions of the country, and 35 EAs were selected from each bigger region, namely Oromia, Amhara, and Southern Nations, Nationalities, and Peoples’ Region (SNNPR). Accordingly, a total of 305 EAs (93 in urban and 212 in rural areas) were chosen for the 2019 EMDHS using a probability proportional to their size. Following that, a household listing procedure was carried out in each selected EA. Finally, 30 households per cluster were selected using equal probability systematic selection from the previously produced list. The DHS guide provides additional information on survey sampling strategies [23].
Source and study population
Eligible women for the 2019 EMDHS include all women aged 15 to 49 who were either permanent residents of the selected households or guests who slept in the household the night prior to the survey. However, our research used all live children born in the two years preceding the survey as the source population. The study population consisted of all youngest children (midx = 1) living with their mother in the selected EAs and born 0 to 23 months before the survey. The last-born child data was collected if the mother have two or more under 24 months old children. However, mothers who had twin birth for the most recent birth were asked for both children. In this study, missing data is managed as not currently breastfeeding in numerator and included in the denominator. Consequently, we included 2,052 under 24 months children data extracted from the 2019 EMDHS datasets (KR file) that satisfy the inclusion criteria.
Eligibility criteria
This analysis included, all mothers who gave birth in the last two years before the survey. In addition, we considered all women who live with their children and were breastfed but not given anything else in the 24 h preceding the interview (Fig. 1).
Eligibility assessment for exclusive breastfeeding among women having 0–23 month’s children in Ethiopia, 2019
Measurement of variables
Dependent variable
The outcome variable of the study was exclusive breastfeeding practice among mothers with children under 24 months in Ethiopia. It was measured in two ways based on the age group of the children. For infants aged 0 to 5 months (< 6 months), mothers were asked if they breastfed and given nothing else except medications in the 24 h before the interview. Mothers of children aged 6 to 23 months were questioned if they breastfed and given nothing else while their children were under six months old [28, 29]. DHS used the information on exclusive breastfeeding collected from mothers’ verbal responses. Thus, the response variable was coded as “1” if the mother breastfed and provided nothing else in the day before the interview for children under 6 months, or if the mother responded as breastfed and given nothing else based on her dietary recall from birth for children aged 6 to 23 months and “0” otherwise.
Independent variables
The individual-level factors included in the analysis were maternal socio-demographic variables (such as maternal age, maternal education, marital status, religion, sex of household head, family size, and household wealth index); and infant related variables (like age of the child, sex of the child, birth order number, birth interval, number of living children, time of breastfeeding initiation, and duration of breastfeeding). And obstetric and health service related variables such as timing of 1st antenatal care (ANC), number of ANC visits, place of delivery, mode of delivery, delivery assistance, postnatal care (PNC), and age of respondent at 1st birth were also included.
Community-level factors like geographic regions, type of place of residence, community women education, community ANC utilization, and community poverty level were considered in this study (Table 1).
Data processing and analysis
We used STATA/SE version 14.0 to process and analyze the data. The study used descriptive statistics to report frequency and proportion of each independent variables. In the DHS survey, children are nested within households, which are then nested within clusters. As a result, children from the same clusters are more similar to each other in terms of the outcome variable as compared to those from different cluster. Individual observations in such data are not independent, which contradicts the assumption of independence of observations. Therefore, we used a multilevel mixed-effects logistic regression model to tackle the violation of independent observation and equal variance assumption of the traditional logistic regression that occurred due to the hierarchical structure of the DHS data. Hence, four models were fitted within the multilevel multivariable mixed-effects logistic regression analysis. The null model, without independent variables, was fitted first. This model was applied to test the null hypothesis that there is no cluster level difference in the outcome variable and to justify the use of multilevel analysis by obtaining the intra-class correlation coefficient (ICC). Secondly, a model with only individual-level factors was fitted. This model assumes that there is no difference in exclusive breastfeeding practice between clusters. Likewise, Model II (with only community-level factors) was constructed to assess community-level determinants using aggregate cluster variations in exclusive breastfeeding practice. Finally, mixed model, Model III, was fitted to test both the fixed and random effects of individual and community-level determinants on the exclusive breastfeeding practice. Variable with p-value < 0.25 were selected for the multilevel multivariable analysis. Finally, Adjusted Odds Ratio (AOR) with 95% confidence interval and p-value less than 5% was used to report statistically significant variables with the exclusive breastfeeding practice. We checked the goodness of fit of the model by Akaike information criterion (AIC), and Bayesian information criterion (BIC). AIC is computed as -2 (log-likelihood of the fitted model) + 2p, where p is the degree of freedom in the model. Similarly, BIC is calculated as -2 (log-likelihood of the fitted model) + ln (N)*p. The value of ICC of each model can be also calculated by using STATA software command (estat icc). The model with the lowest deviance (-2 log-likelihood) was selected as a best explanatory model. Also, multicollinearity amongst the covariates was examined using the Variance Inflation Factor (VIF).
Spatial analysis
The spatial analysis was carried out using ArcGIS 10.7 and SatScan 9.6. The weighted frequency of outcome variable with cluster number was cross tabulated using STATA software and exported to excel to get the case to total proportion. The excel file was then imported into Arc-GIS 10.7, and geographic coordinate data was joined with non-spatial data using each EA’s (Enumeration Area) unique identification code for spatial analysis. To produce the map of Ethiopia, the Ethiopian Poly-conic Projected Coordinate System was used. The units of spatial analysis were DHS clusters (Since geographic coordinates of EDHS were collected at cluster level).
Spatial autocorrelation analysis
To evaluate whether the pattern of outcome variable is clustered, dispersed, or random across the study areas, global spatial autocorrelation was assessed using the Global Moran’s-I statistics. Moran’s I is a spatial statistic used to measure spatial autocorrelation by taking the entire data set and produce a single output value which ranges from -1 to + 1. Moran’s I Values close to − 1 indicate the pattern is dispersed, whereas moron’s I close to + 1 indicates clustered and distributed randomly if the value is zero. A statistically significant Moran’s I (p < 0.05) lead to rejection of the null hypothesis (exclusive breastfeeding is randomly distributed) and indicates the presence of spatial autocorrelation and needs further local analysis. Anselin Local Moran’s I used to investigate at exclusive breastfeeding cluster locations at the local level, whether they were positively correlated (high-high and low-low) or negatively correlated (high-low and low–high). A positive value for ‘I’ indicated that a case had neighboring cases with similar values, part of a cluster. A negative value for ‘I’ indicated that a case was surrounded by cases with dissimilar values.
Host spot and cold spot analysis
Getis-Ord Gi* statistics was computed to measure how spatial autocorrelation varies over the study location by calculating Gi* statistic for each area. Z-score also computed to determine the statistical significance of clustering, and the p-value for the significance. The Getis-Ord Gi* statistic identified spatial clusters of high values (hotspots) and spatial clusters of low values (coldspots). Gi* serves as an indicator of local autocorrelation, i.e. it measures how spatial autocorrelation varies locally over an area and provide statistic for each data points. If z-score is higher, the intensity of the clustering is stronger and Z-score near zero indicates no apparent clustering. A positive z-score indicates clustering of high values and a negative z-score indicates clustering of low values.
Spatial interpolation
Spatial interpolation was done to estimate values for spatial locations with unknown value using known values. Among various deterministic and geo-statistical interpolation methods ordinary Kriging and empirical Bayesian are considered the best method since both incorporates the spatial autocorrelation and statistically optimizes the weight. For this study Ordinary Kriging spatial interpolation method was used to predict of exclusive breastfeeding in un-sampled areas in the country based on the value in sampled EAs.
Spatial scan statistical analysis
Spatial scan statistical analysis Bernoulli based model was employed to test for the presence of statistically significant spatial clusters of exclusive breastfeeding using SaTScan version 9.6. The spatial scan statistic uses a circular scanning window that moves across the study area. Women who give only breast milk for their child were taken as cases and those who give other food in addition to breast milk as controls to fit the Bernoulli model. The numbers of cases in each location had Bernoulli distribution and the model required data for cases, controls, population and geographic coordinates. For each potential cluster, a likelihood ratio test statistic and p-value was computed to determine whether the number of observed exclusive breastfeeding within the potential cluster was significantly higher than expected or not. The scanning window with maximum likelihood was the most likely performing cluster, and p-value was assigned to each cluster using Monte Carlo hypothesis testing by comparing the rank of the maximum likelihood from the real data with the maximum likelihood from the random datasets. The primary and secondary clusters were identified and assigned p-values and ranked based on their likelihood ratio test, on the basis of 999 Monte Carlo replications.
Ethical consideration
The MEASURE DHS team obtained ethical clearance from the Ethiopian Health Nutrition and Research Institute (EHNRI) Review Board and the National Research Ethics Review Committee (NRERC) at the Ministry of Science and Technology of Ethiopia. Thus, this study used secondary data from DHS data files. The authors formally requested the MEASURE DHS team to access the datasets by filling the online request form on their website (www.dhsprogram.com). Subsequently, the ICF international granted us permission to access the data and the letter of authorization. We kept all data confidential, and no effort was made to identify households or individuals. The authors also confirm that all methods were carried out in accordance with relevant guidelines and regulations.
Add Comment