Study design, setting, and period
A community based cross sectional study was conducted from January 18 to June 27 2016 (15). Ethiopia is situated in the Horn of Africa between 3 and 15 degrees of north latitude and 33 and 48 degrees of east longitude. It has great geographical diversity which ranges from the highest peak at Ras-Dashen (4,550 meters above sea level) down to the Afar Depression at 110 meters below sea level (16). Ethiopia is one of the least urbanized countries in the world; only 16 percent of the population lives in urban areas. The majority of the population lives in the highland areas. The main occupation of the settled rural population is farming, while the lowland areas are mostly inhabited by a pastoral people, who depend mainly on livestock production and move from place to place in search of grass and water. More than 80 percent of the country’s total population lives in the regional states of Amhara, Oromia, and SNNP (17). Ethiopia has nine regions and two administrative cities.
Population and sample
All under-five children in Ethiopia were the source population and all under-five children in selected EAs were the study population. Therefore, all under-five children in the selected clusters were included in the study. Respondents were excluded in case of flagging of either nutritional indices and/or absence of geographical location data. In 2016 EDHS 10,752 children aged 0-59 months in 645 EAs (202 in urban and 443 in rural areas) were eligible for height and age measurements. However, only 8855 children in 642 EAs were used for analysis. Hence, a total of 1897 children were excluded from the analysis because of incompleteness and misclassifications (8). A two-stage stratified cluster sampling technique was employed.
In the first stage, 645 EAs were selected with probability proportional to the EAs size and with independent selection in each sampling stratum with the sample allocation. The EA size is the
number of residential households (average 181 each) in the EA as determined in the 2007 PHC (8). In the second stage of selection, a fixed number of 28 households per cluster were selected with an equal probability systematic selection from the newly created household listing. This survey took only the pre-selected interviewed households of EDHS Geo-reference data. Based on a fixed sample take of 28 households per cluster, the survey selected 645 EAs, 202 in urban areas and 443 in rural areas. The survey considered 16,650 residential households, 5,232 in urban areas and 11,418 in rural areas (15) (Figure 1).
Figure 1: Schematic presentation of sampling procedure of spatial distribution and associated factors in Ethiopia, 2018 (15).
Variables of the study
In this study, the dependent variable, childhood stunting was defined as the percentage of children aged 0 to 59 months whose height for age is below minus two standard deviations (moderate and severe stunting) and minus three standard deviations (severe stunting) from the median of the WHO Child Growth Standards (18). The independent variables included: socio-demographic and economic factors (age, sex, residence, occupation, educational status, wealth index, and religion), geographical factors (region, cluster, and temperature), maternal health service utilization factors (ante natal care, place of delivery, and postnatal care), nutritional status of mother (BMI and HFA), birth weight, timing of breast feeding, clinical factors (anemic status of mother, anemic status of child), drinking safe water, and media exposure of respondents. Early initiation of breastfeeding –infants who are sucking the breast milk within one hour of birth (18). Introduction of solid, semi-solid or soft foods (6–8 months) – Percentage of children aged 6–8 months who received solid, semi-solid or soft foods during the past 24 hours (18).
Data collection and extraction procedure
The 2016 EDHS sample was selected based on a two-stage stratified cluster sampling and EAs are the sampling units for the first stage and households was used for the second stage of sampling. Height was measured for children 2 years and above using a measuring board and length was measured for children under the age of 2 years in lying down recumbent position (16). Weight measurements were obtained using light weight, SECA mother-infant scales with a digital screen, designed and manufactured. For all indices of childhood stunting, Z score less than -2 SD from the median of the WHO Growth Reference Population was considered as stunted. Geo-reference coordinates were collected using hand-hold GPS tool.
The EDHS team cleaned the data and calculated the Z-score for all childhood nutritional status indices. Childhood nutritional status data from the 2016 EDHS clusters were characterized by unique latitude and longitude location coordinates. Then, the EDHS cluster nutritional data and location file data were cross linked using the ArcGIS10.1 in making maps (19). The existing childhood nutritional indices Z-score was used to determine childhood nutritional status.
The data were obtained from the 2016 EDHS datasets, which are publicly available on the MEASURE-DHS program website (http://www.dhsprogram.com/Data). The Ethiopia DHS sub-national shape file was downloaded from the spatial data repository website of MEASURE-DHS program (http://www.spatialdata.DHSprogram.com) (20). Then Geo-reference data and other related data were extracted.
Data management and analysis procedure
The collected data was checked for consistency and missing. Descriptive and summary measures were done using STATA version 14 software. Means and percentages were used for continuous and categorical variables, respectively. ArcGIS10.1 was used to visualize the geographical distribution of cases across the regions.
In EDHS data, children within the cluster may be more or less similar to each other and may vary across clusters. Hence, the simple logistic regression model may not be appropriate because of violation of independence. Therefore, Generalized Estimating Equation (GEE) model was fitted to analyze factors associated with childhood stunting at two levels: individual and household levels. Exchangeable correlation structure with binomial logit probability distribution was used. in GEE. A P-value of less than 0.05 was used to define statistical significance. Adjusted odds ratios (AOR) with their corresponding 95% confidence intervals (CIs) were calculated to identify the independent predictors of childhood stunting.
Spatial autocorrelation and spatial scan analysis
The Global Moran’s I spatial autocorrelation test was used to identify the pattern of stunting whether it was dispersed, clustered, or randomly distributed in the study area. Accordingly, if the Moran’s I values was close to 0 indicated stunting is distributed randomly, whereas I value close to −1 indicated stunting was dispersed and I close to +1 indicated stunting was clustered. A statistically significant Moran’s I (p < 0.05) led to the rejection of the null hypothesis and indicated the presence of spatial autocorrelation.
The spatial scan statistical analysis was performed to identify significant clusters. ArcGIS10.3 software was used for identifying significant clusters. Accordingly, clusters were defined as positive autocorrelation (High-High or Low-Low autocorrelation) and negative autocorrelation (High-Low or Low-High). A scanning window that moves across the study area was used. Both primary and secondary clusters were identified using p-values and likelihood ratio test.
Children with stunting were taken as cases and those without the disease as controls to fit the model. Default maximum spatial cluster size of less than 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. To determine whether the number of observed stunting cases within the potential cluster were significantly higher than the expected or not Likelihood ratio test statistic was used. Z-scores and P-values were used to determine the significance of these statistics. A Z-score near zero indicated that no apparent clustering within the study area. A positive Z-score with P-value of less than 0.05 indicated that statistical clustering of hotspots of childhood stunting whereas a negative Z-score with p-value of less than 0.05 indicated statistical clustering of children who are normal. Finally, interpolation was done using Inverse Distance Weighted (IDW) techniques.