Study settings
Ethiopia is the second most populous country in Africa and characterized by enormous diversity. The country has extensive altitudinal and geographic variations. The altitude ranges from 116 meters below sea level in the Danakil Depression to the peak of 4,620 meters above sea level on Mount Ras Dashen. The mean annual rainfall ranges from 500 mm to 2800 mm. Similarly, mean annual temperatures range from below 10 to above 30oC. In the submoist, moist, and sub humid highland areas, there is opportunity for agricultural growth. However, Agricultural production can be low in lowlands of Ethiopia, which is characterized by warm and a dry climate leading to food insecurity. In addition, Ethiopia is gifted with diverse culture with more than 80 Ethnic groups. Despite being one of the world’s poorest countries, Ethiopia’s economic growth is one of the fastest globally (15, 16).
Study Design
The data for the present analysis was obtained from the Ethiopian Demographic and Health Survey (EDHS) 2016. The EDHS is carried out every five years to provide health and health-related indicators at the national and regional levels in Ethiopia. The 2016 EDHS sample was selected using a stratified, two-stage cluster sampling design. In the first stage, 645 clusters of census enumeration areas (EAs), including 202 urban areas and 443 rural areas were selected. In the second stage, 18,008 households were selected. In this study we used data from the 9,268 children who had undergone anemia testing. Due to the non-proportional allocation of the sample to different regions and their urban and rural areas, we applied sampling weight to ensure the actual representative of the survey results at both the national and domain levels. We also applied complex survey design to account for the stratified multi stage sampling methods of EDHS. The detailed sampling procedure is presented in the EDHS report (6). The EDHS 2016 data were downloaded from the DHS website (http://dhsprogram.com) after we secured online permission. Potential predictor variables such as wealth index, educational level, BMI, age, residence (urban vs rural), region and other variables were extracted from the dataset. In addition, ecologic level variables such as temperature, malaria incidence, rainfall, and altitude were extracted from openly available DHS spatially interpolated datasets download from DHS Program Spatial Data Repository (http://spatialdata.dhsprogram.com).
Measurements
Blood specimens for anemia testing were collected from all children age 6–59 months from whom consent was obtained from their parents or another responsible guardian. Blood samples were drawn from a drop of blood taken from the palm side of the end of a finger and in the case of children age 6–11 months, blood was taken from the heel prick. The blood samples were collected on a HemoCue micro cuvette. Blood samples were placed in a HemoCue photometer and the results were recorded on site. Anemia status was defined as follows: mild anemia (10.0- 10.9 g/dl), moderate anemia (7.0- 9.9 g/dl) and severe anemia ( < = 7.0 g/dl). For the purpose of this study, the outcome variable anemia was recoded into a dichotomous variable where a child was considered to be anemic if the blood-hemoglobin count was less than 11.0 g/dl(6).
Data analysis
The statistical analysis was performed using the software packages STATA 14 and SaTScanTM. Descriptive statistics were used to analyze baseline characteristics of children and their caregivers including sex, age, residence, mother’s and father’s education level and wealth index to provide an overall picture of the sample. The prevalence of each risk factor and the 95% confidence interval were also presented.
Analysis Of Spatial Clustering
We made an attribute table containing information for each EA such as EA number, the number of children less than 5 years of age in each EA (population), proportion of anemia cases and EA coordinates. This file was imported into ArcGIS 10·1 for visualization. The visualization was made based on EA median Hb. Hb concentration of less than 110 mg/dl was considered as anemic and Hb concentration of greater than or equal to 110 mg/dl was considered as non-anemic. The coordinates’ projection was defined using the World Geodetic System (WGS) 1984, Universal Transverse Mercator (UTM) Zone 37°N. The shape file created was exported to the software SaTScanTM version 9·1·1 (http://www.satscan.org) for cluster analysis.
We conducted analysis of the spatial clustering of anemia in two steps. The first step aimed at examining the presence and locations of a significant cluster of anemia at national level. For examining the spatial clustering at the national level, we used the data from all regions (eleven) of Ethiopia. The second step aimed at detecting spatial clustering within each regions separately, and if present defined the characteristics of clusters such as size and location. We applied Kulldorf’s spatial scan statistics and used SaTScanTM version 9·1·1 to identify locations and estimate cluster sizes. The scan statistics evaluate whether proportion of anemia cases are distributed randomly over a defined space. If the process is not random, the scan statistics help to identify significant spatial clusters (17, 18). A circular window is used by the Kulldorf spatial scan to identify significant clusters with high cases of anemia over the study area. The statistical significance of this largest likelihood ratio was assessed through Monte Carlo simulation (1000 simulation performed). In order to detect both small and large clusters, we set the upper limit of the window size at 50% of the study population. The spatial relationships among EAs were conceptualized by calculating the spatial weights from the input file containing the proportion of anemia for each EA (the number of anemia cases divided by the total number of tested children in the EA) and the geo-coordinates data for each EA. We assumed that spatial autocorrelation for anemia declined with the distance and therefore a spatial weight matrix conceptualizing the spatial relationship between clusters was generated using an inverse distance approach. In addition to Kulldorf’s spatial scan statistics, we ran LISA (local indicator of spatial association) and the Getis-Ord Gi(d) local statistics to detect and locate clusters (hotspots) of anemia. LISA (local indicator of spatial association) indicates spatial autocorrelation for each location (19). The Getis-Ord Gi(d) statistic(20) was performed using ArcGIS 10.2 to identify the locations of clusters for high occurrence of anemia. The Gi(d) statistic performs the spatial analysis by looking at each cluster within the context of neighboring clusters.
Analysis Of The Determinants Of Anemia Clustering
Although identifying the presence of clustering was our primary objective, we performed further analysis to help identify the underlying process that governs the observed clustering. The observed clustering might be due to the underlying aggregation of known risk factors that are not randomly distributed geographically (or the presence of spatial dependency;‘Tobler’s first law of geography’) (21). We initially ran bivariate analyses to determine the potential risk factors of anemia. We used both individual and ecologic level data such as 1) individual variables: socio-demographic (child age, sex), child disease (fever, diarrhea), Dietary intake (dairy consumption, consumption of vegetables and fruits), child stunting and wasting, Maternal characteristics (women age, women anemia, women education, women BMI, number of ANC visits, iron consumption during pregnancy, number of births), household characteristics (improved water source, improved latrine, wealth status, residence), 2) ecologic level variables: rainfall, temperature, enhanced vegetation index and altitude. We used variables with p-value of less than 0.2 for the analysis of anemia clustering in the multilevel multivariable logistic regression. Regions and households were considered as levels. In the final model, variables such as women anemia, wealth, child stunting, child wasting, women, education, availability of improved toilet, number of births and amount of rainfall were included. We identified risk factors that varied across cases (anemic children) identified within the cluster and cases (anemic children) outside the cluster. A significance level of 0.05 was chosen for all the analyses.