Menu
Menu
Menu
Menu
Background The distribution of under-five mortality (U5M) worldwide is uneven and the burden is higher in Sub-Saharan African countries, which account for more than 53% of the global under-five mortality. In Ethiopia, though U5M decreased substantially between 1990 and 2019, it remains excessively high and unevenly distributed. Therefore, this study aimed to assess geographic variation and factors associated with under-five mortality (U5M) in Ethiopia. Methods We sourced data from the most recent nationally representative 2019 Ethiopian Mini-Demographic and Health Survey for this study. A sample size of 5,695 total births was considered. Descriptive, analytical analysis and spatial analysis were conducted using STATA version 16. Both multilevel and spatial analyses were employed to ascertain the factors associated with U5M in Ethiopia. Results The U5M was 5.9% with a 95% CI 5.4% to 6.6%. Based on the multivariable multilevel logistic regression model results, the following characteristics were associated with under-five mortality: family size (AOR = 0.92, 95% CI: 0.84,0.99), number of under-five children in the family (AOR = 0.17, 95% CI: 0.14, 0.21), multiple birth (AOR = 14.4, 95% CI: 8.5, 24.3), children who were breastfed for less than 6 months (AOR = 5.04, 95% CI: 3.81, 6.67), people whose main roof is palm (AOR = 0.57, 95% CI: 0.34, 0.96), under-five children who are the sixth or more child to be born (AOR = 2.46, 95% CI: 1.49, 4.06), institutional delivery (AOR = 0.57, 95% CI: 0.41, 0.81), resident of Somali and Afar region (AOR = 3.46, 95% CI: 1.58, 7.55) and (AOR = 2.54, 95% CI: 1.10, 5.85), respectively. Spatial analysis revealed that hot spot areas of under-five mortality were located in the Dire Dawa and Somali regions. Conclusion Under-five mortality in Ethiopia is high and unacceptable when compared to the 2030 sustainable development target, which aims for 25 per 1000 live births. Breastfeeding for less than 6 months, twin births, institutional delivery and high-risk areas of under-five mortality (Somali and Dire Dawa) are modifiable risk factors. Therefore, maternal and community education on the advantages of breastfeeding and institutional delivery is highly recommended. Women who deliver twins should be given special attention. An effective strategy should be designed for intervention in under-five mortality hot spot areas such as Somali and Dire Dawa.
Our source was the 2019 EMDHS, the second mini demographic health survey (DHS) conducted in Ethiopia (a land-locked country located in the Horn of Africa that lies between the 30N and 150N Latitude or 330E and 480E Longitude) [21]. Data collection was conducted from March 21, 2019, to June 28, 2019, the nine regions (Tigray, Afar, Amhara, Oromia, Somali, Benishangul Gumuz, Southern nation nationalities and People region (SNNPR), Harari, and Gambella) and two administrative cities (Addis Ababa and Dire Dawa). The study design was a population-based cross-sectional study. A frame of all census Enumeration areas (EAs) was used as a sampling frame for the 2019 EMDHS. 149,093 EAs were created which cover an average of 131 Households (HHS). A two-stage stratified cluster sampling technique was employed and each region was stratified into urban and rural areas, yielding 21 sampling strata were selected independently in each stratum. In the first stage, 305 clusters (93 urban and 212 rural) were selected with probability proportional to EAs size and with independent selection in each sampling stratum. In the second stage, a fixed number of 30 households per cluster was selected. Finally, women aged 15–49 in 9,150 (2,790 urban and 6,360 rural) households from 305 clusters were selected. The whole procedure of sampling is found in the full 2019 EMDHS report [21]. The outcome variable was under-five mortality status, which was categorized as (child alive: Yes = 0 and No = 1). The age was recorded in months. The community-level predictors, were place of residence, region, community place of delivery, community wealth, community media exposure, and community toilet facility. The individual-level predictors were further categorized as socio-demographic and economic factors like the educational level of the mother, sex of the household head, age of household head, number of household members, number of children under the age of five, marital status of the mother, source of water, time to get water, type of toilet facility, household electricity, types of cooking fuel, main floor material, main wall material, and main roof material and maternal and child factors (such as maternal age at first birth, sex of the child), utilization of contraception, order of birth, mode of delivery, duration of breastfeeding and multiple births (Fig 1). The data for spatial analysis was cleaned and merged using STATA version 16 and Microsoft Excel. ArcGIS version 10.8 and saTScan version 9.7 were used for the spatial analysis. Spatial autocorrelation (Global Moran’s I) analysis was conducted to examine whether under-five mortality was dispersed (Moran’s I value closer to -1), clustered (Moran’s I value closer to 1), or randomly distributed (Moran’s I value of 0) in Ethiopia [22]. The under-five mortality was known in enumerated areas, while in areas that were not selected, the under-five mortality rates were predicted. Spatial interpolation was applied using the geostatistical ordinary Kriging spatial interpolation technique to predict under-five mortality from existing sample data points to un-sampled areas [23]. The scan analysis was performed using SaTscan, based on the Bernoulli test for cases (child is not alive) and controls (child is alive). The upper limit used was the default maximum spatial cluster size of less than 50% of the population, allowing both small and large clusters to be detected, while clusters that contained more than the maximum limit with the circular shape of the window were avoided [24]. Most likely clusters were identified using p-values and likelihood ratio tests, which is the ratio of the likelihood of the alternative hypothesis (higher activity level inside the window) over the likelihood of the null hypothesis (same activity level inside and outside). We used STATA version 16 and R statistical software version 4.0.5 to analyze the data. A total of 31 variables were retained for the analysis. Residents who were not De jure were excluded which affected under-five mortality, as they could not respond to most of the socio-demographic and economic characteristics even though they could answer the maternal and child characteristics. This exclusion changed our sample size from 5,753 to 5,695. The outcome variable was re-coded to (child alive: Yes = 1 and No = 0). Four community variables were generated by taking the individual variable, calculating their proportion, and dichotomizing them based on their mean or median according to their distribution. In the end, we had 29 predictor variables, of which 6 were community-level predictor variables and 23 were individual-level predictor variables. Based on EMDHS, respondents in the same cluster showed similar outcomes or functions at the same level and the data has a hierarchical structure. This made binary logistic regression not the most appropriate as it violates the assumption of independence of the residuals. Instead, a model that considers clustering effect should be used [7, 10]. Multilevel logistic modeling separates the within-cluster effects from the between-cluster effects [25]. Therefore, to assess the predictors associated with U5M, a non-weighted multilevel logistic regression model was used. Bivariable multilevel logistic regression was used to screen each predictor variable for a p-value less than 0.2. Significant variables were included in multivariable multilevel logistic models. Twenty predictor variables (3 of which were community-level predictor variables) were included in the multivariable analysis. In the multivariable analysis, a p-value less than 0.05 was considered a factor associated with U5M. The first model fitted was the null model (intercept model), which contained the outcome variable only (under-five mortality status) with the cluster number. The intra-cluster correlation (ICC) was used to assess whether there was a random effect. An ICC of 0.130 which meant there was a minimum of 13% under-five mortality was explained by between-cluster differences. We found that 87% of under-five mortality was explained by within-cluster differences, which was not negligible. The second model was fitted using the outcome, the cluster and the individual-level predictor variables only. The probability of U5M was predicted as a function of individual-level predictors. For the third model the outcome variable, the cluster number, and the community-level predictor variables were accounted for. Then the final model was fitted by taking both the individual-level and the community-level predictor variables into account. The models were compared by using a log-likelihood statistic, where the best model was selected based on smallest deviance. Permission for data access was obtained from a major demographic and health survey through an online request from http://www.dhsprogram.com. The data used for this study were publicly available with no personal identifiers.
