Mapping Geographical Differences and Examining the Determinants of Childhood Stunting in Ethiopia: A Bayesian Geostatistical Analysis

Abstract

Understanding the specific geographical distribution of stunting is essential for planning and implementing targeted public health interventions in high-burdened countries. This study investigated geographical variations in the prevalence of stunting sub-nationally, and the determinants of stunting among children under 5 years of age in Ethiopia. We used the 2016 Ethiopia Demographic and Health Survey (EDHS) dataset for children aged 0–59 months with valid anthropometric measurements and geographic coordinates (n = 9089). We modelled the prevalence of stunting and its determinants using Bayesian geospatially explicit regression models. The prevalence of stunting among children under five years was 36.3% (95% credible interval (CrI); 22.6%, 51.4%) in Ethiopia, with wide variations sub-nationally and by age group. The prevalence of childhood stunting ranged from 56.6% (37.4–74.6%) in the Mekelle Special zone of the Tigray region to 25.5% (10.5–48.9%) in the Sheka zone of the Southern Nations, Nationalities and Peoples region. Factors associated with a reduced likelihood of stunting in Ethiopia included non-receipt of breastmilk, mother’s BMI (overweight/obese), employment status (employed), and higher household wealth, while the enablers were residence in the “arid” geographic areas, small birth size of the child, and mother’s BMI (underweight). The prevalence and determinants of stunting varied across Ethiopia. Efforts to reduce the burden of childhood stunting should consider geographical heterogeneity and modifiable risk factors.

Mapping Local Variations and the Determinants of Childhood Stunting in Nigeria

“…Children were considered stunted if their HAZ score was less than −2.0 standard deviations from the median of the WHO reference population for age. The stunting status was dichotomised as “stunted” (1) or “not stunted” (0), in line with the 2018 NDHS report [ 13 ] and previously published studies [ 22 , 30 ].…”

“…Using a conceptual framework adapted from the UNICEF [ 31 ] and WHO [ 32 ] frameworks for undernutrition [ 32 ] and used in past studies from LMICs [ 22 , 30 ], we broadly grouped explanatory variables into proximal and contextual factors ( Figure 1 ) [ 31 , 32 ]. For this study, child morbidity (including diarrhoea and acute respiratory infection (ARI)) was included as an immediate proximal factor, given the immediate pathophysiological relationships with stunting, where infection can lead to poor nutrient intake, absorption, or utilisation ( Table 1 ) [ 31 , 32 ].…”

“…The overall analytical framework and strategy for the describing the prevalence of childhood stunting at the second administrative level while examining relevant proximal and contextual factors associated with childhood stunting have been described in detail elsewhere [ 30 ].…”

“…We used a five-staged analytical strategy to implement the Bayesian geostatistical modelling, as reported in detail elsewhere [ 30 ]. In stage one, data on environmental and climatic factors were linked with household survey data using a ‘cluster id’ as a unique identifier.…”

“…Bayesian inference was chosen for its ability to consider prior information, make predictions for unsampled locations, provide credible intervals (CrIs), and handle complex models such as spatial and spatiotemporal models [ 36 ]. In the prediction model of the prevalence of stunting in high-resolution grids [ 30 , 37 ], the model used proximal and contextual factors with available raster surfaces, including maternal education, ANC visits, place of birth, type of toilet system, source of drinking water, aridity, number of wet days, DLST, and altitude. In stage five, model validation was conducted by grouping the total dataset into a “training set” (75% of the sample) and a “test set” (the remaining 25% of clusters) based on previously published studies [ 30 , 38 , 39 ].…”