Drivers of stunting reduction in Ethiopia: A country case study

Background: Chronic undernutrition in children continues to be a global public health concern. Ethiopia has documented a significant decline in the prevalence of childhood stunting, a measure of chronic undernutrition, over the last 20 y. Objectives: The aim of this research was to conduct a systematic assessment of the determinants that have driven child stunting reduction in Ethiopia from 2000 to 2016, focused on the national, community, household, and individual level. Methods: This study employed both quantitative and qualitative methods. Specifically, a systematic literature review, retrospective quantitative data analysis using Demographic and Health Surveys from 2000–2016, qualitative data collection and analysis, and analyses of key nutrition-specific and -sensitive policies and programs were undertaken. Results: National stunting prevalence improved from 51% in 2000 to 32% in 2016. Regional variations exist, as do pro-rich, pro-urban, and pro-educated inequalities. Child height-for-age z score (HAZ) decomposition explained >100% of predicted change in mean HAZ between 2000 and 2016, with key factors including increases in total consumable crop yield (32% of change), increased number of health workers (28%), reduction in open defecation (13%), parental education (10%), maternal nutrition (5%), economic improvement (4%), and reduced diarrhea incidence (4%). Policies and programs that were key to stunting decline focused on promoting rural agriculture to improve food security; decentralization of the health system, incorporating health extension workers to improve rural access to health services and reduce open defecation; multisectoral poverty reduction strategies; and a commitment to improving girls’ education. Interviews with national and regional stakeholders and mothers in communities presented improvements in health service access, women and girls’ education, improved agricultural production, and improved sanitation and child care practices as drivers of stunting reduction. Conclusions: Ethiopia’s stunting decline was driven by both nutrition-specific and -sensitive sectors, with particular focus on the agriculture sector, health care access, sanitation, and education. Am J Clin Nutr 2020;112(Suppl):875S–893S. This study involved 4 methods of inquiry including a systematic literature review, retrospective quantitative data analysis, qualitative data collection and analysis, and policy and program analyses. The time period covered was from 2000 to 2016, at three 5-y intervals (2000–2005, 2005–2011, 2011–2016). A conceptual framework was adapted from the 1995 UNICEF Nutrition Framework and the 2008 Lancet Nutrition framework, and informs the quantitative and qualitative data analyses. This framework identifies the key distal-, intermediate-, and proximal-level factors that potentially contributed to Ethiopia’s stunting decline (Figure 2). Ethics approval for the study, inclusive of primary data collection, was obtained through the University of Addis Ababa’s research ethics process. Ethics approval for the broader stunting case study was also obtained through the Research Ethics Board at the Hospital for Sick Children (SickKids), in Toronto, Canada. Conceptual framework showing distal, intermediate, and proximal determinants of stunting in Ethiopia. Framework reflects only indicators that were measurable and available for quantitative analysis. Skilled birth attendance was omitted because estimates were not available for the first year of study, and were only available for the last 2 y studied (2011 and 2016). Flesh foods refers to meat, fish, poultry, and liver/organ meats. Data on early initiation of breastfeeding were also unavailable and were omitted. Other variables included in analyses were child age in months, child sex, child weight, and child height. DPT3, 3 doses of Diphtheria-tetanus-pertussis vaccine. The systematic literature review resulted in 10,789 total articles being found, and consisted of a search for peer-reviewed literature, as well as gray literature, published between 1990 and 2019. Search terms used included “stunting” or “linear growth” or “linear growth stunting” or “HAZ” or “height” or “height-for-age” or “LAZ” or “length” or “length-for-age” or “undernutrition” or “malnutrition” or “nutr*” AND “child*” or “infan*” AND “Ethiopia*.” The initial database search found 10,789 records, and after duplicates were removed and records were screened, a total of 150 were included in our analysis, of which 5 were systematic reviews, 124 were quantitative analyses, and 21 were gray literature. More detailed information on the search terms, as well as the literature review flow diagram, can be found in Supplemental Appendix 2. The sources of data used for the quantitative analyses were 4 Demographic and Health Surveys (DHSs) from 2000, 2005, 2011, and 2016. Table 1 shows the sample size breakdown of under-5 children with available anthropometry data for these 4 surveys. Sample size by survey based on the index child with valid anthropometric data The main study outcomes included child height-for-age z score (HAZ) and stunting prevalence (HAZ < −2SD), as they were estimated from the WHO child growth standards (144). In line with Figure 2, covariables were selected as they were available in DHSs (individual/household variables) and the Agricultural Sample Survey conducted by Ethiopia's Central Statistical Agency (ecological variables at district level). We grouped potential stunting determinants hierarchically as distal, intermediate, and proximal factors that align with “basic causes,” “underlying causes,” and “immediate causes” in Figure 2. Child HAZ kernel density plots were estimated for all survey years in order to examine population shifts in growth faltering over time. “Victora curves” or child HAZ-against-age plots were calculated using smoothed local polynomial regressions to enable examination of the growth faltering process from birth to 60 mo (145). We used piecewise linear splines to estimate the slopes and inflection points of the Victora curve growth trajectories (146). Equity analysis was conducted using standardized and well-established methods in order to study stunting prevalence disaggregated by wealth quintile, maternal education, area of residence (urban compared with rural), and child gender (147, 148). Wealth quintiles were derived using principal components analysis of household asset data. We also calculated the slope index of inequality (SII) and concentration index (CIX) in order to measure absolute and relative socioeconomic inequalities, respectively (147, 148). To assess the relative change (decline) in stunting prevalence of Ethiopia's districts, we calculated compound annual growth rates (CAGRs). A series of stepwise linear regression models and hierarchical modeling were used to conduct 2 sets of multivariable analyses between HAZ and distal-, intermediate- and proximal-level variables as suggested in the literature (149). The DHS 2000–2016 rounds were assembled into a panel data set to conduct difference-in-difference (DID) analysis with time × covariable interaction terms to understand if a change in a proposed predictor of HAZ led to a change in HAZ over the studied time period (150). Regression-based Oaxaca–Blinder decomposition methods were used to decompose the change in mean child HAZ between 2000 and 2016 into the covariables that drove this change. Full methods details are included in Supplemental Appendix 3 and the methods article in this series (Akseer et al.). All analyses were conducted with Stata version 14.0 (StataCorp LLC) and accounted for survey design and weighting. Key nutrition-specific and nutrition-sensitive policies and programs in Ethiopia were incorporated into a timeline using an iterative approach. Literature was identified through a systematic approach in order to suggest a timeline which was then shared with expert stakeholders for them to corroborate and provide insight. Any additional policy/program documents suggested by stakeholders were added to the second iteration of the timeline, and the process was repeated until consensus was reached. Further, rating was done by stakeholders and research team members to evaluate the relative importance of the policies/programs to the observed reduction in stunting. Qualitative research was conducted in 2019 using in-depth interviews with key informants at the national and community levels, and focus group discussions (FGDs) with mothers in communities. In-depth interviews were conducted with national stakeholders, including nutrition experts, representatives from multi- and bilateral international donor organizations, representatives of related ministries, and state agencies. Interviews with both national and community informants asked respondents to describe key nutrition-specific and -sensitive events in Ethiopia. National informant interviews also asked about successful factors and barriers to implementation. Community informant interviews with regional stakeholders, including teachers and health staff, explored how contextual factors such as changes in social and economic situations, access to key resources, and community-level changes to dietary intake affected child nutrition. FGDs with mothers who gave birth to children during 3 time periods (1987–1991; 1995–1999; 2011–2015) sought women's experiences, and perspectives regarding pregnancy, breastfeeding, child nutrition, and perceived drivers of change and general trends in children's nutrition. A total of 12 FGDs were conducted—with 3 age groups each in a rural and an urban area, in 4 districts of the SNNPR and Somali regions. Data generated in interviews and FGDs were analyzed using the UNICEF Nutrition Framework, the Lancet Nutrition framework, and the adapted framework for country case studies (Figure 2). These frameworks guided the qualitative analysis and interpretation of key determinants, contextual factors, and barriers and facilitators to nutrition-specific and -sensitive events. Responses from national- and community-level stakeholders were analyzed separately. Thematic analysis was used to explore key themes that emerged regarding stunting determinants. Full methods for qualitative data collection and analysis are available in Supplemental Appendix 4.