Trends in child growth failure among children under five years of age in Ethiopia: Evidence from the 2000 to 2016 Demographic and Health Surveys

Introduction In a majority of low- and middle-income countries (LMICs), levels of child growth failure (CGF) have steadily declined since 2000. However, some countries show a different trend. Despite continued investment from the government of Ethiopia as well as donors, CGF levels are still high in Ethiopia. This study aimed to assess trends in CGF and associated sociodemographic, economic and water, sanitation, and hygiene (WASH) factors from 2000 to 2016 in Ethiopia. Methods Data were taken from four rounds of the Ethiopia Demographic and Health Survey (EDHS). Children aged between 0 to 59 months were included. CGF indicators were categorised based on height-for-age z-score (HAZ) < -2 Standard deviation (SD), weight-for-age z-score (WAZ) < -2 SD and weight-for-height z-score (WHZ) < -2 SD. CGF trends were estimated for predicted probabilities and odds ratios (ORs) between 2000 and 2016. Results A total sample size of 31978 for HAZ, 32045 for WAZ and 32246 for WHZ were included in the current study. Stunting decreased from an adjusted odds ratio (AOR) = 0.77 (95% CI: 0.67 to 0.88) in 2005 to an AOR = 0.45 (95% CI: 0.39 to 0.53) in 2016 compared with the year 2000. Compared with data in 2000, underweight decreased from an AOR of 0.70 (95% CI: 0.61 to 0.80) in 2005 to an AOR of 0.43 (95% CI: 0.36 to 0.50) in 2016. Wasting declined from an AOR of 0.91 (95% CI: 0.75 to 1.10) in 2005 to an AOR of 0.76 (95% CI: 0.61 to 0.94) in 2016, compared with data in 2000. Conclusions Between 2000 to 2016, there was a decline in CGF levels albeit the levels are still relatively high compared with the World Health Organization (WHO) cut-off levels for public health concern. Observed rates of change varied across sociodemographic, economic and WASH factors which suggest that interventions tailored towards addressing the imbalances across those factors are required.

Introduction at providing education and/or supplementary foods. The implementation of this strategy occurred in two phases from 2008 to 2013 and then from 2013 to 2014 [16]. In 2015, Ethiopia declared the Seqota Declaration to end child malnutrition by 2030 by launching a Multi-Sectoral National Nutrition Program (NNP-II) aimed at guiding sectors and development partners in scaling-up nutrition interventions through the period from 2016 to 2020 [17]. Despite these programs, the country is lagging behind the expected trajectory needed to achieve the global nutrition target by 2025 and to end undernutrition by 2030.
Assessing patterns and trends of CGF indicators helps guide discussions about countrylevel progress and direct interventions to where they are most needed [5, 18,19]. Although country-level estimates are useful for international comparison and benchmarking [12], they may mask disparities in CGF across sociodemographic, economic, and environmental factors. Knowing the level of detail at local levels will provide important insights into the stalled progress to eliminate CGF within the country. It is anticipated the current study findings will help to inform future policy and program decisions towards ending CGF in Ethiopia. As such, this study aimed to assess trends in CGF and associated sociodemographic, economic and WASH factors from 2000 to 2016 in Ethiopia.

Data type
Data for this study were obtained from four rounds of EDHS: 2000, 2005, 2011 and 2016. Each survey was conducted in Ethiopia at 5-year intervals in all nine regions and in two city jurisdictions. Note, the gap between 2000 and 2011 publication was 6 years, due to slight delays in publication of the 2011 report rather than delays in data collection. This indicates the publication of report was a bit longer for 2011 survey compared with others. EDHS is a national representative and large-scale survey. Many countries have conducted the multiple Demographic and Health Surveys (DHS) to establish trends that enable decision makers at the national levels to gauge progress. Children aged between 0 to 59 months and mothers/caregivers 15 to 49 years were included in this study. All children with valid anthropometric measurements were included.

Sampling procedures of the EDHS
Two-stage stratified, and cluster random sampling techniques were used in the EDHS. In the first stage, enumeration areas or clusters were selected for each survey. The sampling frame used for the 2000 and 2005 EDHS was taken from the Population and Housing Census (PHC) which was conducted in 1994, while the 2011 and 2016 EDHS used the sampling frame from PHC 2007. Enumeration areas were selected from both urban and rural areas with a fixed number of households for each survey. Sociodemographic, economic, environmental and other disease related information for selected households were obtained by interviewing women. In addition, anthropometry was collected from children and from women who were interviewed. After identifying a nationally representative sample of households, a total of 64401 (sum of all surveys) eligible women aged 15-49 years were identified for individual interview. Of eligible women in each survey, the response rate was 97.8%, 95.6%, 95.0% and 94.6% in 2000, 2005, 2011 and 2016 EDHS, respectively. A pooled sample of the four surveys yielded an unweighted 43029 child-mother/caregiver pairs. In the current paper, we included 31978 children for HAZ, 32045 for WAZ and 32246 for WHZ analyses, respectively. The weighted sample yielded a total of 101415 children, including 29430 children in 2000, 12918 in 2005, 30120 in 2011 and 28947 in 2016. Detailed survey methods and sampling procedures are found in the respective EDHS reports [13,[20][21][22]. EDHS data were downloaded with permission from the DHS Program website using Integrated Public Use Microdata Series (IPUMS-DHS) extract system [23].

Data preparation and statistical analyses
Child growth failure was classified using HAZ, WAZ and WHZ as defined by WHO 2006 standard and classifications [8,9]. Child growth was considered as failure when child's HAZ, WAZ or WHZ fell below −2 SD from the median of the WHO reference population for a given age. Explanatory variables included: age; sex and birth order of the child; age of mothers at birth; maternal and paternal education; household wealth quintiles; geographic region; area of residence; and access to WASH facilities.
We pooled data from four EDHS datasets. Prior to performing statistical analyses, we checked for missing observations and for any small cell frequencies through cross-tabulation between explanatory and outcome variables. Categories deemed to have small cell frequencies were merged in ways that preserved common sense meanings for each category. We have included sampling weight, clustering and stratification variables which were provided by DHS to account for the complex survey design. The STATA command 'svyset' was used to declare the survey design while all estimations were performed by using survey-specific command 'svy.' A univariable logistic regression model was fitted with each explanatory variable to select candidates with p-values < 0.25 for the multivariable model. After which, a multivariable logistic regression model was fitted to estimate adjusted trends in ORs and predicted probabilities along with confidence intervals (CI). Interaction terms were fitted one at a time between survey year and each predictor variable. Threshold for statistical significance was declared at the outset as p < 0.05. All analyses were performed using STATA version 15. Table 1 presents the overall weighted sample (n) included in the analysis as well as prevalence of stunting, underweight and wasting across sociodemographic, economic and WASH factors. The overall prevalence of stunting (HAZ < -2 SD) was 57.45%, 50.85%, 44.34% and 38.39% in 2000, 2005, 2011 and 2016, respectively. The highest levels (48.81%) of child linear growth failure (stunting) were concentrated in rural areas. Stunting ranged from a high of 55.88% in the Amhara region to a low of 22.65% in Addis Ababa. About 50.38% of stunted children were born to mothers with no formal education which is in stark comparison with the 18.43% stunted children were born to mothers with higher educational. About 51.44% of all stunted children were found in the first wealth quintile and 50.32% were in the second wealth quintile. Children aged 24 to 59 months had a higher proportion of stunting (55.89%) compared with 0 to 5 months (15.73%) ( Table 1).

Overall prevalence of CGF by selected predictors
The percentage of children underweight (WAZ < -2 SD) was 41.13% in 2000, 33.18% in 2005, 28.80% in 2011and 23.73% in 2016. Of all underweight children, 39.56% were living in the Afar region. More than one in three (34.82%) underweight children were born to mothers with no education compared with 8.91% underweight children were born to mothers with higher educational status. As expected, the wealth quintile was inversely related to being underweight, with 36.64% in the lowest quintile and 21% in the highest quintile. Being underweight ranged from a high of 35.42% in children aged 24 to 59 months to a low of 12.59% in children aged 0 to 5 months. Overall, wasting (WHZ < -2 SD) in Ethiopia was 12.49%, 12.21%, 9.85% and 10.09% in 2000, 2005, 2011 and 2016, respectively. Wasting was highest in Somali (22.09%) while the lowest levels were observed in Addis Ababa (4.34%). The percentage Trends in child growth failure  Table). Children who had access to combined improved WASH showed a larger decline of 27 points in the predicted probability of stunting (S1 Table) and there was a slight wide gap in trend lines between 2005 and 2011 ( Fig 2H).

Child growth failure
Stunting showed a statistically significant reduction with survey years (Table 2). In Ethiopia, compared with data from the 2000, the odds of being stunted showed a decreasing trend from an AOR of 0.77 (95% CI: 0. 67  The odds of being stunted decreased with increases in maternal age in years and it was statistically significant for mothers aged 45 to 49 (AOR = 0.73 with 95% CI: 0.57 to 0.94) in 2016. The odds of being stunted decreased with an increase in wealth quintiles; 0.49 times lower in the highest wealth quintile compared with the lowest quintile in 2011 (AOR = 0.49; 95% CI: 0.33 to 0.73) and 47% lower in 2016 (AOR = 0.53; 95% CI: 0.35 to 0.80). In 2000, the odds of being stunted were 14% lower among female children compared with their counterparts (AOR = 0.86; 95% CI: 0.77-0.96). In 2016, children aged 24 to 59 months were 0.62 times less likely to be stunted compared with infants aged 0 to 5 months (AOR = 0.62; 95% CI: 0.41 to 0.94). In 2011, children in households with access to improved sanitation were 46% less likely to be stunted compared with children in households without improved sanitation (AOR = 0.54; 95% CI: 0.38 to 0.78). Also in 2011, children with access to improved combined WASH facilities were 71% less likely to be stunted (AOR = 0.29; 95% CI: 0.12 to 0.70) compared with children without access to improved combined WASH (Table 2).

Underweight
The predicted probability of being underweight has decreased by 17 points in urban areas  Table). The

Wasting
The predicted probability of wasting in the SNNP region significantly declined between 2000 and 2016 compared with other regions. Gambela region had the largest reduction in predicted probability of wasting from 0.21 in 2000 to 0.10 in 2005 compared with other regions (S3 Table).  (Table 4).

PLOS ONE
Child growth failure

Discussion
This paper showed evidence of declining trends in child growth failure at a national level between 2000 and 2016 in Ethiopia. The decline was greatest between 2005 and 2011 compared with other periods. Possible reasons for the difference in the rate of decline need to be considered. Many factors may have contributed to the overall decline, including economic growth [24,25] as well as the implementation of large-scale programs such as the Health Extension Program to improve access to health services, the Enhanced Outreach Strategy, Targeted Supplementary Food and the Community Management of Acute Malnutrition program [16]. However, a systematic review and meta-analysis which pooled studies conducted in various parts of Ethiopia failed to find a statistically significant association between economic growth and a reduction in child undernutrition [26]. On the other hand, evidence indicates that low family income and a poor living environment increases the risk of CGF, mainly due to inadequate access to health care as well as living in contaminated environments [26][27][28].
Although the downward trends in CGF are encouraging, the level of child growth failure in Ethiopia continues to be higher than the WHO cut-off levels for public health concern [29]. The international definition of 'normal growth' defines the first threshold, which includes 2.3% (very low prevalence of CGF) of the area under the normalized distribution [30]. As such 2.3% is used to establish subsequent thresholds for public health concern. In the current study, the decline in stunting was greater than the decline in children who met criteria for underweight or wasting. While the decline in rates of stunting in Ethiopia was comparable with other countries [31,32], such declines in CGF indicators were not equally distributed across the country.
The decline in all CGF indicators between 2000 and 2016 was relatively rapid in urban areas compared with rural areas in Ethiopia. Previous studies have found that urban children had lower odds of stunting and being underweight compared with their rural counterparts [33,34]. The observed association with urban areas could be attributed to better socioeconomic conditions, food security, maternal prenatal and postnatal care, access to quality

Handwashing
Improved n/a n/a n/a n/a 0.67 (0.30, 1.49) 0.325 0.91 (0.76, 1.08) 0.268 unimproved n/a n/a n/a n/a Ref Ref   complementary feeding, child immunization, WASH conditions, and fewer livestock resulting in cleaner environments and/or fewer infections. Region was another factor related to an unequal distribution in CGF indicators.

WASH
Since 2000, the predicted probabilities of having stunted children declined in all regions except in the Northern (Tigray and Afar between 2005 and 2011) and Eastern regions Harari and Dire Dawa. The analysis of trends in odds ratios also showed increased odds of stunting in Dire Dawa compared with other regions. The observed association could be attributed to regular drought conditions in the eastern part of Ethiopia [35]. Drought which causes water scarcity and crop failure would have negative impacts on food security and socioeconomic conditions in these areas. The current study shows a larger burden of stunting in the northern regions of the country which is consistent with previous studies [36][37][38]. The observed hotspots of stunting in the northern parts of the country are likely attributable to difference in child feeding practices [39,40] and nutrition knowledge of mothers or caregivers [37]. Also, increased rates of stunting in some areas is likely attributable to climate variability in zones that were dependent on rain-fed agriculture [36,[41][42][43], which has an impact on household food security.
In addition to the downward trend in the adjusted predicted probabilities, the odds ratios indicated non-linear decreasing in stunting, underweight and wasting with an increase in maternal years of education. Previous studies have shown a statistically significant association between the education level of maternal/caregivers with child stunting [26,36,44,45], underweight children [46] and child wasting [26]. These associations may be attributable to educated mothers/caregivers having greater access to and use of health services, healthier feeding behaviours and greater household income [47,48]. In the current study, the odds of stunting decreased with increases in maternal age, which is a similar result to a previous study that found children of mothers older than 20-24 years of age had a lower probability of stunting compared with younger mothers [49]. A possible explanation for this is that women with more life experience may be more aware of, and have greater capacity to be, meeting the needs of their children, through the provision of adequate food and health care practices, which were not assessed in the current study [50][51][52]. CGF showed variation across household wealth quintiles between 2000 and 2016. The decline in predicted probabilities by wealth quintiles was higher for child stunting and being underweight compared with child wasting. Wealthier households showed a greater reduction in the odds of stunting and being underweight compared with households in lower wealth quintiles. This finding is in agreement with previous studies [27,[53][54][55][56].
There was a notable difference in CGF indicators between boys and girls. Although the predicted probabilities of child stunting declined similarly for boys and girls between 2000 and 2016, the odds ratios showed consistent decreases in wasting for girls compared with boys. A possible explanation for these variations are the differences in morbidity rates between girls and boys, as well as having a higher proportion of preterm births in boys compared with girls [57][58][59]. For instance, poor diet leads to an increased risk of infections, and an infection has profound effects on child nutritional status. This shows a vicious cycle between infections and wasting. A severe infection can cause wasting in children which potentially leads to a loss of appetite [60]. As wasting aggravates, children become more vulnerable to infections [60]. There were also CGF differences by child age and WASH facilities.
Increased odds of stunting, being underweight and wasting were associated with increased age of children, which is consistent with Hagos et al. [36] and Nguyen et al. [61] for stunting. Available evidence shows that lower risk at the younger age may be due to the protective effects of being breastfed which mostly continues during the first year of life [61][62][63]. CGF appears to develop during the period of weaning and then rises sharply at a later age which may be due to cumulative undernutrition associated with poor child feeding practices, inadequate complementary food, household conditions, household food insecurity and interacting with unsafe environment [32,[64][65][66].
The predicted probabilities of being stunted or underweight between 2000 and 2011 decreased with access to improved water, sanitation and combined WASH facilities. This may be attributable to large-scale investments by the Ethiopian government in sanitation during these times which may have plausibly improved child growth [16,24,26,[67][68][69]. The current findings have shown that sanitation alone and integrated improved WASH facilities were associated with decreased odds of stunting, while improved sanitation facilities alone were associated with decreased odds of children being underweight. Earlier studies found that access to improved WASH, either combined and/or individual components of WASH facilities, contributed to the reduction in child growth failure [26,45,[70][71][72][73][74][75].
The current study used four large population-based surveys with large sample sizes that enable the findings to be representative of Ethiopia. However, the study had some limitations. As is often the situation in observational studies, the current study may be susceptible to recall and measurement biases. In the present study, it is difficult to make causal inferences and investigate temporal relationships between study outcomes and predictors. Unmeasured or residual confounding factors may significantly change the current estimates.

Conclusions
This study shows that while CGF levels have decreased between 2000 and 2016 in Ethiopia, these levels remain high. CGF levels in the country shows substantial variation across sociodemographic, economic, and environmental characteristics, indicating benefits for some groups more than others. This study provides insights into trends in prevalence, predicted probabilities and determinants to inform policy makers to focus on target populations with relevant intervention programmes.