Determinants of stunting among children under age five in Burundi: Evidence from the 2016–2017 Burundi Demographic and Health Survey (BDHS 2016–17)

Abstract Burundi has one of the highest prevalence of stunting in the world. This study aimed to identify determinants of stunting among children under age five in Burundi. A total of 4993 children with anthropometric measurements from the 2016–2017 Burundi Demographic and Health Survey were included in the study. Stunting was assessed from the height‐for‐age Z‐scores (HAZ). Logistic regression models were analyzed to identify demographic, maternal, child‐related, and structural variables that influence stunting. In total, 56.9% of children under age five were stunted, of those 31.0% moderately and 25.9% severely. Multivariable logistic regression indicated that older children, male children (adjusted OR (aOR) = 1.41, 95% CI 1.24–1.61), and children who were perceived as small (aOR = 2.00, 95% CI 1.55–2.59) or very small at birth (aOR = 2.37, 95% CI 1.57–3.59) were significantly more likely to be stunted. Moreover, children of single mothers, with lower levels of education, who were underweight at the time of the survey (aOR = 1.95, 95% CI 1.42–2.68), who had short stature (aOR = 3.76, 95% CI 2.50–5.66) or who conceived more than four children (aOR = 1.22, 95% CI 1.05–1.42) were more commonly stunted. Stunting was more prevalent in rural areas (aOR = 2.53, 95% CI 1.72–3.73) and in households with no access to improved types of toilet facilities (aOR = 1.27, 95% CI 1.10–1.45). The results of this study show that the prevalence of stunting in children under age five in Burundi is alarmingly high and underscores the urgent need for decisive and determined action.

neurological diseases play a decisive role in the emergence of undernutrition, as they have a negative impact on how nutrients are digested, absorbed, transported, and utilized (Batte et al., 2017;Black et al., 2008Black et al., , 2013Silverstein, 2018;UNICEF, 2015).
A sufficient supply of nutrients during pregnancy and the first 1000 days of a child's life are of crucial importance for the healthy development of a child's brain and its physical growth (Black et al., 2013;UNICEF, 2015). Acute and chronic episodes of undernutrition in this period increase the risk for impaired cognitive, psychological, and physical development (Akombi, Agho, Hall, Wali, et al., 2017;Black et al., 2013;UNICEF, 2015;Victora et al., 2008).
In addition, stunting significantly increases the risk of mortality and morbidity (Black et al., 2013;UNICEF, 2015). Black et al. estimated that in 2008, 2.2 million deaths and 21% of the disabilityadjusted life years (DALYs) years were attributable to stunting, acute wasting, and intrauterine growth retardation among children under age five (Black et al., 2008).
Stunting in early childhood has long-term effects that reach into adulthood. A systematic review found that undernutrition in early stages of life was significantly associated with shorter adult stature, fewer years of school education, lower income, lower offspring birthweight, and higher offspring perinatal mortality (Victora et al., 2008). Moreover, the review showed that children with episodes of childhood undernutrition had an elevated risk for increased blood pressure, fasting glucose levels, and blood concentration of low-density fatty acids (Victora et al., 2008). It is assumed that children who were chronically or acutely undernourished during childhood have a significantly higher risk of being overweight in adulthood, which in turn, increases the risk of cardiovascular, oncological, and metabolic diseases (UNICEF, 2015;Victora et al., 2008).
As these interdependencies demonstrate, high prevalence of stunting not only leads to devastating consequences for the affected children and their families but also hinders the sustainable economic and social development of countries.
Despite great efforts by the Burundian government and its development partners, prevalence of stunting of children under age five in Burundi is among the highest in the world (ENSNSAB, 2019).
The prevalence only slightly decreased in the last decade from 58%, to 55. 9% to 54.2% in 20109% to 54.2% in , 20169% to 54.2% in /20179% to 54.2% in , and 20199% to 54.2% in , respectively (ENSNSAB, 2019ISTEEBU et al., 2012ISTEEBU et al., , 2017. The adverse effects of climate change and the return of more than 100,000 political ref- ugees are expected to further aggravate the high level of food insecurity in Burundi in the future (UNHCR, UNDP, 2019). To be able to sustainably reduce the prevalence of stunting and to evaluate, prioritize, and implement future measures to combat undernutrition, scientific evidence on the factors associated with stunting is required.
The aim of this study is to identify the determinants of stunting of children under age five in Burundi and to provide key stakeholders, development actors, and policy makers with sufficient data to make informed decisions in their efforts to reduce the burden of stunting in Burundi.

| ME THODS
For this study, we analyzed data from the third Burundi Demographic and Health Survey 2016(MPBGP et al., 2016. The BDHS 2016-17 is a cross-sectional population-based household survey that assembled nationally representative data on various sociodemographic and health-related parameters. It was conducted by the Institute of Statistics and Economic Studies of Burundi in collaboration with the National Institute of Public Health and with technical assistance from ICF International. The data collection phase was carried out between October 2016 and March 2017, and it was funded in a joint effort between the Burundian Government and its development partners. In order to simplify statistical analysis, the collected data are provided in separate data files for different units of analysis. The data set analyzed in this study was derived from the Children Recode File (KR). This data file lists every child under age five that was born to interviewed woman within the selected households and contains general, socioeconomic, and health-related information about the child, mother, and household. We constrained the analysis to children under age five (i) whose mothers lived in the household and could provide child-related information, (ii) for whom anthropometric data were collected, and (iii) who were alive at the time of data collection.

| Dependent variables
The anthropometric indicator stunting was used to assess the nutritional status of the children. In the BDHS 2016-17, stunting was calculated by comparing the children's height with the median height of the WHO reference population of the same age. The deviance from the reference population is expressed in standard deviations and is referred to as a Z-score. In this analysis, stunting was coded dichotomously ("stunted" and "not-stunted") according to the benchmark recommendations of the WHO Nutrition Landscape Information System (WHO, 2010). Children with Zscores below −2 SD from the median of the reference population were classified as stunted. A distinction was further made between moderately (between −2 SD and −3 SD) and severely stunted (below −3 SD) children. Anthropometric measurements of height and weight of children in the BDHS were carried out within randomly selected households. Weight was measured using the "SECA 878 flat" electronic scale. Size was measured using graded measuring rods (Shorr Board®). Children over 24 months were measured in an upright standing position, while infants under 24 months were measured in a reclining position.

| Included variables
Factors associated with undernutrition were selected on the basis of the "UNICEF Conceptual Framework of the Determinants of Child Undernutrition" from 2015 (UNICEF, 2015). A literature review on undernutrition studies (Akombi, Agho, Hall, Wali, et al., 2017;Amare et al., 2019;Ettyang & Sawe, 2016;Khan et al., 2019;Mukabutera et al., 2016;Poda et al., 2017) based on Demographic and Health Surveys (DHS) data served to detect appropriate indicator variables associated with stunting that were obtained within the BDHS 2016-17. Child-related risk factors included the sex, age, birth order, perceived size at birth, the 14-day prevalence of fever, diarrhea, and acute respiratory infection (ARI), breastfeeding status, place of birth, anemia status, and the number of children sleeping under an insecticide-treated mosquito net. Maternal risk factors considered marital status, education, weight, stature, age at first birth, and the parity. On the household level, risk factors contained sex of household head, type of place of residence, region, and wealth quintile.
Almost all the selected independent variables from the BHDS 2016-17 were obtained from interviews with the children's mothers using ordinal or categorical survey scales. Exceptions were the determination of the anemia status that was assessed by drawing blood samples with self-retractable lancets from children 6 months and older and analyzing it in a photometer (ISTEEBU et al., 2017).
Children were considered anemic if they had hemoglobin levels <11.0 g/dL. If hemoglobin levels were <7.0 g/dL, children were severely anemic; between 7 and 11.0 g/dL, they were considered to be mildly/moderately anemic (Croft et al., 2018). Maternal height and weight were measured using the same anthropometric instruments to determine the height and weight of the children. The wealth index was constructed within the survey as a component measure that was derived from a number of different "easy-to-collect" proxy variables that reflect on the cumulative living standard of the household (Croft et al., 2018). Those proxy variables include information from household's ownership of a number of consumer items such as a car or television, dwelling characteristics such as flooring material, type of drinking water source, toilet facilities, and other characteristics that are related to wealth status (Croft et al., 2018). A detailed description of the variable names used from the BDHS 2016-17, their collection, response options, and the coding performed for the purpose of this study are shown in Appendix S1.

| Statistical analysis
The analyses were conducted using a weighted survey design to account for the two-stage sampling design. The sample design applies a specific weighting factor to every child, according to its primary sampling unit and strata. The application of weights ensures nationally representative results. For the analyses, the R package 'survey' was used (Lumley, 2019). Prevalence estimates and percent distributions were calculated for all variables. Uni-and multivariable logistic regression models on stunting were performed. Factors that were statistically significant in univariable analyses were included in the multivariable logistic regression models. Variation inflation factors were analyzed to assess (multi-)collinearity between independent variables. If two variables showed high collinearity, the more information-revealing variable was kept. The other one was excluded from multivariable analyses. Anemia was excluded from multivariable analyses because of missing data. p-values below .05 were considered significant. All statistical analyses conducted in this study were performed using R Studio 3.6.2. The characteristics of the sample population are described in Table 1. All children were almost equally distributed among the five age groups and both sexes. The majority of children were born in second to fourth birth order (53.8%) and were perceived as large or average sized at birth (85.6%). Fever, diarrhea, or symptoms of acute respiratory infections (ARI) occurred among 41.0%, 22.1%, and 6.7% of children in the 2 weeks preceding the survey, according to their mothers. Almost all children were born in health facilities (87.1%) and had ever been or were still being breastfed (99.3%). Approximately half of the children (48.0%) did not sleep under insecticide-treated mosquito nets and more than 60% had either moderate or severe anemia.

| Characteristics of the sample population
Of the children's mothers, roughly 93% were living with a partner and almost half had no formal education. Approximately 15% of the mothers were underweight and about 4% had a stunted height (<150 cm). The majority of mothers (71.8%) conceived their first child between the age of 18 and 24 years and little over 40% of the mothers had given birth to more than four children.
The households of the included children were predominantly male headed (81.2%) and situated in rural areas (91.2%). Almost all households were equally distributed across the 18 provinces and the five wealth quintiles, except for a slight underrepresentation of the richest quintile. While 81.0% of the households had access to improved sources of drinking water, only half of the households had access to improved toilet facilities. Table 2, the prevalence of stunting accounted for 56.9%, with 31.0% of the children being moderately and 25.9% being severely stunted. Table 3 summarizes the results of the univariable and multivariable logistic regression models. In the univariable logistic regressions, several child-, mother-and household-related factors were significantly associated with stunting. These were the child's age, sex, perceived size at birth, place of delivery, and anemia status. Furthermore, all included mother-related factors were associated with stunting, and of the household-related factors place of residence, wealth quintile, source of water, and the type of toilet facility. All these factors apart from anemia (because of missing values) and wealth quintile (because of collinearity with mother's education) were included and adjusted for in the multivariable logistic regression model. Multivariable logistic regression indicated that children, in the four age groups from 12 to 59 months had significantly higher odds of being stunted compared to children between 0 and 11 months, with children in the age groups of 24-35 (adjusted OR (aOR) = 4.09, 95% CI 3.29-5.08) and 36-47 months (aOR = 4.48, 95% CI 3.59-5.59) showing the highest odds.

| DISCUSS ION
In this study among children under age five in Burundi, we found a prevalence for stunting of 56.9%, with 31.0% of the children being moderately and 25.9% being severely stunted. Several child-, mother-, and household-related factors were associated with stunting. Stunting was more common among boys, increased with age, peaking between 3 and 4 years, and was associated with small size at birth. Mothers who were underweight, short, single, or separated, and who had no, or little education were more likely to have stunted children. Generally, stunting was more common in rural areas, decreased with wealth of the household, and was more common in households with no access to improved toilet facilities.
Despite efforts to tackle undernutrition, the prevalence could not be reduced notably in the past few years and remains at an alarming level. Between 2010 and 2016-2017, the prevalence of stunting only decreased by 1.8%, and is considerably higher than in bordering countries like Tanzania, Rwanda, and the Democratic Republic of Congo with 34%, 38%, and 43%, respectively (Ministry of Health et al., 2016;MPSMRM et al., 2014;NISR et al., 2015). According to a meta-analysis based on the most recent DHS data of 32 Sub-Saharan African countries, Burundi even shows the highest prevalence of stunting in all of Sub-Saharan Africa (SSA) .
We found several factors to be associated with the risk of stunting. Those include sociodemographic, maternal, child-related, and structural factors.

| Maternal factors associated with stunting
Maternal underweight was found to be a predictor for stunting in this analysis, as children of underweight mothers had significantly higher odds of being stunted compared to children of overweight mothers. Similar results could be observed in several studies based on DHS data across the globe (Akombi, Agho, Hall, Wali, et al., 2017;Khan et al., 2019;Poda et al., 2017). We further observed that children of mothers with primary or no education had significantly higher odds of being stunted compared to children of mothers with secondary or higher education. Maternal educational attainment (and wealth) is highly intertwined with various factors that positively affect the nutritional status of both, mother and child through improvements in income, maternal selfdetermination, dietary diversity and knowledge, hygiene awareness, child disease management, and access to healthcare and childcare services (Akombi, Agho, Hall, Wali, et al., 2017;Amare et al., 2019;Khan et al., 2019).
Elevated parity increased the odds of stunting in this analysis and is, moreover, associated with lower maternal education (Keats, 2018).
Similar results could be observed in several cross-sectional studies (Islam et al., 2013;Poda et al., 2017;Wong et al., 2014). With a growing number of children in a household, the risk of stunting accumulates, possibly because of the financial burden that is posed on households with an increasing number of children. These results support the current scientific understanding that a decrease in the prevalence of stunting can be accomplished through a reduction of maternal parity (Giroux, 2008).

| Household factors associated with stunting
In Burundi, undernutrition is more common in rural areas. Similar   Wali, et al., 2017;Nkurunziza et al., 2017). However, there are also countries with contrary results such as Iran and Pakistan (Kavosi et al., 2014;Khan et al., 2019). In Burundi, the vast majority of people live in rural areas, most of which are densely populated. Thus, negative consequences of urbanization may be present in the rural areas as well. Households in urban areas, moreover, are less dependent on agriculture and tend to be employed in more economically profitable occupations (Bernoussi Marlène Kanga et al., 2013). This association may also be reflected by the low prevalence of stunting in Bujumbura Mairie which is mainly urban.
Regarding regional differences, prevalence of stunting was significantly higher in the northern, compared to the southern regions. This association could be caused by the higher population density in the northern regions (World Bank Group, 2018).
Moreover, the southern regions of Burundi are less populated, at lower altitude, and flatter which is favorable for agriculture and competition for arable land. Furthermore, Lake Tanganyika in the southwest is a source for fishing and offers opportunities for irrigation of plants.
The importance of sustained food supply is also demonstrated by the fact that children are less often undernourished if they are living in wealthier households where food scarcity occurs less often. In the context of the DHS, the wealth index of a household is, in part, determined by the availability and type of a toilet facility. The lack of improved toilet facilities was associated with a higher risk for stunting in multiple studies in SSA and South Asia (Akombi, Agho, Hall, Wali, et al., 2017;Amare et al., 2019;Fink et al., 2011;Rah et al., 2015). The lack of improved toilet facilities increases the risk of children being exposed to fecal bacteria that can cause diarrheal diseases and intestinal worm infections and thus increases the susceptibility for stunting (Prüss-Üstün, 2008;Rah et al., 2015). b Not included in multivariable analyses because of multicollinearity with mother's education.

TA B L E 3 (Continued)
to overcome the aftermath of the civil war that struck the country between 1992 and 1999 and the subsequent political turmoil (World Bank Group, 2018). Another important factor explaining the prevalence of stunting is the high level of household food insecurity that affects over half of the population (World Bank Group, 2018). The influence of household food insecurity is also reflected in the strong association of infant birth weight, maternal stature, and maternal weight on stunting. A sufficient supply of nutrients during the pregnancy and the first 1000 days of a child's life are of crucial importance for the manifestation of stunting (Akombi, Agho, Hall, Wali, et al., 2017;UNICEF, 1990UNICEF, , 2015  survey could have resulted in additional bias, as this could mean that particularly disadvantaged children were not included in the analysis (Giroux, 2008). A general problem with surveys is that they are prone to recall bias. This form of bias may not have had a large effect in this study, as most questions were answered by mothers and did not require long recall periods. Despite the questioning of the households in the national language Kirundo, incorrect information may have arisen due to misunderstandings in households that speak indigenous or tribal languages. Furthermore, measuring the height of children is also susceptible to measurement bias, particularly for children under 6 months of age where measurement was done in a reclining position (Amare et al., 2019). A further limitation of this study is that no information on dietary intake and household food insecurity was available. Future research is needed to address the influence of food insecurity and dietary intake on stunting.

| CON CLUS ION
The results of our analysis show that more than half of the children under age five in Burundi are affected by stunting. Despite great efforts by the Burundian government and its development partners, the burden of stunting has not been reduced significantly in the last decade. The many determinants for stunting observed in this study, including child, maternal, and household factors, accentuate that stunting is influenced in multifactorial ways. Thus, it is of crucial importance that measures consider the multifactorial origin of stunting.
Our findings further suggest that stunting is significantly associated with variables that reflect the high food insecurity within the country. project administration (supporting); supervision (supporting); validation (supporting); writing -review and editing (supporting).

ACK N OWLED G M ENTS
We would like to thank Gunther Schauberger, Chair of Epidemiology, Department of Sport and Health Sciences, TUM for his support in the statistical analyses. Open Access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTE R E S T S TATE M E NT
All authors involved in this study declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data and protocol of the BDHS 2016-17 were retrieved from the DHS Program repository and are publicly available upon reasonable request: https://dhspr ogram.com/data/datas et/Burun di_Stand ard-DHS_2016.cfm?flag=1.

E TH I C S S TATEM ENT
Since this study is a secondary analysis, no additional ethical approval was required. Prior to the start of the BDHS 2016-17, ethical approval was obtained from the ICF Institutional Review Board and of the National Council of Statistical Information of Burundi.