Factors associated with diet diversity among infants and young children in the Eastern and Southern Africa region

Abstract This study explores common factors associated with not meeting minimum dietary diversity (MDD) among 27,072 children aged 6–23 months in Eastern and Southern Africa using data from nine Demographic and Health Surveys from 2013 to 2016. MDD was defined as consumption of more than or equals to five of eight food groups including breast milk in the past 24 h. Equity gaps were calculated as the difference in MDD prevalence between the top and bottom wealth quintiles. Logistic regression was conducted to identify common factors for not meeting MDD at the household, maternal and child levels across two or more countries to inform regional policies to improve children's diets. Kenya had the highest MDD wealth equity gap (40.4 pts), and South Africa had the smallest (14.4 pts). Equity gaps for flesh foods or eggs (up to 39.8 pp) were larger than for grain or legumes (up to 20 pp). Common risk factors for not reaching MDD included younger child age (6–11 months) (n = 9 countries), no formal maternal occupation (n = 6), not receiving vitamin‐A supplementation (n = 3), younger maternal age (n = 3), lower maternal education (n = 3), no media (n = 3) or newspaper (n = 3) exposure, lower household wealth quintile (n = 3), use of nonefficient cooking fuel (n = 2), longer time to get to the water source (n = 2), not listening to the radio (n = 2) and higher birth order (n = 2). Priorities for improving MDD in the region include introducing diverse foods at a young age from 6 months with early nutrition counselling, promoting higher maternal education, increasing food purchasing power and ensuring the support of younger mothers.


| INTRODUCTION
The quality of diet during early childhood is a critical determinant of child's growth and development (Kuklina et al., 2004;Ruel, 2003;Miller et al., 2020;Prado et al., 2017). International guidelines recommend that children aged 6-23 months should continue to breastfeed and are introduced to age-appropriate complementary foods from diverse food groups (World Health Organization, 2003). Dietary diversity, particularly the inclusion of foods rich in iron and zinc, is particularly important for this age group due to physiological requirements of rapid growth, depletion of stores acquired during gestation and small gastric capacity (Beluska-Turkan et al., 2019;Dewey, 2013).
Minimum dietary diversity (MDD) in children aged 6-23 months is a simple qualitative indicator of the nutrient adequacy of the diet.
An MDD indicator for children aged 6-23 months based on consumption of at least four out of seven food groups has been validated against the micronutrient density of child diet across several contexts (Moursi et al., 2008;Working Group on Infant and Young Child Feeding Indicators, 2006). The standard definition was recently updated to facilitate comparisons across breastfed and nonbreastfed children. Breast milk was added as an eighth food group and the threshold for MDD was raised to five out of eight food groups (World Health Organization & United Nations Children's Fund, 2017). Children 6-23 months old in sub-Saharan Africa have some of the least diverse diets in the world (Choudhury et al., 2019).
According to 2019 estimates published by UNICEF, one in four children 6-23 months old in Eastern and Southern Africa (ESA) attains MDD, which is only slightly better than one in five in South Asia (UNICEF, 2019). In the UNICEF Eastern and Southern Africa region, there has been a renewed effort to strengthen the role that regional economic communities or bodies such as the Southern African Development Community (SADC) and the Intergovernmental Authority on Development (IGAD) play to support country-level efforts related to nutrition and food systems (Southern African Development Community, 2019).
Understanding the contextual factors that influence dietary diversity is important for informing action and programme design.
As previously wealth and place of residence have been found to consistently correlate with diet diversity through 80 DHS analyses, and disparities in wealth and place of residence do exist in ESA, it becomes imperative to sift already available information considering these socioeconomic inequalities (Gatica-Domínguez et al., 2021).
Socioeconomic variables such as wealth, maternal education and exposure to media were associated with MDD and minimum acceptable diet among 77,887 weighted samples in sub-Saharan countries (Belay, Aragaw, et al., 2022;Belay, Taddese, et al., 2022).
Differences also exist for urban versus rural residence: in a global analysis, the proportion of children from urban households reaching MDD was nearly twice that of rural children (39% vs. 23%) (UNICEF, 2021).
For selected countries in the ESA region, country-specific analyses have identified risk factors or determinants associated with low diversity (Belew et al., 2017;Custodio et al., 2019;Nkoka et al., 2018), but variability in methods limit the cross-country comparability of findings. There has been no systematic effort to date to identify common patterns in the risk factors of poor dietary diversity for children 6-23 months across ESA countries. Such an endeavour can be useful to guide regional efforts to share country learning and support countries with common challenges. This paper was prepared from a collaboration between United Nations International Children's Emergency Fund (UNICEF), John Hopkins University (JHU) and Global Alliance for Improved Nutrition (GAIN) focussed on the East and Southern Africa region, and was designed to develop the food policy guidance in those regions as a part of an effort to support cross-country policymaking by regional governance bodies such as SADC and Intergovernmental Authority on Development (EGAD) (Ryckman et al., 2021;White et al., 2021).
This study has two related aims; (1) to use an equity lens to examine how MDD and consumption of specific food groups vary by wealth quintiles and urban/rural residence across nine countries in the ESA region and (2) to explore the common child, maternal and household-level risk factors of not meeting MDD across countries.

| METHODS
Nine countries out of 22 countries in the ESA region were selected based on the availability of DHS data, diversity in economic and food system characteristics, political actor's interest and link with another set of research activities being conducted by GAIN (Ryckman et al., 2021). For our analysis, UNICEF programming priorities and the desire to capture a variety of food system typologies were considered as it was part of an effort to input into regional governance bodies (SADC, EGAD). From Eastern Africa, Ethiopia • A number of common risk factors predict dietary diversity across countries.
• Risk factors for low dietary diversity include wealth, education and media access.
• Wide equity gaps were apparent for most food groups in ESA.
• Interventions addressing affordability and access to diverse diets are needed. one available in each country at the time the study was conducted.
The DHS collects data on foods consumed in the past 24 h by the youngest child aged 6-23 months in each household. The selected countries and final sample size of children aged 6-23 months with MDD data are presented in Table 1.

| Analytical approach
Based on a review of the literature (Blaney et al., 2015;Stewart et al., 2013) and inputs from the UNICEF Eastern and Southern Africa Regional Office, we identified a list of available DHS at the child-(n = 10), maternal-(n = 29), household-(n = 9) or communitylevels (n = 10) (Supporting Information: Table 1). The investigative team developed a conceptual framework (Figure 1) describing the relationships potentially influencing complementary feeding practices at each of these levels. We selected 35 candidate variables of greatest relevance to our study based on prior studies and our knowledge/UNICEF programming interest (Supporting Information: Table 2). The list of these variables at each level is presented below: 2.2.1 | Child characteristics A total of six child-level variables were identified: age (0-5, 6-23 or 24-59 month); birth order (1st, 2nd-4th or ≥5th), episodes of diarrhoea/cough/fever in the past 2 weeks, and vitamin-A supplementation in the past 6 months.

| Maternal characteristics
The following 12 maternal-level variables were identified: age (15-24, 25-34 or 35-49 years), highest education level, occupation (employment in agriculture, no formal occupation, waged occupation), body mass index (BMI) (<18.5, 18.5-25.0 or ≥25.0 kg/m 2 ), number of antenatal visits during last pregnancy, places of recent delivery, delivery assisted by a health professional, watching TV, reading a newspaper, or listening to the radio at least once a week, exposure to any media (TV, newspaper or radio) at least once a week and women's empowerment. Maternal occupations included professional, clerical, sales, services and skilled and unskilled manual jobs. A score of women's empowerment (range: 0-5) was generated for each country using established methods, based on two domains of maternal participation in decision-making and overall attitude towards domestic violence (Jennings et al., 2014). The score was then dichotomised into low (<50th percentile) and high (>50th percentile).

| Community characteristics
The following three community-level variables were included; the proportion of women within the community who completed primary or higher education, the proportion of women within the community who gave birth at health facilities and the proportion of 'empowered women' within the community. The empowered women are those who belong to 'high (>50th percentile)' group, based on maternal participation in decision-making and overall attitude towards domestic violence. The unit of community was the cluster (primary sampling unit) in the DHS data set. Each community variable was divided into tertiles (low, medium and high).

| Statistical methods
For aim 1, the prevalence of MDD and consumption of specific food groups were estimated for the national sample and by wealth quintile (top vs. bottom) and by residence (rural vs. urban). The equity gap was calculated by taking the arithmetic difference in the prevalence of MDD and consumption of specific food groups between estimates for the upper and lowest wealth quintiles, and between rural and urban areas.
For aim 2, bivariate and multivariable logistic regression was conducted to estimate odds ratios and 95% confidence intervals (CI).
Each of the candidate variables was included in bivariate logistic regression models as risk factors for not achieving MDD. Those that met a threshold for statistical significance of p < 0.05 were selected for inclusion in the final multivariable models for each country.  Improved drinking water sources piped into dwelling piped to yard/plot, public tap/standpipe, piped to the neighbour, tube well or borehole, protected well, protected spring, rainwater.
c Efficient cooking fuel includes electricity, gas, and LPG while inefficient ones include charcoal, animal dung, wood, and others.
In some country models, decisions had to be made about retaining conceptually correlated variables. These included (1) for delivery place and receipt of assisted delivery by health professionals and (2) specific exposure to TV, newspaper or radio and to any of these media in the past week. For these sets of variables that were all significant in univariate analysis, the selection of which to include in the multivariable regression was based on at first, whether or not they remained significant in the multivariable model and then, which had the smaller p value when all sets of variables were significant in the multivariable model.
Only covariates with less than 10% missing value were included in the multivariable regression analysis. Variables that were statistically significant (p < 0.05) but with more than 10% missing in univariate regression were (1) women's empowerment in Kenya, South Africa and Zimbabwe, and (2) maternal BMI in Rwanda. Multicollinearity for the final multivariable logistic regression was checked by calculating variance inflation factors (VIF) and defined using a threshold of 4.0 for VIF and −0.2 for tolerance (Hair et al., 2010).
All analyses adjusted for DHS sampling design and survey weights were applied to generate nationally representative point estimates. We defined 'common regional risk factors of not meeting MDD' across countries as those factors that were significant in the multivariable regression models of two or more countries.

| Ethics and institutional review
The DHS data sets are publicly available from the DHS programme website (dhsprogram.com/data/). This secondary data analysis using deidentified data was deemed exempt from ethics review by the Institutional Review Board (IRB) at Johns Hopkins Bloomberg School of Public Health, USA.
F I G U R E 1 Conceptual framework of determinants of feeding practices among children above 6 months of age (modified from Blaney et al., 2015).

| Country demographics and socioeconomic status
Countries in the region exhibited similar patterns in their socioeconomic characteristics (Table 1). They were predominantly rural, less than 1/3 of households were female-headed and more than half of households had at least two children under 5 years of age. Nearly all families used inefficient cooking methods while access to improved drinking water and sanitation was more variable. South Africa was the exception to the pattern with a high proportion of urban residents, a majority of female-headed households and high access to improved drinking water in their houses and use of efficient cooking fuel. The mean child age was 14.0-14.4 months (range: 6-23)

| Proportion of MDD and food groups
The proportion of children 6-23 months old attaining MDD was low across all nine countries. Only two countries exceeded 30%, South Africa (38.5%) and Kenya (35.4%) (

| Equity gaps in MDD and food group consumption
In all nine countries, a greater proportion of children met the threshold for MDD in wealthier households compared with poorer households (Figures 3 and 4, Table 2 and Supporting Information:   Table 4).

| Risk factor analysis
In bivariate models, the direction of the association between MDD and wealth quintiles (except for Rwanda), maternal age, media and In multivariable models, a number of common risk factors for not meeting MDD were identified across two or more countries ( Figure 6 and Supporting Information:

F I G U R E 3 Equiplots of eight food groups and minimum dietary diversity between richest and poorest household wealth quintiles in nine
Eastern and Southern African countries. All values were estimated accounted for survey design and sampling weights.
F I G U R E 4 Equiplots of eight food groups and minimum dietary diversity by urban/rural residence in nine Eastern and Southern African countries. All values were estimated accounted for survey design and sampling weights. Our findings were similar to those from a study from 80 low-andmiddle-income countries which observed wealth inequities in child consumption of certain food groups including dairy, flesh foods, eggs and to other fruits and vegetables but no differences for cereals, legumes and VAFV (Gatica-Domínguez et al., 2021).
Consumption of animal-source foods (ASF) was generally low in most households across the region except for South Africa and egg consumption was less frequent than other types of animal-source foods. There has been a recent focus on improving egg consumption F I G U R E 6 Multivariable models of the odds of not reaching minimum dietary diversity 1 in nine Eastern and Southern African countries. Odds ratios >1.0 indicate less likely to attain MDD compared with the reference group. Table only presents variables that were significant in ≥2 country-specific multivariable regression models; n/a indicates that the variable was not significant in bivariate analysis. All values were estimated accounted for survey design and sampling weights. The threshold used for each colour was decided arbitrary through discussion by the authors. MDD, minimum dietary diversity; N/S; not significant at bivariate regression; OR, odds ratio.
to improve child nutrition in low-and middle-income countries (C. K. Lutter et al., 2018). Data suggest that the supply of eggs is relatively scarce in Africa, where only 2.5 kg of eggs per person were available compared with the global average of 8.9 and 9.1 kg in Asia (FAO, 2010). Given the challenges of keeping chickens healthy inhome production systems, investments in commercial production may be needed to lower the price of eggs and increase accessibility (C. Lutter et al., 2020;C. K. Lutter, 2020).
Fish are an important but underutilized resource for addressing nutrient gaps, particularly in coastal countries in the region (FAO, 2020). Poultry is relatively more affordable in Southern Africa than in Eastern Africa which may explain the differences we observed in reported consumption of flesh foods (OECD/FAO., 2016). Ownership of cows or livestock is associated with higher consumption of dairy and meat foods among household members but may be less accessible to young children (Haileselassie et al., 2020;Hetherington et al., 2017). Given that different animal source foods can provide different under-consumed nutrients, the ESA region should prioritise identifying strategies to increase consumption of multiple ASFs as fit for the local context rather than focussing on a single type of ASF.
Additionally, given that the links between farm production and dietary diversity are positive but typically small (Ruel et al., 2018;Sibhatu & Qaim, 2018), market-based interventions possibly including conditional cash transfers may be needed to enhance access to diverse diets (Manley et al., 2022).
Interesting patterns were also apparent for provitamin-A-rich fruit and vegetable consumption. Malawi and Rwanda both had relatively high and equitable consumption relative to other countries.
Kenya and Tanzania also had over 60% consumption, but consumption patterns were not as equitable by wealth or urban/rural residence. It would be useful to better understand the household production, accessibility and affordability of these fruits and vegetables across contexts and to identify whether there are specific strategies in Malawi and Rwanda that can be replicated. Unlike other countries, in Zimbabwe and Uganda, children from poorer households and rural households consumed more VAFV than wealthier or urban households. Uganda ranks as the second largest producer of fruits and vegetables in sub-Saharan Africa, with horticulture in all districts (Dijkxhoorn et al., 2019). It may be that a more robust fruit farming sector could help explain the different trends in these countries (Dijkxhoorn et al., 2019;Zimbabwe National Statistics Agency, 2019). A study examining the cost of purchasing nutritious complementary foods in six ESA countries reported that vitamin A is an affordable nutrient through purchases of dark leafy greens, orange-fleshed vegetables and liver (Ryckman et al., 2021).
Breast milk was consistently more commonly consumed by children from both poorer and rural households. Conversely, the timely introduction of complementary feeding seemed to be higher in urban and wealthier households (Supporting Information: Table 6). This trend indicates breastfeeding or delay in complementary foods are driven by accessibility and affordability of foods for the child. In Kenya, continued breastfeeding, timely introduction of complementary foods and dietary diversity are more balanced than other countries (Supporting Information: Table 6 and Figure 2). This is likely driven in part by employment patterns: other studies have shown that having an occupation in the public or private sector is associated with early cessation of exclusive breastfeeding and earlier introduction of complementary foods (Coulibaly et al., 2014;Habtewold et al., 2019). At a regional level, efforts to support countries by strengthening maternity protection laws or even creating family-friendly workspaces that enable women to continue to breastfeed their children can be a way of supporting breastfeeding for working women in urban areas (UNICEF, 2019).
The Young children are supposed to be excluded from this requirement, but their diets may be subject to wider household practices. Religious leaders have been considered a key target of IYCF programming for this reason (Sanghvi et al., 2013). Ethiopia is also the farthest north of the ESAR countries and, similar to Kenya, includes large pastoralist regions. Livestock ownership is high but studies suggest that they are not slaughtered for household consumption (Bundala et al., 2020); however, they may be a good food source to provide milk or eggs to children. Promoting the consumption of ASF in Ethiopia would be a reasonable choice to provide animal protein and micronutrients such as iron.
Rwanda ranks as having among the highest consumption of beans in the world (~29 kg per capital per year), and about 20% of the beans produced are iron-rich varieties (Larochelle & Alwang, 2014).
Given the low consumption of this food group across other countries, learning from Rwanda about how to enhance the production and consumption of beans and legumes is an important opportunity for other countries in the region.
Consistent with global findings, younger children (6-11 months) had a less diverse diet compared with older ages (12-24 months). During this early period, children often lack teeth to chew and there may be concerns about feeding young children foods that require chewing (Mura Paroche et al., 2017). Younger children would be at greater risk of MDD failure if the timely initiation of complementary feeding is delayed (Supporting Information: Table 6). As premastication has become less common, foods such as meat or fish require extra preparation, which can be taxing in contexts where mothers lack labour-saving technologies (Birch et al., 2007;Pelto et al., 2010 (Cetthakrikul et al., 2022). Thus, it is important that strategies to use media to promote proper infant and young child feeding practices encompass the examination of a broad set of opportunities that are aligned with WHO guidelines.
Low incomes and heavy workload for women are linked with poor child feeding across the region (Ahishakiye et al., 2019;Chakona, 2020;Solomon et al., 2017). Findings from this study highlight that an investment in maternal education, keeping girls in school and ensuring completion of secondary education should be prioritised in the region as part of the collaboration with the education sector. In three countries (Tanzania, Malawi, Uganda), we found that children of the youngest maternal age group (15-24 years old) had significantly higher odds of not meeting MDD than children of older women. ESA has one of the highest rates of adolescent pregnancy and young motherhood in the world (Africa, 2020) and young mothers are less likely to use timely antenatal and postnatal care and have limited access to nutrition information (Mekonnen et al., 2019). Approaches that reach adolescent mothers using existing community systems and structures and social media channels could be an entry point for ensuring these vulnerable girls are reached with IYCF counselling and support. Additionally, given that elders or grandparents may play a particularly significant role in the lives of adolescent mothers and their infants, it may be worth developing approaches to reach them with counselling and support messages as well (Bray & Dawes, 2016 Key strengths of this study included the use of nationally representative data spanning nine countries in the ESA region and the use of the revised 8-item MDD score. However, the generalisability of the findings to the remaining countries in the region is uncertain. Due to the limited information in the DHS related to food systems (including access to food), we were unable to examine a number of risk factors for MDD that we would have anticipated being possible drivers of low MDD, including proximity to markets, market prices and availability of foods in the market. The crosssectional nature of the analysis also limits the ability to draw causal inferences between risk factors and MDD. Since MDD is compared across countries at different timepoints, we may not exclude potential biases related to seasonal or yearly differences in food availability. For example, an assessment made at harvest seasons results in narrower wealth gaps in some crops. While we strived to use the most available DHS data for the countries at the time of the analysis, in three countries, a more recent data set is now available, and the generalisability of our findings to that survey is uncertain.

| CONCLUSIONS
Meaningful differences in the rate of MDD and wide equity gaps were apparent for most food groups between high and low-wealth quintiles and rural and urban populations in ESA. We identified many common socioeconomic factors associated with child dietary diversity across countries. Such common factors across countries provide grounding for regional actions to address infant and young child feeding. Interventions to address when complementary foods are introduced, support adolescent mothers or mothers with poor KANG ET AL.
| 13 of 16 education, and improve food security may be prioritised for child nutrition programmes.

AUTHOR CONTRIBUTIONS
Yunhee Kang was involved in conceptualization, methodology, formal analysis, writing of the manuscript-original draft, and investigation.
Rebecca A. Heidkamp was involved in conceptualisation, writing of the manuscript-review and editing, and investigation. Kudakwashe Mako-Mushaninga was involved in conceptualization and writing of the manuscript-review and editing. Aashima Garg was involved in conceptualization and writing-review and editing. Joan N. Matji was involved in conceptualization and writing-review and editing. Mara Nyawo was involved in writing of the manuscript-review and editing. Hope C. Craig was involved in writing of the manuscript-review and editing. Andrew L.
Thorne-Lyman was involved in conceptualisation, methodology, writing of the manuscript-original draft, funding acquisition, and project administration.