Stunting and Anemia in Children from Urban Poor Environments in 28 Low and Middle-income Countries: A Meta-analysis of Demographic and Health Survey Data

Child malnutrition remains a global concern with implications not only for children’s health and cognitive function, but also for countries’ economic growth. Recent reports suggest that global nutrition targets will not be met by 2025. Large gaps are evident between and within countries. One of the largest disparities in child malnutrition within counties is between urban and rural children. Large disparities also exist in urban areas that have higher rates of child malnutrition in the urban poor areas or slums. This paper examines stunting and anemia related to an urban poverty measure in children under age 5 in 28 low and middle-income countries with Demographic and Health Survey data. We used the United Nations Human Settlements Programme (UN-HABITAT) definition to define urban poor areas as a proxy for slums. The results show that in several countries, children had a higher risk of stunting and anemia in urban poor areas compared to children in urban non-poor areas. In some countries, this risk was similar to the risk between the rural and urban non-poor. Tests of heterogeneity showed that these results were not homogeneous across countries. These results help to identify areas of greater disadvantage and the required interventions for stunting and anemia.


Introduction
Child stunting and malnutrition remain a global concern. Globally, among children under age five, at least one in three are not growing as they should because of malnutrition that includes stunting. At least one in two children under age five also suffer from deficiencies in vitamins and essential nutrients [1]. Despite declines in stunting among children globally, progress has been too slow to achieve the 2025 global nutrition target of reducing stunting by 40% to a target of 100 million; it is projected that 130 million children will be stunted by 2025 [2]. There are an estimated 149 million (or almost a quarter) of children under age five who are stunted [2]. There have also been global reductions in anemia prevalence over the last several decades; although, the total cases of anemia increased from 1.42 billion in 1990 to 1.74 billion in 2019 [3]. The greatest burden of anemia is found in western and Central Sub-Saharan Africa and South Asia [3,4], with the highest combined prevalence of anemia (39.7%) found in children under age five [3]. Malnutrition in children not only has a direct effect on their growth, but also on their cognitive, motor, and language development, mental health, susceptibility to diseases, and other long-term development such as lower levels of learning capacity that can continue into adulthood [2,[5][6][7][8]. Malnutrition also influences the economic development and growth of countries and contributes to the intergenerational transmission of poverty [2].

Data
Data from countries that had a Demographic and Health Survey (DHS) from 2014 to 2018 were included in the analysis. Countries with a population less than 10 million were excluded, which left 30 surveys that were considered for the analysis. These surveys use a randomized, multi-stage sample design that produces estimates that are representative of the country at the time of the survey. The descriptive analysis revealed that Jordan and Zimbabwe did not have any children in urban poor clusters according to the definition used in this paper. Therefore, these two countries appear in the results for the description of the urban poor cluster variable in Figure 1 only. Table 1 shows the countries in the analysis and their sample sizes.

Variables
The outcomes of interest are stunting, and moderate to severe anemia in children. Stunting is defined as the proportion of de facto children under age five who have a height-for-age z-score below the median of the WHO 2007 reference population by more than two standard deviations. Moderate to severe anemia in children is defined as the proportion of de facto children age 6-59 months with a hemoglobin level less than 10 g per deciliter. Hemoglobin levels are adjusted for altitude.
The main variable of interest is the urban poverty cluster variable. To construct this variable, we used the UN-HABITAT definition of a slum household, which is a household that lacks one or more of the following: durable housing of permanent nature, sufficient living space for not more than three persons per room, access to safe water, access to adequate sanitation, and security of tenure that prevents forced evictions [31,32]. The DHS data includes information on all items except the security of tenure. Therefore, we defined an urban poor household as households in an urban cluster that are lacking two or more of the following: durable material for the floor, wall, and roof; access to improved water; access to improved sanitation; and fewer than three persons per sleeping room. To define the variable at the cluster level, an urban poor cluster was classified as a cluster with more than half of the households classified as urban poor households, as recommended by United Nations Human Settlements Programme [9]. Clusters that contained fewer than half urban poor households were defined as urban non-poor areas. This aggregation to the cluster level was used to define areas that Nutrients 2020, 12, 3539 4 of 13 may be considered slum-like (the urban poor). Therefore, the urban poverty cluster variable has three categories: rural, urban poor, and urban non-poor.
The control variables used in the regressions include the child's sex, age in months categorized in 6-month intervals, the mother's education level, and the region at the administrative level 1 for the country. The wealth index could not be included as a control because it was highly correlated with the urban poverty cluster variable.

Methods
The analysis includes descriptive and logistic regression analysis. A crosstabulation of the urban poverty cluster variable and each outcome was performed along with Chi-square tests of association. In addition, adjusted logistic regressions were fit for each outcome and the variable of interest along with the controls. The results were summarized with forest plots for the main variable of interest that also produce a pooled odds ratio using meta-analysis and the "metan" command in Stata [33]. The metan command also performs a heterogeneity test that examines the variability between the surveys. This test produces an I-squared index and p-value. An I-squared value of 25-50% is considered a low level of heterogeneity, 50-75% is moderate, and greater than 75% is high [34]. A significant p-value would indicate that we reject the null hypothesis that the surveys are homogeneous for the result and that the surveys are heterogeneous. For the meta-analysis, the surveys were assigned equal weight so that large surveys such as India did not predominate the results.
All analyses accounted for the sampling design and sampling weights and were performed with Stata 16 software. The estimates are reported with 95% confidence intervals (CI). Figure 1 shows the percentage distribution of the population of urban non-poor, urban poor, and rural clusters for the countries in the analysis. Several countries had a highly rural population with more than three-quarters of the population in Burundi, Cambodia, Chad, Ethiopia, Malawi, Mali, Rwanda, and Uganda living in rural clusters. The highest proportion of people who live in urban poor clusters were found in Kenya and Benin (both 17%), followed by the Democratic Republic of the Congo (DRC) and Chad (both 16%). Jordan was highly urban with no urban poor clusters. In contrast, Zimbabwe was highly rural with no urban poor clusters. Therefore, the remaining analyses do not include Jordan and Zimbabwe. Nineteen of the 30 countries shown in Figure 1 have fewer than 5% of the population living in urban poor clusters. Table 2 shows the percentage of children under age five who are stunted. More than one-third of children under age five were stunted in 15 countries during the time of the survey: Angola, Bangladesh, Burundi, Chad, DRC, Ethiopia, Guatemala, India, Malawi, Nepal, Nigeria, Pakistan, Rwanda, Tanzania, and Zambia. Generally, the lowest percentage of stunting was found in urban non-poor clusters and the highest in rural clusters. However, in Bangladesh, Guatemala, Haiti, India, Nepal, Pakistan, Senegal, Tanzania, and Uganda, stunting was highest in urban poor clusters with the highest found in Pakistan (56%), Guatemala (55%), Bangladesh (48%), and Burundi (45%). Differences in stunting across the urban poverty cluster distribution was significant except for Egypt and South Africa.

Results
Since four surveys in Table 3 did not perform anemia testing in children, the analysis with the anemia outcome was performed on 24 surveys. The highest level of moderate to severe anemia in children age 6-59 months was found in Mali (57%), followed by Guinea and Benin, both at 43%. Over one-third of children from 12 of the 24 countries had moderate to severe anemia. In general, anemia was lowest in children who live in urban non-poor clusters and higher in the rural areas. However, in several surveys, there was either no significant difference in anemia between the urban non-poor and rural clusters or anemia was the highest in the urban poor clusters. More than 40% of children age 6-59 months who live in urban poor clusters in Benin, Ghana, Haiti, Mali, Nigeria, and Senegal had moderate to severe anemia. In Angola, Myanmar, South Africa, Uganda, and Zambia, there were no significant differences in moderate to severe anemia across the urban poverty clusters.      urban non-poor reference category. In Figure 2, there are more significant findings for rural compared to urban non-poor than urban poor compared to urban non-poor. With Egypt as an exception, children in rural clusters are more likely to be stunted than children in the urban non-poor clusters. The pooled odds ratio from the meta-analysis was 1.5 (95% CI 1.5, 1.6). Overall, living in a rural area increases the risk of stunting in children by 50% compared to living in an urban non-poor cluster. However, the I-squared results indicate that the surveys were heterogeneous. Several countries (16 from the 28) exhibited significantly higher odds of stunting in children who live in urban poor clusters compared to urban non-poor with the adjusted odds ratios ranging from 1.3 to 2.0, except for Egypt and Ghana. In Ethiopia, Guatemala, Pakistan, and Tanzania, children living in urban poor clusters had almost twice the odds of being stunted compared to children in the urban non-poor clusters. For India, Mali, and Nigeria, the odds ratios for both the rural and urban poor cluster categories were very similar, which indicated that children living in these areas had a similar risk of stunting in comparison to children living in urban non-poor clusters. The pooled odds ratio from the meta-analysis was 1.3 (95% CI 1.2, 1.5), which indicated that children living in urban poor clusters have a 30% increased odds of stunting compared to children living in urban non-poor clusters. The I-squared results also indicate that the surveys were heterogeneous. In Egypt and Ghana, children in urban poor clusters had almost 50% lower odds of being stunting compared to children in the urban non-poor clusters. In Egypt, children in rural clusters also had a 50% lower odds of being stunting compared to children in urban non-poor clusters.
Nutrients 2020, 12, x FOR PEER REVIEW 8 of 14 Figures 2 and 3 summarize the regression results for the stunting and anemia outcomes and the urban poverty cluster variable. These estimates can also be found in Appendixes A and B. The figures show the adjusted odds ratios for the rural and urban poor categories compared to the urban nonpoor reference category. In Figure 2, there are more significant findings for rural compared to urban non-poor than urban poor compared to urban non-poor. With Egypt as an exception, children in rural clusters are more likely to be stunted than children in the urban non-poor clusters. The pooled odds ratio from the meta-analysis was 1.5 (95% CI 1.5, 1.6). Overall, living in a rural area increases the risk of stunting in children by 50% compared to living in an urban non-poor cluster. However, the Isquared results indicate that the surveys were heterogeneous. Several countries (16 from the 28) exhibited significantly higher odds of stunting in children who live in urban poor clusters compared to urban non-poor with the adjusted odds ratios ranging from 1.3 to 2.0, except for Egypt and Ghana. In Ethiopia, Guatemala, Pakistan, and Tanzania, children living in urban poor clusters had almost twice the odds of being stunted compared to children in the urban non-poor clusters. For India, Mali, and Nigeria, the odds ratios for both the rural and urban poor cluster categories were very similar, which indicated that children living in these areas had a similar risk of stunting in comparison to children living in urban non-poor clusters. The pooled odds ratio from the meta-analysis was 1.3 (95% CI 1.2, 1.5), which indicated that children living in urban poor clusters have a 30% increased odds of stunting compared to children living in urban non-poor clusters. The I-squared results also indicate that the surveys were heterogeneous. In Egypt and Ghana, children in urban poor clusters had almost 50% lower odds of being stunting compared to children in the urban non-poor clusters. In Egypt, children in rural clusters also had a 50% lower odds of being stunting compared to children in urban non-poor clusters. As shown in Figure 3, in 13 of 24 surveys, children living in rural clusters had a higher odds of having moderate to severe anemia compared to children living in urban non-poor clusters. In 12 of 24 surveys, children living in urban poor clusters had greater odds of having moderate to severe anemia compared to children living in urban non-poor clusters. For several countries, the odds of children living in urban poor clusters having anemia compared to children in urban non-poor clusters were almost twice as high. For Burundi, Nigeria, and Senegal, the odds ratios for the rural and urban poor categories were very similar, which indicated that children living in these clusters had a similar As shown in Figure 3, in 13 of 24 surveys, children living in rural clusters had a higher odds of having moderate to severe anemia compared to children living in urban non-poor clusters. In 12 of 24 surveys, children living in urban poor clusters had greater odds of having moderate to severe anemia compared to children living in urban non-poor clusters. For several countries, the odds of children living in urban poor clusters having anemia compared to children in urban non-poor clusters were almost twice as high. For Burundi, Nigeria, and Senegal, the odds ratios for the rural and urban poor categories were very similar, which indicated that children living in these clusters had a similar risk of anemia when compared to children living in urban non-poor clusters. The pooled odds ratio from the meta-analysis for children living in rural and urban poor clusters compared to urban non-poor clusters were both 1.4. This means that children in rural and urban poor clusters had 40% greater odds of being anemic compared to children who live in the urban non-poor clusters. However, the I-squared index and p-value indicate that this result has moderate to high level of heterogeneity. One unexpected finding was the 60% lower odds of moderate to severe anemia in children living in urban poor clusters compared to the urban non-poor children in Guinea.
Nutrients 2020, 12, x FOR PEER REVIEW 9 of 14 risk of anemia when compared to children living in urban non-poor clusters. The pooled odds ratio from the meta-analysis for children living in rural and urban poor clusters compared to urban nonpoor clusters were both 1.4. This means that children in rural and urban poor clusters had 40% greater odds of being anemic compared to children who live in the urban non-poor clusters. However, the Isquared index and p-value indicate that this result has moderate to high level of heterogeneity. One unexpected finding was the 60% lower odds of moderate to severe anemia in children living in urban poor clusters compared to the urban non-poor children in Guinea.

Discussion
This paper examined the prevalence of stunting and moderate to severe levels of anemia among children under age five in 28 countries. Globally, the world's population continues to grow, and urbanization continues to increase with 84% of the world's population living in less developed regions [11]. It is important to examine disparities within countries that account for the nuanced needs of people based on where they live. This study examined stunting and anemia through an urbanicity measure [21,34], referred to here as the urban poverty cluster variable, which goes beyond the traditional urban-rural residence dichotomy used in other studies. With this multi-country DHS meta-analysis that used the urban poverty cluster measure, we were able to focus on areas of greater disadvantage based on these two malnutrition conditions in children that have long-term developmental consequences for the child, as well as consequences for the society as a whole [2,[5][6][7][8]22].
In more than half the countries in our analysis, children in urban poor clusters had a greater odds of being stunted than those in urban non-poor clusters. Similarly, in half of the countries in our analysis with available data on anemia, children in urban poor clusters had a higher odds of having moderate to severe anemia than children in urban non-poor clusters. In an 11-country DHS analysis focused on intra-urban differentials, researchers found that child stunting was as much a problem in urban areas as in rural areas when socioeconomic status (SES) was considered within these areas [35]. In this same study, children in Guatemala were seven times more likely to be stunted in urban low SES areas, compared to the urban high SES areas. Similar relationships were found in Pakistan and Tanzania, but at lower magnitudes. These three countries were included in our study, and displayed greater odds of stunting among children in urban poor clusters compared to urban non-poor clusters. Previous work that used data from 73 countries with DHS data found that children in slums have

Discussion
This paper examined the prevalence of stunting and moderate to severe levels of anemia among children under age five in 28 countries. Globally, the world's population continues to grow, and urbanization continues to increase with 84% of the world's population living in less developed regions [11]. It is important to examine disparities within countries that account for the nuanced needs of people based on where they live. This study examined stunting and anemia through an urbanicity measure [21,35], referred to here as the urban poverty cluster variable, which goes beyond the traditional urban-rural residence dichotomy used in other studies. With this multi-country DHS meta-analysis that used the urban poverty cluster measure, we were able to focus on areas of greater disadvantage based on these two malnutrition conditions in children that have long-term developmental consequences for the child, as well as consequences for the society as a whole [2,[5][6][7][8]22].
In more than half the countries in our analysis, children in urban poor clusters had a greater odds of being stunted than those in urban non-poor clusters. Similarly, in half of the countries in our analysis with available data on anemia, children in urban poor clusters had a higher odds of having moderate to severe anemia than children in urban non-poor clusters. In an 11-country DHS analysis focused on intra-urban differentials, researchers found that child stunting was as much a problem in urban areas as in rural areas when socioeconomic status (SES) was considered within these areas [36]. In this same study, children in Guatemala were seven times more likely to be stunted in urban low SES areas, compared to the urban high SES areas. Similar relationships were found in Pakistan and Tanzania, but at lower magnitudes. These three countries were included in our study, and displayed greater odds of stunting among children in urban poor clusters compared to urban non-poor clusters. Previous work that used data from 73 countries with DHS data found that children in slums have worse health risks compared with children in other urban areas, but lower risks than their rural counterparts [20]. The authors concluded that a large share of these differences can be explained by differences in the mother's education, wealth, and access to health services. In another study that examined health outcomes in Bangladesh, Kenya, Egypt, and India, children living in urban-slums fared worse in terms of health outcomes, including anemia and stunting, compared to their urban non-slum and rural counterparts [37].
In addition to expected findings, there were also some noteworthy, unexpected findings. We found that children in urban poor clusters had lower odds of being stunted in two countries-Egypt and Ghana-compared to children in urban non-poor clusters. Rural children in Egypt also displayed lower odds of being stunted compared to children in urban non-poor clusters. This aligns with the finding in another study in Egypt where stunting is decreasing among the poorest wealth quintiles but increasing in the richest wealth quintiles [38]. In addition, food programs in Egypt have focused on energy-dense foods, especially among poor and rural populations, given that rising food prices and food insecurity have led to an increase in consumption of calorie-dense foods [38,39]. Egypt and Ghana are also experiencing the double burden of malnutrition [39,40], and in Ghana, concurrent overweight and stunting was found to be significantly higher in children from the fourth wealth quintile compared to children from the poorest wealth quintile [40]. More research is needed to understand the changes in nutrition transition and dietary diversity in these countries that may shed more light on these findings. Another area for further exploration is the high provision of services in these countries' urban poor areas, that may counter the negative effects of living in these slum-like conditions [21,37,41,42]. For example, one study found that living in a slum neighborhood was associated with poor child outcomes, except when mothers had access to and received health services (such as antenatal care) from trained providers [21]. Further research is required to explore these country-specific findings. The relationship between urbanization and child malnutrition is complex and may be modified by some advantages and disadvantages to living in an urban environment [14,29,43].
Our study has some limitations. The urban-rural variable was used when constructing the urban poverty cluster variable, our main variable of interest. Since the urban-rural dichotomous variable available in the DHS data uses each country's specific definition of what is urban or rural, this definition is country-specific and not standardized [44]. Incorporating this DHS urban-rural residence variable has the potential to introduce misclassification and error into the urban poverty cluster variable used in our analysis. Another potential limitation of our study is that the UN-HABITAT's definition of a slum household [31], which we used in the construction of the urban poverty cluster variable, may be more pertinent to lower-income countries than to the middle-income countries. There is potential for missed opportunities in capturing within-country inequalities because some domains in the UN-HABITAT's definition may not be as relevant. Some countries' circumstances require other domains to be incorporated for better comparability in the definition of a slum household. Finally, the displacement procedures employed by the DHS mask the real locations of urban clusters by a displacement of up to 2 km. Therefore, it is not possible to know if the urban poor cluster we identify is actually a slum area or another disadvantaged area. Thus, we exercised caution when identifying the urban poor clusters in our analysis as actual slums.
Based on this study, we recommend that researchers, policymakers, and program implementers go beyond the usual urban-rural measurement to examine poor child health outcomes, as recommended by previous researchers [20,36,45]. We suggest considering the inclusion and use of an urban poverty cluster variable based on available data per domain to provide a better understanding of urban and rural environments. Country-specific programming and policies focused on the rural and urban poor clusters are essential. Given the similarities in the odds of stunting and moderate to severe levels of anemia across these two cluster categories in some countries, programming might consider accounting for these findings in complementary ways.

Conclusions
This study further informs the relationship between more nuanced areas of urban and rural environments and children's risk of being stunted and experiencing moderate and severe levels of anemia. In several countries, we found that children living in urban poor areas experienced worse child malnutrition outcomes compared to those in the urban non-poor areas. In some countries in our analysis, this disparity was similar to the disparity between the rural and urban non-poor. Future research with the urban poor cluster variable should consider contextual factors in urban poor clusters and slum areas, such as the availability and use of health services and accessibility to nutrient-rich foods. Intervention programs should consider urban areas that may be further disadvantaged if they are identified as slums or areas with poor access to water, sanitation, and other issues such as overcrowding, which can have direct consequences on children's health outcomes including malnutrition.