Child health inequities in developing countries: Differences across urban and rural areas

Objectives: To document and compare the magnitude of inequities in child malnutrition across urban and rural areas, and to investigate the extent to which within-urban disparities in child malnutrition are accounted for by the characteristics of communities, households and individuals. Methods: The most recent data sets available from the Demographic and Health Surveys (DHS) of 15 countries in sub-Saharan Africa (SSA) are used. The selection criteria were set to ensure that the number of countries, their geographical spread across Western/Central and Eastern/Southern Africa, and their socioeconomic diversities, constitute a good yardstick for the region and allow us to draw some generalizations. A household wealth index is constructed in each country and area (urban, rural), and the odds ratio between its uppermost and lowermost category, derived from multilevel logistic models, is used as a measure of socioeconomic inequalities. Control variables include mother’s and father’s education, community socioeconomic status (SES) designed to represent the broad socio-economic ecology of the neighborhoods in which families live, and relevant mother- and child-level covariates. Results: Across countries in SSA, though socioeconomic inequalities in stunting do exist in both urban and rural areas, they are significantly larger in urban areas. Intra-urban differences in child malnutrition are larger than overall urban-rural differentials in child malnutrition, and there seem to be no visible relationships between within-urban inequities in child health on the one hand, and urban population growth, urban malnutrition, or overall rural-urban differentials in malnutrition, on the other. Finally, maternal and father’s education, community SES and other measurable covariates at the mother and child levels only explain a slight part of the within-urban differences in child malnutrition. Conclusion: The urban advantage in health masks enormous disparities between the poor and the non-poor in urban areas of SSA. Specific policies geared at preferentially improving the health and nutrition of the urban poor should be implemented, so that while targeting the best attainable average level of health, reducing gaps between population groups is also on target. To successfully monitor the gaps between urban poor and non-poor, existing data collection programs such as the DHS and other nationally representative surveys should be re-designed to capture the changing patterns of the spatial distribution of population. © 2006 Fotso; licensee BioMed Central Ltd.

This research uses the most recent data sets available as of January 2005 from the Demographic and Health Surveys (DHS) of the following 15 countries: Burkina Faso, Cameroon, Chad, Côte d’Ivoire, Ghana, Nigeria, and Togo from Western and Central Africa, and Kenya, Madagascar, Malawi, Mozambique, Tanzania, Uganda, Zambia and Zimbabwe from Eastern and Southern Africa. The selection criteria were not only based on the availability of data on child nutritional status, but more importantly, were set to ensure that the number of selected countries, their geographical spread across Western/Central and Eastern/Southern Africa, and their socioeconomic diversities, could allow us to draw some generalizations. Indeed, Column (Col.) 1 of Table ​Table11 shows that according to the human development index (HDI2), four countries (Ghana, Zimbabwe, Cameroon and Kenya) can be classified as high-HDI (ranking below 20 out of 48 African countries); six others (Madagascar, Togo, Nigeria, Zambia, Côte d’Ivoire and Tanzania) are middle-HDI (ranking between 20 and 30); and the five remaining (Burkina Faso, Mozambique, Chad, Malawi and Uganda) can be classified as low-HDI (ranking 31 and higher). Further, in each of the above categories of ranking, there is almost the same number of countries from either region (Central/Western and Eastern/Southern Africa). Human development index, urban population and gross domestic product in 15 selected countries aRanking within 48 African countries. Countries are ranked in decreasing order of human development index. Source: United Nations Development Program, 2000. bSource: United Nations, 2004. cAt constant 1995 US$. Available data for Uganda and Tanzania start in 1982 and 1988 respectively. Source: World Bank, 2004. dNAp: Not applicable;eNAv: Not available. Table ​Table11 also illustrates the economic diversity of the selected countries with regard to levels of urbanization and per capita gross domestic product (GDP) in 2000. It shows that the percentage of urban population (Col. 2) differs significantly among the selected countries. It varies from 12–17% in Uganda, Malawi and Burkina Faso, to close to or more than 45% in Cameroon, Nigeria, Ghana and Côte d’Ivoire. The average value for SSA is 34%. As for GDP per capita, Côte d’Ivoire, Cameroon and Zimbabwe emerge as the most affluent countries with values higher than $600, whilst by contrast Malawi, Mozambique, Tanzania, Chad and Madagascar are the most deprived (less than $250). The selected countries also display marked socioeconomic diversities in terms of per capita food production, per capita health expenditures, and adult literacy rates (not shown). Overall, we make no pretence that the sample countries are representative of the entire SSA, but their number and geographical and socioeconomic diversities constitute a good yardstick for the region and help to strengthen the findings from the study. Moreover, the selected countries typify rapid urbanization amidst declining economies. Table ​Table11 shows that between 1980 and 2000, the urban population grew by 5.4% per year in the selected countries as a whole, against an average of 3.5% for developing countries. The fastest growths are recorded in Kenya (7.4%), Tanzania (7.2%) and Mozambique (6.6%). By contrast, Zambia (2.2%), Chad (4.0%) and Côte d’Ivoire (4.4%) witnessed the slowest growth rates of their urban populations. At the same time, GDP per capita dropped by 0.7% on average in the selected countries. The most marked reductions are in Togo, Zambia, Cote d’Ivoire and Madagascar (1.7–1.9%), whereas improvements are recorded in Uganda (+2.1%) and Burkina Faso (1.2%), and to a lesser degree in Mozambique (0.9%) and Chad (0.7%). Among various growth-monitoring indices, the three most commonly used profiles of malnutrition in children are stunting, wasting and underweight, measured by height-for-age, weight-for height, and weight-for-age indexes, respectively. The present study focuses on stunting (or growth retardation) in young children. Stunting results from recurrent episodes or prolonged periods of nutrition deficiency for calories and/or protein available to the body tissues, inadequate intake of food over a long period of time, or persistent or recurrent ill-health [15,18]. Since the height-for-age measure is less sensitive to temporary food shortages, stunting is considered the most reliable indicator of a child’s nutritional status, especially for the purpose of differentiating socioeconomic conditions within and between countries [20,21]. As recommended by the WHO, children whose indices fall more than two standard deviations below the median of the NCHS/CDC/WHO reference population are classified as stunted [17]. Despite the growing number of studies attesting evidence of poorer health among people with less education and income, lower status jobs, and poorer housing [12,21-25], there is still debate about the meaning of health inequalities [26-28]. Kawachi et al. arguably state that priority must be given to analysing health inequalities between groups, referred to as health inequities [29]. There is also a great deal of discussion on the appropriate measures to capture such inequities [30,31]. The concentration index is increasingly used in the literature on socioeconomic inequalities in health [12,21,22,25]. The concentration curve plots the cumulative proportions of the population (beginning with the most disadvantaged) against the cumulative proportion of the health outcome under study. The resulting concentration index which varies from -1 to +1 measures the extent to which a health outcome is unequally distributed across groups [25]. Though this measure takes into account what is going on in all the groups, it is mainly used for descriptive purposes, and adjustment for control variables is not straightforward. The odds ratio between the uppermost and the lowermost categories of the socioeconomic variable is used in this paper as a proxy for socioeconomic inequalities. The main advantage of this approach is the use of a single number which makes it easier to compare the magnitude of inequalities across populations or over time, even though it overlooks the health outcome in the intermediate groups of the socioeconomic variable. This measure is particularly appropriate when a linear trend has previously been observed in the association between the socioeconomic variable and the health outcome under consideration [30]. Poverty -and thus SES- has been recognized to be multi-faceted, and to exert its influences on health at various levels (individual, household, community and nation). Poverty includes, but is not limited to, inadequate income, shelter and assets for individuals and households, and inadequate provision of infrastructure and basic services such as health services, roads, schools and vocational training [19,32]. This paper privileges the economic and material dimension of poverty at the household level. DHS data do not provide information on income or expenditures. Thus, along the lines of Gwatkin et al. and Filmer and Pritchett [33,34], we build on our previous work [35] and construct a household wealth index in each country and area (urban, rural). The wealth index is constructed from household’s possessions, source of drinking water, type of toilet facilities and flooring material using principal components analysis. It is then re-coded as poorest (bottom 30%), middle (next 40%), and richest (top 30%), with poorest as the reference category. The key control variables used in the study include urban-rural place of residence, and maternal education, known to have some effects on child health and nutrition that are independent of the effects of other measures of SES [23,36]. Maternal education is coded as no education (reference category), primary, secondary or higher. The controls also include a community SES constructed in each country and area, from the proportion of households having access to clean water and electricity, as well as the proportion of wage earners and that of educated adults (level of primary education or higher). The variable, which is in line with the multilevel nature of the health determinants [16,37-39], is designed to represent the broad socio-economic ecology of the neighborhoods in which families live, besides the broad rural-urban location of residence. Father’s education is also used in this study. In some societies of the developing world, certain behaviors and practices which may affect child health and nutrition are highly dependent on characteristics of the father, particularly his level of education [22]. The other control variables used in this study include: (i) at the mother level: age at birth of the index child, marital status, religion, and nutritional status; and (ii) at the child level: current age, sex, low birth weight, antenatal care, place of delivery, age-specific immunization status, birth order and interval, and breast feeding duration. DHS data have a hierarchical structure, with children nested within mothers, mothers clustered within households, and households nested within communities. As a result, observations from the same group are expected to be more alike at least in part because they share a common set of characteristics or have been exposed to a common set of conditions, thus violating the standard assumption of independence of observations inherent in conventional regression models. Consequently, unless some allowance for clustering is made, standard statistical methods for analyzing such data are no longer valid, as they generally produce downwardly biased variance estimates, leading for example to infer the existence of an effect when, in fact, that effect estimated from the sample could be ascribed to chance [40,41]. Multilevel models provide a framework for analysis which is not only technically stronger, but which also has a much greater capacity for generality than traditional single-level statistical methods [42]. Given that the number of children per household in the data for this analysis is very small (between 1.1 and 1.3), we carry out two-level (child and community) logistic regression analyses in each country and area. Models are fitted using the MLwiN software with Binomial, Predictive Quasi Likelihood (PQL) and second-order linearization procedures [41].

N/A

• Dima Health