Menu
Menu
Menu
Menu
Background. Malnutrition is a major public health and development concern in the developing world and in poor communities within these regions. Understanding the nature and determinants of socioeconomic inequality in malnutrition is essential in contemplating the health of populations in developing countries and in targeting resources appropriately to raise the health of the poor and most vulnerable groups. Methods. This paper uses a concentration index to summarize inequality in children’s height-for-age z-scores in Ghana across the entire socioeconomic distribution and decomposes this inequality into different contributing factors. Data is used from the Ghana 2003 Demographic and Health Survey. Results. The results show that malnutrition is related to poverty, maternal education, health care and family planning and regional characteristics. Socioeconomic inequality in malnutrition is mainly associated with poverty, health care use and regional disparities. Although average malnutrition is higher using the new growth standards recently released by the World Health Organization, socioeconomic inequality and the associated factors are robust to the change of reference population. Conclusion. Child malnutrition in Ghana is a multisectoral problem. The factors associated with average malnutrition rates are not necessarily the same as those associated with socioeconomic inequality in malnutrition. © 2007 Van de Poel et al; licensee BioMed Central Ltd.
Nutritional status was measured by height-for-age z-scores. An overview of other nutritional indices and why height-for-age is the most suited for this kind of analysis is provided in [36]. A height-for-age z-score is the difference between the height of a child and the median height of a child of the same age and sex in a well-nourished reference population divided by the standard deviation in the reference population. The new WHO child growth population is used as reference population [33]. To construct height-for-age z-scores based upon these standards, we used the software available on the WHO website [37]. To check sensitivity of the results to this change in reference group, the analysis is also done by using the US National Center for Health Statistics (NCHS) reference population [35]. Generally, children whose height-for-age z-score is below minus two standard deviations of the median of the reference population are considered chronically malnourished or stunted. In the regression models, the negative of the z-score is used as dependent variable (y). This facilitates interpretation since it has a positive mean and is increasing in malnutrition [32]. For the purpose of our analysis, using the z-score instead of a binary or ordinal variable indicating whether the child is (moderately/severely) stunted is preferred as it facilitates the interpretation of coefficients and the decomposition of socioeconomic inequality. However, binary indicators of stunting are also used in the descriptive analysis and to position Ghana within a set of other Sub-Saharan African countries. Assume yi is the negative of the height-for-age z-score of child i. The concentration index (C) of y results from a concentration curve, which plots the cumulative proportion of children, ranked by socioeconomic status, against the cumulative proportion of y. The concentration curve lies above the diagonal if y is larger among the poorer children and vice versa. The further the curve lies from the diagonal, the higher the socioeconomic inequality in nutritional status. A concentration index is a measure of this inequality and is defined as twice the area between the concentration curve and the diagonal. If children with low socioeconomic status suffer more malnutrition than their better off peers the concentration index will be negative [38]. It should be noted that the concentration index is not bounded within the range of [-1,1] if the health variable of interest takes negative, as well as positive values. Since children with a negative y are better off than children in the reference population, they cannot be considered malnourished. Therefore their z-score is changed into zero, such that the z-scores are restricted to positive values with zero indicating no malnutrition and higher z-scores indicating more severe malnutrition. Further, the bounds of the concentration index depend upon the mean of the indicator when applied to binary indicators, such as stunting [39]. This would impede cross-country comparisons due to substantial differences in means across countries. To avoid this problem, we used an alternative but related concentration index that was recently introduced by [40] and does not suffer from mean dependence, when comparing Ghana with other Sub-Saharan African countries. More formally, a concentration index of y can be written as [38]: where yi refers to the height-for-age of the i-th individual and Ri is its respective fractional rank in the socioeconomic distribution. As will be discussed further in the following section, the present paper uses a continuous wealth variable, developed by principal component analysis, as a measure of socioeconomic status [see e.g. [41]]. If yi is linearly modelled [32] showed that the concentration index of height-for-age can be decomposed into inequality in the determinants of height-for-age as follows: where μ is the mean of y, x¯k is the mean of xk, Ck is the concentration index of xk (with respect to socioeconomic status) and GCε is the generalized concentration index of the residuals. The latter term reflects the socioeconomic inequality in height-for-age that is left unexplained by the model and is calculated as GCε=2n∑i=1nεiRi. As the DHS data have a hierarchical structure, with children nested in households and households nested within communities, we have also considered using multilevel models to estimate the associations of variables with childhood malnutrition (see e.g. [42]). Allowing for random effects on the household and/or community level yielded coefficients that were similar to the ones from OLS regression corrected for clustering. Because of this similarity and because the use of multilevel models would complicate the decomposition of socioeconomic inequality in malnutrition, the remainder is based on results from linear regression corrected for clustering on the community level. All estimation takes account of sample weights (provided with the DHS data). Statistical inference on the decomposition results is obtained through bootstrapping with 3000 replications. The bootstrap procedure takes into account the dependence of observations within clusters. Data is used from the 2003 Ghana Demographic Health Survey (DHS) and are restricted to children under the age of 5. Anthropometric measures are missing for 12.3% of children in this age group. The final sample contains information on 3061 children. We did examine possible selection problems due to the high proportion of missing observations. A logit model explaining the selection in the sample and a Heckman sample selection model (using different exclusion restrictions) were used to check for this [43]. Both tests did not reveal large sample selection problems, and coefficients in the Heckman model were very similar to those in the model presented here. The nutritional status of a child is specified to be a linear function of child-level characteristics such as age, sex, duration of breastfeeding, size at birth; maternal characteristics such as education, mother’s age at birth, birth interval, marital status, use of health services, occupation and finally household-level characteristics such as wealth, type of toilet facility, access to safe water, number of under-five children in the household, region and urbanization. We preferred not to include information on the type of toilet and water source into the wealth indicator, as these variables can be expected to have a direct relation with children’s growth apart from being correlated with household socioeconomic status [44]. The explanatory variables are described in the last column of Table Table1.1. All have well documented relevance in the literature [5,22-26,31,32,45,46]. Mean, standard deviation and description of all variables Reference categories for categorical variables used in the regression model are in bold. No information on mother’s nutritional status was included in the set of explanatory variables. Since about 10% of women in the dataset were pregnant at the time of interview, their BMI did not provide an accurate measure of their nutritional status. Furthermore, BMI reflects current nutritional status and may not be relevant for children born 5 years prior to the interview. Inclusion of mother’s height-for-age had no significant effect on results.
N/A
