Menu
Menu
Menu
Menu
Background: Millions of people in low and low middle income countries suffer from extreme hunger and malnutrition. Research on the effect of food insecurity on child nutrition is concentrated in high income settings and has produced mixed results. Moreover, the existing evidence on food security and nutrition in children in low and middle income countries is either cross-sectional and/or is based primarily on rural populations. In this paper, we examine the effect of household food security status and its interaction with household wealth status on stunting among children aged between 6 and 23 months in resource-poor urban setting in Kenya. Methods: We use longitudinal data collected between 2006 and 2012 from two informal settlements in Nairobi, Kenya. Mothers and their new-borns were recruited into the study at birth and followed prospectively. The analytical sample comprised 6858 children from 6552 households. Household food security was measured as a latent variable derived from a set of questions capturing the main domains of access, availability and affordability. A composite measure of wealth was calculated using asset ownership and amenities. Nutritional status was measured using Height-for-Age (HFA) z-scores. Children whose HFA z-scores were below -2 standard deviation were categorized as stunted. We used Cox regression to analyse the data. Results: The prevalence of stunting was 49 %. The risk of stunting increased by 12 % among children from food insecure households. When the joint effect of food security and wealth status was assessed, the risk of stunting increased significantly by 19 and 22 % among children from moderately food insecure and severely food insecure households and ranked in the middle poor wealth status. Among the poorest and least poor households, food security was not statistically associated with stunting. Conclusion: Our results shed light on the joint effect of food security and wealth status on stunting. Study findings underscore the need for social protection policies to reduce the high rates of child malnutrition in the urban informal settlements.
The study was conducted in two informal settlements—Korogocho and Viwandani—located in Nairobi, Kenya. The two study sites are part of the Nairobi Urban Health and Demographic Surveillance System (NUHDSS). The NUHDSS was initiated in 2002 to collect health and demographic statistics from an urban poor population. From 2003, the NUHDSS framework has provided opportunities for nesting studies, including the current study. For detailed description of the NUHDSS see Beguy et al. [25]. Data for this study come from the Maternal and Child Health (MCH) study (2006–2010), which was a sub-study of the broader Urbanization, Poverty and Health Dynamics project, and the INDEPTH Vaccination Project (IVP) study (2011 – 2013). The latter was a continuation of the MCH study. The MCH study was nested within the NUHDSS framework. The project targeted all women of reproductive health residing in the two study communities who gave birth between the duration of study—2006 to 2013. Under the MCH project, mothers and their new-borns were recruited upon delivery. The mothers and their children were then followed prospectively until the child was 5 years or until when they exited from the study either through death or out migration. Data were collected on the mother’s social demographic characteristics, her health seeking behaviour during and after delivery, feeding practices, immunization of the child as well as the anthropometric measures for both the child and mother. Upon recruitment, three follow-up visits, in this study referred to as updates, were made each calendar year. The following variables were extracted from two data sets: 1) Child characteristics that include date of birth, date of recruitment and subsequent visits, gender of the child, immunization, anthropometric measures, and birth weight and; 2) maternal characteristics that included the mother age at birth, parity, education level and health seeking behaviour. By 2013, 7452 children had been recruited to the study. During analysis, we excluded children with missing information on stunting between the ages 6 and 23 as well as those who were lost to follow-up before they attained the age of 6 months. The final sample consisted of 6858 children contributing to 101,686 person months. The dependent variable is stunting, which is Height for Age (HFA). Stunting is used here because it is a measure of long term food deprivation (chronic malnutrition) and illness making it a good indicator of child nutrition [26]. We calculated z-scores for the HFA using the ‘WHO Child Growth Charts and WHO Reference 2007 Charts’ for children aged up to 2 years. This was suitable because analysis was restricted to children aged between 6 and 24 months. The entry age was set at 6 months, which marks the end of exclusive breastfeeding and introduction to complimentary feeding. The Z-scores show the number of standard deviations of a child on a particular anthropometric measure in relation to a mean or median value. In this regard, those with z-scores of 2 standard deviations of height for age below the WHO reference median were categorized as stunted. Those with a score of above 2 standard deviations were categorized as normal (not stunted) [27]. Child anthropometric measures were obtained during each visit and therefore, stunting was calculated at the each points of visit. The primary independent variable was household food security status. Food security exists “when all people at all times have access to sufficient, safe, nutritious food to maintain a healthy and active life” [28]. Food security status was computed from a set of questions that captured the domains of food access as described in Radimer framework [29] . The questions assessed the frequency in the 30 days preceding the survey with which households: did not have adequate food; were worried about food availability; lacked enough money to purchase food; and children and adults had to forgo food for a whole day because there was not enough food. The response were coded as either ‘0 = never true’, ‘1 = sometimes true’ and ‘2 = often true.’ The respondent to the food security component was the household head who in his or her absence, the spouse or someone who had enough information and was credible enough was interviewed. Responses were recoded into binary responses: ‘often true’ were coded as ‘1’ and the rest ‘0’ and as described by [30]. We tested for agreement between the items and found a Cronbach’s Alpha of 0.72, indicating a good item reliability [31]. A composite score was generated by summing the items and categorized as 1 = food secure (score of 0); 2 = moderate food insecure (score of 1 or 2); and 3 = severely food insecure (score of more than 2). The second independent variable was the household asset wealth index, a latent variable computed from a composite measure of household assets and amenities. Principal Component Analysis (PCA) was used to reduce the multidimensional nature of the data to a single score that was categorized into three groups: Poorest, middle poor and least poor [32]. Data were managed and analysed in STATA 13.1. Both descriptive and inferential statistics were used for analysis. Frequencies and percentages were computed to describe the key socio-demographic characteristics of the study sample. In addition, descriptive statistics were used to estimate the prevalence of stunting in the sample. As multiple measures on stunting exist for each child, the exposure, which is the age of the child, was calculated for each visit. We used Cox regression models to estimate the survival time from age six to first stunting and to assess whether the survival time significantly varied by household food security status. The Cox regression models allowed us to control for other known determinants of stunting. In our study we restricted analysis to time to the first stunting. We tested the assumption of proportional hazard in the Cox regression, which is that the hazards are constant between the food security status and wealth status categories being compared. To test this assumption, we used Kaplan-Meier Curves and the log rank test. We also tested the assumption by interacting survival time with time varying covariate. Different Cox regression models were fitted: 1) unadjusted models with the key independent variables that included food security and household; 2) adjusted Cox regression with two main covariates—food security and wealth index; 3) a fully adjusted model, including all the covariates; and 4) a fully adjusted model with all covariates and the interaction between wealth index (poorest, middle poor and least poor) and food security status. The latter model was used to determine whether the effect of food security on stunting was the same across the wealth quintiles. For the Cox regression model, age in months was the main dependent variable with a dummy variable indicating whether the child is stunted or not. Each child was observed from birth between 2006 and 2013 until either the child was stunted, was censored due to loss of follow-up, out-migration or end of the follow-up for the child who aged above 23 months. Ethical clearance for both the MCH and IVP studies were granted by the Kenya Medical Research Institute (KEMRI). In addition, ethical clearance to use the data for secondary analyses was obtained from the University of the Witwatersrand, Human Research Ethics Committee and AMREF Kenya. Informed consent was obtained from all individual participants included in the study. All procedures were conducted in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
N/A
