Objective To investigate the predictors of wasting, stunting and low mid-upper arm circumference among children aged 6-59 months in Somalia using data from household cross-sectional surveys from 2007 to 2010 in order to help inform better targeting of nutritional interventions. Design Cross-sectional nutritional assessment surveys using structured interviews were conducted among communities in Somalia each year from 2007 to 2010. A two-stage cluster sampling methodology was used to select children aged 6-59 months from households across three livelihood zones (pastoral, agro-pastoral and riverine). Predictors of three anthropometric measures, weight-for-height (wasting), height-for-age (stunting) and mid-upper arm circumference, were analysed using Bayesian binomial regression, controlling for both spatial and temporal dependence in the data. Setting The study was conducted in randomly sampled villages, representative of three livelihood zones in Somalia. Subjects Children between the ages of 6 and 59 months in Somalia. Results The estimated national prevalence of wasting, stunting and low mid-upper arm circumference in children aged 6-59 months was 21 %, 31 % and 36 %, respectively. Although fever, diarrhoea, sex and age of the child, household size and access to foods were significant predictors of malnutrition, the strongest association was observed between all three indicators of malnutrition and the enhanced vegetation index. A 1-unit increase in enhanced vegetation index was associated with a 38 %, 49 % and 59 % reduction in wasting, stunting and low mid-upper arm circumference, respectively. Conclusions Infection and climatic variations are likely to be key drivers of malnutrition in Somalia. Better health data and close monitoring and forecasting of droughts may provide valuable information for nutritional intervention planning in Somalia.
In Somalia, a nutrition intervention group comprised of UNICEF, the World Food Programme and other government and non-governmental agencies was formed in 2006 to strengthen coordination of efforts against malnutrition( 6 , 19 ). This group developed a nutrition strategy for the period 2011–2013 in response to persistently high rates of malnutrition in the country. So far, several nutrition initiatives have been rolled out including the out-patient therapeutic feeding programmes for the management of severe acute malnutrition implemented by UNICEF and other agencies and the targeted supplementary feeding programmes for the management of moderately malnourished under-5s and pregnant and lactating women supported by the World Food Programme( 19 ). UNICEF currently targets 100 000 children aged 6–36 months with blanket distribution of ready-to-use food every two months in areas showing the highest malnutrition rates. The World Food Programme is also providing food assistance to vulnerable groups through institutional feeding and school feeding to about 90 000 beneficiaries. General food rations, consisting of cereals, corn–soya blend, sugar, fortified oil and iodized salt when available, are distributed to vulnerable rural populations, the urban poor and internally displaced persons. Despite these efforts, these interventions are thought to cover only a small proportion of the children under the age of 5 years who are likely to be malnourished in Somalia. The FSNAU cross-sectional surveys were conducted biannually during the long (April to June) and short (October to November) rainy seasons between 2007 and 2010. A stratified, multistage cluster sampling design was used where the sampling frame of a selected district was based on three livelihood definitions (pastoral, agro-pastoral and riverine), within which thirty rural communities and thirty households within each community were selected at random( 20 ). Surveys were undertaken in all three zones of Somalia (Fig. 1). (colour online) Map showing the distribution of clusters sampled during the Food Security and Nutritional Analysis Unit nutrition surveys conducted between 2007 and 2010 in Somalia. The country is divided into three main zones: North West, North East and South Central Sample sizes for the surveys (number of households and number of children) were calculated by Standardized Methodology for Survey in Relief and Transition (SMART) methods( 9 ). A list of all villages and population within each of the assessed livelihoods was used to estimate the total population for the assessment area. The selection of households within the village was done randomly from a list of eligible names or a map of households where possible. Where these were not available, the number of households in the village was estimated from the population figures (the total population divided by the mean household size)( 21 ). Detailed descriptions of the survey methods and data collection are provided elsewhere( 9 ). The spatial coordinates for each cluster were derived from several spatial databases( 22 ). Anthropometric measures were used to compute wasting and stunting using WHO 2006 references( 23 ). A child was defined as wasted or stunted when his/her Z-score for weight-for-age or height-for-age, respectively, was below −2. Additionally, children with MUAC below 125 mm were classified as having ‘low MUAC’. These measures were treated separately during analysis. The predictors for the present study were selected using both the WHO conceptual framework on childhood stunting( 24 ) and the UNICEF conceptual framework of child health and survival( 18 ). The underlying predictors were related to household, maternal and environmental factors. At the child level, vitamin A supplementation in the last 6 months, diarrhoea, acute respiratory infection and incidence of febrile illness in the last 2 weeks before the survey, polio and measles vaccination history, sex and age of the child were examined in the present study. In addition, information was collected on child age, weight, height, MUAC and access to staple foods as well the mother’s age and MUAC. For each household, information recorded included the household size and age structure, sex of the household head and access to different types of foods in the last 24 h. Detailed description of the variables can be found in the online supplementary material (Table S1). The effect of a set of five distal environmental covariates associated with vector-borne diseases( 25 ) and food security( 26 ) on the indices of malnutrition were examined. These were rainfall, enhanced vegetation index (EVI), mean temperature, distance to water features and urbanization. Rainfall and mean temperature were derived from the monthly average grid surfaces obtained from the WorldClim database( 27 ). The EVI values were derived from the MODerate-resolution Imaging Spectroradiometer (MODIS) sensor imagery( 28 ) for the period 2000–2010 while the urbanization information was obtained from the Global Rural Urban Mapping Project (GRUMP)( 29 ). All the environmental covariates were extracted from 1 km×1 km spatial resolution grids. Rainfall, temperature and EVI were summarized to compute seasonal averages using the four main seasons in Somalia: (i) December to March, the Jilal season, a harsh dry season; (ii) Gu which is the main rainy season from April to June; (iii) from July to September is the second dry season, the Hagaa; and (iv) the short rainy season known as Deyr from October to November. Further details of the covariates are provided in the online supplementary material (section S.1, ‘Data description’). Ethical approval was provided through permission by the Ministry of Health Somalia, Transitional Federal Government of Somalia Republic (ref. MOH/WC/XA/146./07, dated 02/02/07). Informed verbal consent was sought from all participating households and individuals. Three separate Bayesian hierarchical spatial-temporal regression models were used to analyse the predictors of stunting, wasting and MUAC among children under the age of 5 years. Model parameters were estimated using the Integrated Nested Laplace Approximation (INLA) algorithm for inference and was implemented in R project version 3·0·1 using R-INLA library( 30 ). Cluster-level effects were incorporated in the model to allow for the structured (spatial and temporal) and unstructured heterogeneity of malnutrition, using a convolution prior. District random effects were also included in this model. An assumption of additional flexibility in the model was made to allow for effects of non-linear predictors. Seasonality was controlled in the model as a factor with two unordered levels (April to June; October to November). A detailed description of the model procedures is provided in the online supplementary material (section S.2, ‘Spatial-temporal binomial regression model’).