A descriptive, cross-sectional study with multivariate analysis, using data from the 2010 Colombian National Nutritional Survey (ENSIN, from its initials in Spanish), was designed.
Data and sample
ENSIN 2010 was a joint effort of Colombian governmental and non-governmental organizations, which was supported by the United Nation’s World Food Program and the Pan-American Health Organization. The survey was applied to a nationally representative sample of 50,670 urban and rural households, which represent more than 99% of the Colombian population [15].
For this study, the initial sample included 4498 children, which comprised children from 12 to 59 months of age, who were included in the ENSIN. For the analysis, children with more than 10% of information missing in the survey (n = 223) were excluded from the analysis, for a final sample of 4275.
Outcome measure
Trained bacteriologists went to the children’s houses, after signing an informed consent; they applied the surveys to the parents and collected the blood samples from the children, between 6 and 9 mL, by venepuncture of the median cubital vein. ENSIN determined the zinc levels using atomic absorption spectrophotometry (AA6300 Shimadzu) following the Colombian National Institute of Health standardized protocols [15]. For the purpose of the present study, zinc deficiency was recoded as a dichotomous variable, for which a serum level of less than 65 μg/dl on a non-fasting serum sample was deemed to be a deficient serum level (zinc deficiency, 1 = Yes and 0 = No).
Independent variables
1) Self-reported information of enrolment in any nutritional support program, whether regional or national wide. This variable shows if a child is a beneficiary of a subsidized nutritional support program that provides at least one meal a day (one = Yes and 0 = No). 2) The wealth of the child’s household. This measure was created by the World Bank and Macro International to systematically determine a household’s relative economic status [19]. It gives each household a score based on a principal component analysis of the income, availability and quality of utilities, number of rooms, dwelling materials, type of cooking fuel, and availability of durable consumer goods. For the analysis, it was divided into quintiles (very rich, rich, average, poor and very poor). 3) Food security, which was assessed using the 2009 Latin-American and Caribbean household food security scale (ECLA), which is a validated scale based on household experiences [22]. For the analysis, food security was coded as a dichotomous variable (1 = Secure and 0 = Insecure).
Control variables
The following control variables were included: ethnicity (recoded as a dummy variable, namely, Majority, Native-Colombian, Afro-Colombians and others), health coverage (1 = Yes and 0 = No), age in years, sex (1 = Girls and 0 = Boys), body mass index (BMI), maternal education level (recoded as a dummy variable, namely, Lack of education, Elementary, High school and Superior education), and area of residence (1 = Urban and 0 = Rural).
The serum vitamin A, ferritin, haemoglobin and C-reactive protein (CRP) levels and the weight and health status were included in the initial analysis but excluded from the final analysis because they failed to show any association.
Data analysis
SPSS 22.0 (IBM) was used for the data processing. Initially, descriptive statistics were obtained, and logistic bivariate regressions were estimated for zinc deficiency on wealth, food security and enrolment in a nutritional support program. Finally, stepwise logistic multivariate regressions of zinc deficiency were performed. In the first models, wealth and food security were included, then enrolment in a nutritional support program was added, and finally, a complete model was computed by adjusting the previous models by all of the control variables.
Moderator analyses to find the possible effects of enrolment in nutritional programs on the associations of wealth and food security with zinc deficiency were conducted by multiplying the variables and introducing the terms into the regression models.