Study setting
Ethiopia is a country with 94 million people, second largest among African countries and among the least urbanized countries in the world. The majority of the population resides in the highland areas [6]. The source of livelihood of the settled rural population is farming while the lowland areas are mostly inhabited by nomads, who depend mainly on livestock production and move from place to place in search of pasture and water.
There are 11 administrative regions in Ethiopia (9 regional states and two administrative cities); Tigray, Afar, Amhara, Oromia, Somali, Benishangul-Gumuz, Southern Nations Nationalities and Peoples (SNNP), Gambella, Harari, Addis Ababa, and Dire Dawa. Regions are divided into zones, and zones, into administrative units called woredas. Each woreda is further subdivided into the lowest administrative unit, called kebele. More than 80% of the country’s total population lives in the regional states of Amhara, Oromiya, and SNNP.
The 2011 EDHS is the third Demographic and Health Survey conducted in Ethiopia. It is intended to measure levels, patterns, and trends in demographic and health indicators. EDHS provides data on fertility, family planning, maternal and child health, childhood mortality, nutrition, malaria, HIV knowledge and behavior, and HIV prevalence [6].
Data source
Secondary analysis was performed using data from the 2011 Ethiopian Demographic Health Survey (EDHS), which is a nationally representative cross-sectional household survey of women of reproductive age and children less than five years old in Ethiopia. The data have been weighted to cater for the different sample proportions [6]. The survey was conducted from September 2010 to January 2011 and included three structured questionnaires: the Household Questionnaire, the Woman’s Questionnaire, and the Man’s Questionnaire.
EDHS sample design and procedure
A representative probability sample of 18,720 households was selected using a multistage stratified two- stage cluster sampling design in which samples of households within clusters (enumeration areas) are selected. This sample was constructed to allow for separate estimates of health indicators for each of the 11 geographic/administrative regions (nine regional states and two city administrations), as well as for urban and rural areas separately. A total of 624 clusters, 187 urban and 437 rural were selected from the sampling frame (The 2007 Population and Housing Census) in the first stage. In the second stage, a fixed number of 30 households were selected for each enumeration area. Of all the selected 18,720 households, 5610 are in urban areas and 13,110 are in rural areas [6].
Analytic sample and population
The EDHS sample design considers different parameters for the indicators to estimate the final sample size. In view of that, we have used children recode file with 7764 women and 11,654 children. But our study focused only on children whose mothers are not pregnant and living with their biological mothers.
Study variables
Outcome variables
Child undernutrition was defined along three anthropometric indices: underweight, stunting and wasting. Weight measurements were obtained using lightweight, SECA mother-infant scales with a digital screen, designed and manufactured under the guidance of UNICEF. Height measurements were carried out using a measuring board manufactured by Shorr Productions. Children younger than 24 months were measured for height while lying down, and older children, while standing [6]. The WHO 2006 growth standards were used to transform children’s weight and length/height measurements into sex- and age-specific Z-scores: height-for-age Z-score (HAZ), weight-for-age Z-score (WAZ) and weight-for-height Z-score (WHZ) [7]. Stunting was defined as HAZ below -2SD, underweight was defined as WAZ below -2SD while wasting was defined as WHZ below -2SD from the respective WHO 2006 growth standards reference median.
Exposure variables
The nutritional status of women was assessed by use of height and body mass index (BMI). To derive BMI, EDHS measured the height and weight of women age 15–49 years. BMI is used to measure thinness or obesity. BMI is defined as weight in kilograms divided by height in meters squared (kg/m2). A BMI below 18.5 kg/m2 indicates thinness or acute undernutrition. A BMI below 17 kg/m2 indicates severe undernutrition. A BMI of 25.0 kg/m2 or above indicates overweight or obesity. Height was also classified in a single cut off point < 145 cm as short stature.
Covariates
We included a number of theoretically important covariates that have been considered before in other studies on childhood undernutrition [8]. Child’s sex and age, maternal age, maternal education, place of residence (urban and rural). We also included a number of additional covariates: maternal smoking status, maternal parity and household wealth index.
Data analysis
DHS has developed recode files in order to facilitate data analysis. Recode files have standard data definitions across countries and across DHS phases. There are seven common types of recode data files associated with the core questionnaires. The datasets are available in the standard recode file formats in SPSS, SAS, Stata and CSPro; only completed questionnaires are included in these files. Among the types of the recode data files, children recode file is one of them. This is a dataset that has one record for every child of interviewed women, born in the five years preceding the survey. Therefore, we used the children recode data file in the form of Stata for analysis.
The available sample in the child recode file was 11, 654 children under age five and 7764 mothers. Of the 11,654 children, we excluded from the analysis children from pregnant mothers (1303); children not alive at the time of interview (n = 713); children not living with their mothers (n = 260); children not measured (n = 612); and values that are flagged and out the plausible limit (n = 261). The final data set comprised 8505 children aged 0–59 months (Fig. 1). We have not used child-mother pairs instead we have used all children and repeated their mothers. For example, if the mother had three children, she was repeated three times. It’s not recommended to match mother with the younger child by doing this we will introduce bias because the youngest child born in the last five years tend to be healthier than other children [9]. It would have been better to select the matched child at random but that means it’s impossible to match our results exactly. Statistical analysis was performed using the STATA software package, version 14.0 (Stata Corp., College Station, TX, USA). Its survey commands (svy) account for the complex sample survey data composition: strata, clusters, and weights.
We estimated weighted prevalence of stunting, underweight and wasting by maternal, child and household variables. Overall differences across the categories were tested with design-based Pearson chi-squared test. We also carried out correlation analysis to investigate possible associations between nutritional indices of the mother and the children and statistical significance was considered at the significance level of 5%. STATA does not give confidence intervals for correlations, so new command (corrci) was used to estimate confidence intervals. We then calculated sensitivity, specificity, predictive values and Area Under Roc Curve of the nutrition indices of the mother and the children using diagt command. Maternal BMI and height was used to predict the nutrition status of the children. The area under the ROC curve (AUC) determines the overall level of accuracy, with a value of 0.50 indicating purely random performance and 1.00 indicating the maximal value possible. According to an arbitrary guideline, one could distinguish between non-informative (AUC = 0.5), less accurate (0.5 < AUC ≤0.7), moderately accurate (0.7 < AUC ≤ 0.9), highly accurate (0.9 < AUC < 1) and perfect tests (AUC = 1, [10]).
Regression analysis
In evaluating the predictors of child undernutrition, logistics regression for binary outcomes was used. Unadjusted and adjusted odd ratios from logistic regression with corresponding 95% confidence interval was used to assess the significance and the magnitude of the effects of the given exposure.