South Asia continues to be the global hub for child undernutrition with 35% of children still stunted in 2017. This paper aimed to identify factors associated with stunting among children aged 0–23 months, 24–59 months, and 0–59 months in South Asia. A weighted sample of 564,518 children aged 0–59 months from the most recent Demographic and Health Surveys (2014–2018) was combined of five countries in South Asia. Multiple logistic regression analyses that adjusted for clustering and sampling weights were used to examine associated factors. The common factors associated with stunting in three age groups were mothers with no schooling ([adjusted odds ratio (AOR) for 0–23 months = 1.65; 95% CI: (1.29, 2.13)]; [AOR for 24–59 months = AOR = 1.46; 95% CI: (1.27, 1.69)] and [AOR for 0–59 months = AOR = 1.59; 95% CI: (1.34, 1. 88)]) and maternal short stature (height < 150 cm) ([AOR for 0–23 months = 2.00; 95% CI: (1.51, 2.65)]; [AOR for 24–59 months = 3.63; 95% CI: (2.87, 4.60)] and [AOR for 0–59 months = 2.87; 95% CI: (2.37, 3.48)]). Study findings suggest the need for a balanced and integrated nutrition strategy that incorporates nutrition-specific and nutrition-sensitive interventions with an increased focus on interventions for children aged 24–59 months.
This study is based on analysis of existing datasets in the Demographic Health Survey (DHS) repository that are freely available online with all identifier information removed. It utilised datasets from the most recent 2014–2018 DHS conducted within 5 South Asia countries including Bangladesh, India, Nepal, Maldives, and Pakistan. Data for other South Asian countries were not available through DHS due to the following reasons: Afghanistan does not collect anthropometric data for children under 5 years of age, data for Bhutan are unavailable on DHS and finally, data for Sri Lanka have restricted access and are not publicly available for research purposes. Data were obtained from a password-enabled DHS website [25]. The DHS data were nationally representative and population-based surveys, collected by country-specific ministries of health or other relevant government-owned agencies, with technical support largely provided by Inner City Fund (ICF) International. These surveys were comparable, given the standardised nature of the data collection methods and instruments [26]. The datasets were pooled to ascertain the most significant factors associated with child stunting and severe stunting across the South Asian countries. The DHS is a nationally representative survey that collects data on the health status of people, including reproductive health, maternal and child health, mortality, nutrition, and self-reported health behaviour among adults [26]. Information was collected from eligible women, that is, all women aged 15–49 years who were either permanent residents in the households or visitors present in the households on the night before the survey. Child health information was collected from the mother based on the youngest child aged less than five years, with response rates that ranged from 96% to 99%. Detailed information on the sampling design and questionnaire used is provided in the respective country-specific Measure DHS reports [25]. Our analyses were restricted to 564,518 children aged 0–59 months for 5 South Asian countries. The outcome variable was stunting (height-for-age). Stunting is an indicator of linear growth retardation and cumulative growth deficits in children. The height-for-age Z-score (HAZ), as defined according to 2007 WHO growth reference, expresses a child’s height in terms of the number of standard deviations (SD) above or below the median height of healthy children in the same age group or in a reference group. This study focused on children with a height-for-age Z-score below minus two standard deviations (−2 SD) as stunted and height-for-age Z-score below minus three standard deviations (−3 SD) as severely stunted. Prior to computing the prevalence and further analyses were undertaken, biologically implausible values (HAZ 6 SD) were excluded [27]. The choice of confounding factors used in this study was informed by the UNICEF framework [9]. The framework includes immediate factors including individual-level factors of diet and disease occurrence, underlying factors including household factors, and basic factors such as place of residence and country. The confounding factors were organised into five groups: (i) Immediate factors: dietary diversity score and child’s disease occurrence (episodes of diarrhoea and fever in the last two weeks), feeding practices such as currently breastfeeding and duration of breastfeeding, Vitamin A supplementation, Vaccination coverage, child’s age and sex; underlying factors (ii) Mother’s characteristics: such as age, age at birth, height, BMI, marital status, birth order and interval, maternal and paternal education, women’s power over household earnings, household decision-making and health care autonomy, (iii) Household factors: Pooled household wealth index, access to source of water and type of toilet which were categorised into improved and unimproved sources, (iv) Access and utilisation of services: Healthcare utilisation factors such as place and mode of delivery, combined birth rank (the position of the youngest under-five child in the family), and birth interval (the interval between births; that is, whether there were no previous births, birth less 24 months prior, or birth more than or equal to 24 months prior), delivery assistance, antenatal clinic visits (ANC) and access to media services, listening to the radio, watching television, and reading newspapers or magazines; (v) Basic factors such as country and place of residence (urban or rural). In order to reduce collinearity, we combined place of birth and mode of delivery and, birth order and birth interval. The combined mode of delivery and place of birth was divided into three categories as delivered at home, delivered at a health facility with non-caesarean section and delivered at a health facility with a caesarean section while, the combined birth order and the birth interval was classified as birth rank and birth interval, which is consistent with previous studies [28]. Maternal height was divided in the 5 following categories: <145 cm, 145–149.9 cm, 150–154.9 cm, 155–159.9 cm, and ≥160 cm, with <145 cm defined as short maternal height [29]. The household wealth index for the pooled dataset was constructed using the “hv271” variable. The hv271 variable used that principal components statistical procedure which was used to determine the weights for the wealth index based on information collected about 22 household assets and facilities and produce the standardised scores (z-scores) and factor coefficient scores (factor loadings) of wealth indicators. In the household wealth index categories, the bottom 20% of households were arbitrarily referred to as the poorest households, and the top 20% as the richest households, and was divided into poorest, poor, middle, rich, and richest. Dietary diversity (DD) was calculated by summing the 7 food groups consumed during the last 24 h. These foods are grains roots and tubers, legumes and nuts, Milk/dairy products, flesh foods (meat, fish, poultry and liver/organ meats), vitamin-A rich fruits and vegetables other fruits and vegetables and eggs, and were categorised into two groups, namely, the child had ≥4 food groups and the child had <4 food groups [30]. To examine factors associated with stunting among children aged 0–23 months, 24–59 months and children 0–59 months, the dependent variables were expressed as a binary outcome, i.e., category 1 [stunted (≥2 SD) or severely stunted (≥3 SD)]. For the combined 5 South Asian countries, a population-level weight, unique country-specific clustering, and strata were created to avoid the effect of countries with a large population (such as India with over 1.4 billion people in 2017 [31] offsetting countries with a small population (such as the Maldives with about 437,535 people in 2017 [32]. Population-level weights were used for survey (SVY) tabulation that adjusted for a unique country-specific stratum, and clustering was used to determine the percentage, frequency count and estimating the rates and 95% confidence intervals of child stunting in each country. Using three stages as described in Figure 1, the associations were further tested by odds ratios (OR) using univariate survey logistic regression analyses, and then hierarchical multiple survey logistic regression analyses. In the first stage model, basic factors were entered into the model. In the second stage model, underlying factors were added to the basic factors. A similar procedure was employed for the third stage model, which included the basic, underlying factors, as well as access to immediate factors. The aim of this hierarchical multiple logistic regression analyses was to allow for a comparison of the relationship between each of the different sets of factors described in Figure 1 in examining factors associated with stunting among children under 5 years. All statistical analyses were conducted using STATA/MP Version.14.1 (StataCorp, College Station, TX, USA) and adjusted odds ratios (AORs) and their 95% confidence intervals (CIs) obtained from the adjusted hierarchical multiple logistic regression model were used to measure the factors associated with child stunting. Conceptual framework of the determinants of child undernutrition.