Association between stunting and early childhood development among children aged 36–59 months in South Asia

Abstract Stunting (length‐for‐age z score < −2) before 2 years of age has shown associations with poor child developmental indicators, but information at the population level is scarce in South Asia, the region with the highest burden of stunting. We examined associations between z scores (i.e., height for age [HAZ], weight for age [WAZ], and weight for height [WHZ]) and undernutrition (i.e., stunting [HAZ < −2], wasting [WHZ < −2], and underweight [WAZ < −2]) with learning/cognition and social–emotional development among children 36–59 months of age. Data from Multiple Indicator Cluster Surveys in Bangladesh (n = 8,659), Bhutan (n = 2,038), Nepal (n = 2,253), and Pakistan (Punjab n = 11,369 and Sindh n = 6,718) were used. Children were considered developmentally “on‐track” in learning/cognition or social–emotional domains if they met specific early child development criteria. Meta‐analysis was conducted to examine regional associations, adjusting for socio‐economic status, early childhood education, and quality of care. In a pooled sample, on‐track learning/cognition development was positively associated with HAZ (OR = 1.17, 95% CI [1.07, 1.27]) and WAZ (OR = 1.18, 95% CI [1.07, 1.31]) and negatively associated with stunting (OR = 0.72, 95% CI [0.60, 0.86]) and underweight (OR = 0.75, 95% CI [0.66, 0.86]) but not associated with WHZ or wasting. On‐track development of social–emotional domain was not associated with any z scores or undernutrition indicators. Across several countries of South Asia, stunted children were less likely to be developmentally “on track” for learning/cognition. It is likely that interventions that prevent stunting may benefit child development, leading to significant individual and societal gains given the large burden of child stunting in regions like South Asia.

, and, more broadly, the stimulation and stress that covary with socio-economic well-being (Walker et al., 2011).
Human neurodevelopmental processes begin within weeks after conception and continue through infancy (Couperus & Nelson, 2006). Animal models, supported by human studies, reveal that gestational nutrient deficiencies may impair neuron proliferation, dendritic branching, and synapse formation and function during pregnancy and infancy (Prado & Dewey, 2014). Postnatally, severe wasting, stunting, and deficiencies in some micronutrients (e.g., iron, zinc, vitamin B12, and iodine) have been shown to be associated with developmental deficits during early to midchildhood (John et al., 2017) and may extend intergenerationally. For example, Jamaican children born to parents who had been stunted before age 2 years scored lower on cognitive development scales than did peers born to parents who were not stunted in early childhood (Walker, Chang, Wright, Osmond, & Grantham-McGregor, 2015). Trials evaluating prenatal and childhood micronutrient supplementation have revealed positive effects on cognitive and executive functioning performance, educational attainment, learning, and reading ability at a later age compared with controls Walker et al., 2011).

Nationally representative Demographic and Health Survey and
Multiple Indicator Cluster Surveys (MICS) data from 35 low-and middle-income countries (LMICs) estimate that approximately 81.0 million (33.0%) children aged 3 and 4 years have low cognitive and/or socialemotional development (McCoy et al., 2016). In an analysis of crosssectional data from 11 LMICs, each unit increase in height-for-age z score was associated with a 0.24 SD increase in cognitive score before age 2 years, although the association was attenuated to 0.09 SD from 2 to 12 years of age . Globally, the largest number of stunted children resides in South Asia, where~33.3%, or 58.7 million, of children under 5 years of age are affected (UNICEF, WHO, & World Bank Group, 2018). Furthermore, the region also harbours the second-highest number (27.8 million) of children with low cognitive and socioemotional Early Childhood Development Index (ECDI) test scores after the Sub-Saharan Africa region (McCoy et al., 2016), justifying our interest to explore associations between preschool child nutritional status and development in South Asia.
Accessing population-level data from recent-year MICS in Bangladesh, Bhutan, Nepal, and Pakistan, the present study examined associations between undernutrition and learning/cognition and socialemotional development indicators among South Asian children aged 36 to 59 months. We further tested the consistency of effect estimates across the region by meta-analysis.

| Available variables
An ECDI assessment, introduced into the fourth and fifth rounds of MICS, was designed to assess development in children aged 36-59 months . The ECDI is based on responses to 10 yes/no questions by primary caregivers. An initial validation study for ECD was conducted in Jordan and the Philippines, comprising 900 children aged 3-4 years in each country, with reliability testing for a subsample (Loizillon, Petrowski, Britto, & Cappa, 2017). After a series of factor analyses, ECD was constructed as long (48-item) and short (18-item) sets of items covering six child development domains (Janus & Duku, 2012). The 18-item set draft ECDI was field tested in Kenya (Janus & Duku, 2011;Plowman, 2014), and the final revisions were made, leading to the current 10-item set for four child development domains: literacy-numeracy (three items), physical (two items), learning/cognition (two items), and social-emotional (three items). Recently, McCoy et al. (2016) demonstrated that ECDI items designed to assess literacy/numeracy were too advanced for 3-and 4-year-old children and that the sensitivity of the physical domain module was low, detecting only more extreme conditions. For these reasons, our current analysis was restricted to the learning/cognition and socioemotional domains.
Based on ECDI criteria (UNICEF, 2014), learning/cognition development is considered to be on track "if a child can follow simple directions on how to do something correctly and/or when given something to do, is able to do it independently"  • Interventions that promote linear growth and prevent child stunting may contribute to improve child development outcomes in South Asia. defined as the percent of children whose HAZ, WAZ, and WHZ, respectively, were two standard deviations below the WHO reference median for children of the same month of age and sex (WHO Multicentre Growth Reference Study Group, 2006).

| Confounding variables
We selected socio-economic, early childhood education, and quality of care variables as potential confounders to see how they affected the associations between undernutrition and child development. Socioeconomic variables were selected at the individual, maternal, and household levels if they were conceptually related to child nutritional status and development and were available in the five surveys. Household variables included rural/urban residence, household wealth index, household size (greater than five members or not), gender of household head, use of improved drinking water sources, and use of improved toilet facilities. Household wealth index quintiles were generated by principal components analysis using information on the dwelling characteristics, including water sources and toilet facilities, fuel types for cooking, durable assets, livestock, electricity, and other characteristics related to household wealth. Maternal education attainment was also included, as were child sex and age. Reported episodes of diarrhoea and cough in the past 2 weeks were also included as individual level characteristics.
Early childhood education and quality of care were assessed in the child development section of the MICS-Rounds 4 and 5 (UNICEF, 2013), based on the mother's or caretaker's report, as they are considered to be essential to child development. Attendance at early childhood education was defined as attending an early childhood education programme (i.e., private or government facility including kindergarten or community child care). Support for learning was defined as any family member over 15 years of age having engaged in four or more activities to promote learning and school readiness in the 3 days prior to the interview: (a) reading a book(s) or looking at a picture book(s) with the child; (b) telling a story to the child; (c) singing song(s), including lullabies, to or with their child; (d) taking the child outside the home, compound, yard, or enclosure; (e) playing with the child; and (f) naming, counting, or drawing things with the child. Availability of children's books was defined as having three or more children's books in the home. Availability of playthings was defined as the child playing with two or more types of playthings out of homemade toys, toys from a shop, and household/outside objects (sticks, rocks, animal shells, or leaves). Inadequate care was defined as the child's being left alone or in the care of another child younger than 10 years for more than 1 hr at least once in the last week.

| Statistical analysis
The two-stage cluster sampling design of the original surveys, which randomly selected households with children under 5 years in enumeration areas (primary sampling unit) from the national census, was taken into account in all analyses. "svy" commands were used to estimate summary statistics and associations at the population level. Learning/ cognition and social-emotional variables were dichotomized as being on track versus not. To see the associations between z scores/undernutrition and developmental indicators, we first explored the bivariate logistic regression analysis (Model 1). Next, the models were adjusted for socio-economic variables (Model 2), and finally models were adjusted for early childhood education and quality of care in addition to the socio-economic variables (Model 3). We conducted meta-analysis and created forest plots. Adjusted odds ratios and 95% CIs from Model 2 and Model 3 were log transformed, and the "metan" command with the option of "eform effect (OR)" was used. Heterogeneity (I 2 ) of association across pooled country datasets was measured; I 2 greater than 50% or I 2 P value < 0.05 indicated significant heterogeneity (Higgins, Thompson, Deeks, & Altman, 2003). Heterogeneity was incorporated into a random-effects model that allows different population estimates of effect size to be combined as recommended by the Cochrane guidelines (Higgins et al., 2003). Stata Version 14 (StataCorp LP, College Station, TX) was used for all statistical analyses.

| Ethical considerations
This secondary data analysis was deemed exempt from ethics review by the Institutional Review Board, Johns Hopkins School of Public Health, as no human subjects work was conducted as part of this project.

| General characteristics
This study included nationally representative children 36 to 59 months of age in Bangladesh (n = 8,659), Bhutan (n = 2,038), Nepal (n = 2,253), and Punjab (n = 11,369) and Sindh (n = 6,718) provinces in Pakistan (Table 1). The proportion of households from rural areas varied from 53.7% in Sindh province in Pakistan to 86.7% in Nepal. The proportion of households with five or more members was greatest in Sindh, Pakistan (92.3%), and smallest in Bangladesh (54.0%). The proportion of female-headed households was highest in Nepal (27.4%), followed by Bhutan (22.8%), and low in both provinces in Pakistan (4.4% and 5.8% in Sindh and Punjab, respectively). In all countries, most households had access to improved drinking water (89.5-96.4%), and roughly half of households used improved toilet facilities (50.0-60.0%). The proportion of mothers who did not complete primary education or had no formal education was lowest in Bangladesh (40.7%) and highest in Bhutan (67.1%). The proportion of children experiencing diarrhoea and cough in the past 2 weeks varied from 1.8% in Bangladesh to 18.3% in Sindh, Pakistan, and 7.7% in Punjab, Pakistan, to 45.4% in Bhutan, respectively.
The proportion of children attending early childhood education programmes was low overall, ranging from 10.3% in Bhutan to 51.3% in Nepal (Table 1). The proportion reporting support for child learning from family members was lowest in Pakistan (51.6% in Punjab and 56.2% in Sindh) and highest in Bangladesh (87.4%). Availability of three or more children's books was low at <20% in all countries, whereas the availability of playthings was moderately high at 61.1-74.3%. The proportion of children with inadequate care was largely low, ranging from 7.9% in Punjab, Pakistan, to 26.0% in Nepal.

| Early childhood development indicators
The proportion of children who had on-track development in the learning domain was high, ranging from 82.4% in Nepal to 93.2% in Bhutan and 94.3% in Punjab, Pakistan (Figure 2). The proportion of When testing with continuous nutritional status z scores, we found high heterogeneity in the effect size across samples, but the pooled OR was accepted by using random-effects meta-analysis (    Where examined, wasting shares an inconsistent relationship with child development indicators . In our analysis, only Pakistan showed significant associations between wasting or WHZ and learning/cognition development, possibly due to higher burden of wasting (>11%) in the country than other countries (4.1~8.3%).
When early learning and better family care were introduced into our risk factor analyses, there were slight attenuations in otherwise Model 2: Adjusted for location (rural vs. urban), household wealth quintile, household size, type of household head (male vs. female), improved drinking water source (yes/no), improved toilet facility (yes/no), maternal education, child age and sex, episodes of diarrhoea and cough in the past 2 weeks (yes/no), and study design effect. d Model 3: In addition to all confounding variables used in Model 2, attendance to early childhood education, support for learning, availability of children's books, availability of playthings, and inadequate care were adjusted for in Model 3. e I 2 is defined as the percentage of total variation across studies that is due to heterogeneity. strong negative associations between stunting and developmental indices, suggesting that early childhood care practices, as practised and reported in the surveyed home environments, may only partially ameliorate longer acting, negative effects of undernutrition on developmental capacity. The present findings support the implementation of interventions that may improve growth and reduce stunting in the first years of life to positive effects on early childhood development (Bhutta et al., 2013). Examples of such interventions include improved maternal nutrition to reduce the risk of maternal anaemia and small for gestational age or low weight birth, optimal breastfeeding and complementary feeding practices, micronutrient supplementation and food fortification, good hygiene and safe food handling practices, and timely detection and treatment of common childhood illnesses. The present findings also suggest that the integration of child stimulation, early learning, and better family care (Nguyen et al., 2017) with nutrition interventions is likely to yield the best results (Walker, Chang, Powell, & Grantham-McGregor, 2005).

| Strengths and limitations
Among limitations of this study is the cross-sectional nature of the data, making it not possible to infer causality or mediation in the observed associations between child undernutrition and development indicators. It remains to be causally established that linear growth deceleration leading to stunting may adversely affect child development (Larson & Yousafzai, 2017;Perkins et al., 2017

| CONCLUSIONS
In the South Asia region, stunting and underweight were associated with poorer learning developmental achievement among children Model 2: Adjusted for location (rural vs. urban), household wealth quintile, household size, type of household head (male vs. female), improved drinking water source (yes/no), improved toilet facility (yes/no), maternal education, child age and sex, episodes of diarrhoea and cough in the past 2 weeks (yes/no), and study design effect. d Model 3: In addition to all confounding variables used in Model 2, attendance to early childhood education, support for learning, availability of children's books, availability of playthings, and inadequate care were adjusted for in Model 3. e I 2 is defined as the percentage of total variation across studies that is due to heterogeneity. 36-59 months old. On the contrary, none of nutrition indicators were associated with social-emotional development. Associations varied by domain and country. Our findings are consistent with emerging evidence that interventions that improve linear growth and reduce stunting may have a positive impact on development outcomes in infancy and early childhood. Integration of nutrition interventions and strategies to improve early stimulation and learning are likely to have synergistic positive impact on early childhood development and should be an investment priority in the South Asia region.