Infant and young child feeding practices and child linear growth in Nepal: Regression–decomposition analysis of national survey data, 1996–2016

Abstract Suboptimal infant and young child feeding (IYCF) practices have profound implications on child survival, health, growth, and development. First, our study analysed trends in 18 IYCF indicators and height‐for‐age z‐score (HAZ) and stunting prevalence across Nepal's Family Health Survey 1996 and four rounds of Nepal Demographic and Health Surveys from 2001–2016. Second, we constructed multivariable regression models and decomposed the contribution of optimal IYCF practices on HAZ and stunting prevalence over the 1996–2016 period. Our findings indicate that most age‐appropriate IYCF practices and child linear growth outcomes improved over the past two decades. At present, according to the World Health Organization's tool for national assessment of IYCF practices, duration of breastfeeding is rated very good, early initiation of breastfeeding and exclusive breastfeeding (EBF) are rated good, whereas minimal bottle‐feeding and introduction of solid, semi‐solid or soft foods are rated fair. Our study also reports that a paucity of age‐appropriate IYCF practices—in particular complementary feeding—are significantly associated with increased HAZ and decreased probability of stunting (p < .05). Moreover, age‐appropriate IYCF practices—in isolation—made modest statistical contributions to the rapid and sustained reduction in age‐specific child linear growth faltering from 1996–2016. Nevertheless, our findings indicate that comprehensive multisectoral nutrition strategies—integrating and advocating optimal IYCF—are critical to further accelerate the progress against child linear growth faltering. Furthermore, specific focus is needed to improve IYCF practices that have shown no significant development over the past two decades in Nepal: EBF, minimum acceptable diet, and minimal bottle‐feeding.


| INTRODUCTION
Child malnutrition is a major public health concern worldwide. At present, an estimated 149 million children under five are stunted, whereas the lives of over 49 million children continue to be threatened by wasting (UNICEF, WHO [World Health Organization], & World Bank Group, 2019). These malnourished children are at a higher risk of mortality and poor health, growth, and development ).
The causes of child linear growth faltering are complex, multidimensional, and interrelated. UNICEF's (1991) framework, updated by Black et al. (2013), propagates that nutrition outcomes are the end result of a causal chain of determinants. Among the most proximal and immediate determinants are suboptimal infant and young child feeding (IYCF) practices. Optimal IYCF is essential for child nutrition, linear growth, and cognitive development WHO, 2008WHO, , 2010).
An estimated 14% of deaths in children aged 0-23 months might be prevented by scaling up breastfeeding to universal levels, and at least 6% of mortality in children under five might be prevented by adequate complementary feeding (Black et al., 2013;Victora et al., 2016). Furthermore, age-appropriate IYCF practices are directly linked to the attainment of at least four of the United Nation's Sustainable Development Goals: nutrition, health, poverty reduction, and inequity reduction (Rollins et al., 2016).
In South Asia, at least one in three children under five is stunted (34.4%), and it is the only region with a very high wasting prevalence (15.2%;UNICEF et al., 2019). Over the past two decades, South Asian nations-in particular Nepal-have achieved unprecedented progress on reducing chronic child undernutrition (Cunningham, Headey, Singh, Karmacharya, & Rana, 2017;Headey, Hoddinott, & Park, 2017). However, recent research has reported that maternal and caregiver IYCF knowledge is suboptimal (Cunningham, Headey, et al., 2017;Senarath et al., 2012), and most breastfeeding (Benedict, Craig, Torlesse, & Stoltzfus, 2018) and complementary feeding practices (Na et al., 2018) have improved only marginally in recent years.
In this study, we address the research gap with regard to long-term trends in age-appropriate IYCF practices and their contribution to the rapid and sustained progress against child linear growth faltering in Nepal. We aim to answer three interrelated questions: (a) What are the long-term trends in 18 IYCF indicators, height-for-age z-score (HAZ) and stunting prevalence in Nepal? (b) which IYCF indicators are associated with age-specific HAZ and stunting? and (c) which IYCF practices potentially account for variations in child linear growth faltering observed from 1996-2016? We also briefly discuss the policy and programme implications of our findings to inform actions to protect, promote, and support IYCF practices in Nepal.

| Data sources
The estimates reported in this research paper were constructed using individual child-level data from the following cross-sectional surveys: Nepal's Family Health Survey (NFHS) 1996 and four rounds of Nepal Demographic and Health Survey (NDHS) 2001. These data are well suited to our research, as the five surveys are high quality, nationally representative, standardised across rounds and cover a broad-albeit non-exhaustive-range of hypothesised nutrition-sensitive and nutrition-specific determinants of child linear growth and anthropometric measurements.

| Outcomes
HAZ and stunting (HAZ < −2 SD) were measured against the median of the WHO (2006) Child Growth Standard. Linear growth is generally regarded as the single most relevant indicator of overall child nutrition status, and the reduction of stunting is the standard metric of longterm nutritional, educational, and economic progress (Dewey & Begum, 2011;Hoddinott et al., 2013).

Key Messages
• Nepal made substantial progress on infant and young child feeding (IYCF) practices and linear growth outcomes from 1996-2016.
• A paucity of IYCF indicators-in particular complementary feeding practices-are significantly associated with height-for-age z-score and stunting prevalence.
• Age-appropriate IYCF practices-in isolation-made modest statistical contributions to the rapid and sustained reduction in age-specific child linear growth faltering over the past two decades.
• Nevertheless, the integration and monitoring of IYCF policies, programmes, health systems, workplaces, and communities remains critical to further accelerate the progress against chronic child undernutrition in Nepal.
for non-breastfed children. Our analysis also examined three additional indicators: avoidance of prelacteal feeding in the first 3 days following delivery, optimal early breastfeeding, and consumption of animal-source foods (ASF). Furthermore, covariates from child-, parental-, and household-level were selected based on the Black et al. (2013) framework and a review of previous regression-decomposition analyses of chronic child undernutrition (Cunningham, Headey, et al., 2017;Headey et al., 2017;Headey, Hoddinott, Ali, Tesfaye, & Dereje, 2015;Headey, Hoddinott, & Park, 2016;Menon, Headey, Avula, & Nguyen, 2018). These time-variant independent variables included a number of hypothesised nutrition-sensitive and nutritionspecific determinants of child linear growth faltering, namely, household asset index (Filmer & Pritchett, 2001), maternal education, paternal education, prenatal doctor visits, 4 + antenatal care visits, iron during pregnancy, childbirth in a medical facility, maternal body mass index and height, child vaccination, birth order, birth interval, improved household sanitation, and drinking water source (Table S1).
Our multivariable regression models also included time-invariant control variables, namely, dummy variables for child age (months), child sex, region, agroecological zone, religion, ethnicity, maternal age (5year intervals), and survey rounds. 8. Consumption of iron-rich or iron-fortified foods Proportion of children 6-23 months of age who receive an iron-rich or iron-fortified food that is specially designed for infants and young children, or that is fortified in the home Note. Indicators 2-8, 10-12, and 14-15 are based on a 24-hr recall period. Indicators 1, 2, 7, and 8 are considered top priorities for reporting among the core indicators. Indicator 2 can be disaggregated for ages 0-1, 2-3, 4-5, and 0-3 months. The seven food groups mentioned under indicator 5 are grains, roots, and tubers; legumes and nuts; dairy products (milk, yogurt, cheese); flesh foods (meat, fish, poultry, and liver/organ meats); eggs; vitamin A-rich fruits and vegetables; other fruits and vegetables. Minimum number of times mentioned under indicator 6 is defined as: two times for breastfed infants 6-8 months; three times for breastfed children 9-23 months; four times for non-breastfed children 6-23 months. Indicator 7 is the sum of two fractions: (1) the proportion of breastfed children 6-23 months of age who had at least the minimum dietary diversity and the minimum meal frequency during the previous day; plus (2) the proportion of non-breastfed children 6-23 months of age who received at least two milk feedings and had at least the minimum dietary diversity and the minimum meal frequency during the previous day. Indicator 11 is the sum of exclusive breastfeeding under 6 months plus the proportion of children 6-23 months of age who received breast milk as well as solid, semi-solid, or soft foods during the previous day.

| Statistical analysis
Data management and statistical analysis were conducted in Stata In contrast to previous studies, our research applied a quantitative and dynamic regression-decomposition approach, which included IYCF indicators as determinants of child linear growth faltering. Our statistical decompositions comprised two distinct steps: (a) use multivariable regression models-ordinary least squares (OLS) for HAZ and linear probability model for stunting-to assess age-specific associations (β-coefficients) between age-appropriate IYCF practices and linear growth outcomes among Nepalese children and (b) decompose the potential contribution of age-appropriate IYCF practices on agespecific progress against chronic child undernutrition over the 1996-2016 period. Independent variables with variance inflation factors ≥ 4 were omitted from our regression models to avoid potential collinearity. p values < .05 were considered statistically significant for associations between IYCF indicators and child linear growth outcomes.
Our multivariable regression models are represented in Equation (1) below, assessing the associations between linear growth outcomes In a simple decomposition at means described in Equation (2), in which β-coefficients are assumed to be stable over time, the estimated contribution of an IYCF indicator (X) on a linear growth outcome (N) is the product of its β-coefficient and the change in its mean over time.
Hence, an IYCF indicator will make a substantial contribution if its βcoefficient is large and if its mean score changes considerably over time. For our analysis, we selected the earliest (NFHS 1996) round (t = 1) and the most recent (NDHS 2016) round (t = k) available.
To assess the assumption of stable IYCF β-coefficients, we conducted an Oaxaca-Blinder decomposition testing for systematic differences in β-coefficient between the NFHS 1996 and NDHS 2016 rounds (Jann, 2008). Furthermore, we also estimated models that excluded potentially endogenous determinants of optimal IYCF, including wealth, parental education, and healthcare-related service variables. Lastly-to determine potential attenuation of β-coefficients by younger children-we tested the associations between age-appropriate complementary feeding practices and linear growth outcomes for children aged 18-23 months (Alderman & Headey, 2018).
Our research paper did not aim to assess the multifactorial determinants of optimal IYCF and child linear growth in Nepal (Benedict et al., 2018;Headey et al., 2017;Na et al., 2018), nor emulate the detailed definitions, implications, and rationales behind the WHO and UNICEF IYCF indicators (Jones et al., 2014;UNICEF, 2016;WHO, 2008WHO, , 2010.
3.1 | Trends in age-appropriate IYCF practices     how much of the changes in age-specific child linear growth faltering is explained by IYCF practices over the past two decades in Nepal?

| Regression-decomposition
The estimated β-coefficients are reported in   Our findings in Table 5   Source: Author's estimates.
Note. Clustered robust standard errors are reported in parentheses. Stunting (%) refers to HAZ < −2 SD. The regressions include a number of time-variant independent variables, including household asset index, maternal education, paternal education, prenatal doctor visits, four or more antenatal care visits, iron during pregnancy, born in a medical facility, maternal BMI, maternal height, vaccinations status, birth order, birth interval, open defecation, drinking water source, and time-invariant controls, including regional and agroecological fixed effects for seven groups, an urban dummy, dummy variables for religion and ethnicity, dummy variables for various maternal age groups (in 5-year intervals), dummy variables of child age (monthly), a child sex dummy and Family Health Survey and Demographic and Health Surveys round dummy variables (Table S1).
Note. Predicted nutritional change is based on a linear decomposition at means, in which changes in the mean of an age-specific IYCF indicator (Table 2) is multiplied by the corresponding adjusted β-coefficient (Table 4). Stunting (%) refers to HAZ < −2 SD. Actual changes in child linear growth outcomes are reported in Table 3.
IYCF has been increasingly recognised as a critical driver of improved child nutrition , and recent evidence has shown associations between age-appropriate IYCF practices and child linear growth outcomes ( and consumption of ASF-are associated with improved child linear growth outcomes, and second, the statistical contributions of ageappropriate IYCF practices on long-term age-specific increases in HAZ and reductions in stunting are modest. Hence, our findings support the notion that the aetiology of child linear growth faltering is complex and multifactorial Menon et al., 2018), and improving IYCF in isolation is insufficient to ensure further progress against chronic child undernutrition. Our findings also mirror previous studies that have indicated the critical role of MDD-reflecting micronutrient density adequacy of complementary foods-in relation to child linear growth outcomes (Arimond & Ruel, 2004;Jones et al., 2014;Menon et al., 2015). Nevertheless, child dietary diversity and diet quality remains a persistent concern in Nepal. Cunningham, Headey, et al. (2017) reported that solid or semi-solid complementary foods in Nepal consisted mainly of nutrient-poor rice, jaulo (rice and lentil porridge), soup, lito (roasted cereal and lentil porridge), Cerelac, and fruit juices, although ASF tended to be introduced only after 12 months of age (Cunningham, Headey, et al., 2017). Furthermore, there are widespread cultural beliefs that cereals are sufficient for child nutrition, with little emphasis on the importance of MDD or IRON for child linear growth and development (Gautam, Adhikari, Khatri, & Devkota, 2016). Moreover, ASF, green leafy vegetables, fruits, and yoghurt are generally not given to young children in Nepal due to cultural taboos and the common perception that these foods are difficult to digest (Locks et al., 2015;Siegel et al., 2006). Similar to recent studies, our findings indicate that consumption of ASF is associated with improved child linear growth outcomes (Headey, Hirvonen, & Hoddinott, 2018;Krasevec et al., 2017;Shapiro et al., 2019). However, the lack of sensitivity and specificity of many IYCF indicators might contribute to the relationships observed in our analyses. Therefore, stronger proxies of diet quality and quantity are warranted to elucidate how age-appropriate IYCF practices relate to child linear growth faltering over time (Jones et al., 2014).
Our findings on the lack of association between breastfeeding indicators and HAZ or stunting-and subsequent limited statistical contribution to the progress over time as compared with complementary feeding-should not be interpreted as breastfeeding being unimportant for child survival, growth, and development (Darmstadt et al., 2005;Jones, Steketee, Black, Bhutta, & Morris, 2003;Kramer & Kakuma, 2012). To illustrate, associations between EBF and child linear growth outcomes are often only apparent in later childhood because nutrition outcomes such as stunting are cumulative by nature and do not fully set in until the second year of life (Victora et al., 2010). Nevertheless, a review by Bhutta et al. (2008) showed that strategies for breastfeeding promotion have only small effects on child stunting.  .
Positive national-level changes in IYCF practices and child linear growth outcomes might also reflect improved socioeconomic status, parental education, and/or sanitation, as well as exposure to and utilisation of healthcare-related services (Cunningham, Headey, et al., 2017;.
Interpretations of causality in our research are limited by the use of cross-sectional data, potentially confounded by omitted variables.
Furthermore-although infant and young children's diets are often monotonous in low-and middle-income countries (Ruel, 2003)-ageappropriate complementary feeding practices were based on a single 24-hr recall per child-introducing random intraperson error-which might not be representative of "usual" dietary intake and thus attenuate our β-coefficients (Thorne-Lyman, Spiegelman, & Fawzi, 2014).
Hence, the directions of the relationships observed are only hypothesised, that is, age-appropriate IYCF practices are positively associated with child linear growth outcomes. The causal sequence of these relationships cannot be determined for the available data.
However, the plausibility of our findings are strengthened by the adjustment for several potential nutrition-sensitive and nutritionspecific confounders-hypothesised to be associated with child linear growth-and time-invariant controls. Furthermore, our analyses used the most comprehensive and highest quality national survey data sets for five time points between 1996 and 2016 in Nepal. The findings are also largely consistent with theoretical expectations and previous studies on the topic. Examining the associations between our model time-variant covariates and child linear growth outcomes was beyond the scope of this study.

| CONCLUSIONS
This is the first statistical decomposition to comprehensively investigate the contributions of the spectrum of IYCF indicators on agespecific HAZ and stunting prevalence over time. Long-term trends suggest substantial improvements in IYCF practices and child linear growth over the past two decades in Nepal. Nevertheless, our findings also indicate that only a few IYCF indicators-in particular complementary feeding-are significantly associated with improved child linear growth outcomes. Moreover, age-appropriate IYCF practices, in isolation, have modest contributions to the rapid and sustained reduction in age-specific child linear growth faltering from 1996-2016 in Nepal.
However, given the inextricable link between successful breastfeeding and complementary feeding, it important to aim to improve the entire continuum of IYCF practices in the first 2 years of life. Our findings indicate that multisectoral nutrition strategies that include approaches to protect, promote, and support IYCF are critical to further accelerate the progress against chronic child undernutrition in Nepal. Furthermore, specific focus is needed to improve age-appropriate IYCF practices that have shown no significant development over the past two decades, including EBF, MAD, and minimal bottle-feeding.

ACKNOWLEDGMENTS
The authors acknowledge UNICEF Nepal for funding this research and thank Harriet Torlesse (Nutrition Section, UNICEF Regional Office for South Asia) for her critical appraisal of the paper. H-C., A.A., P.D., and S.C. wrote the paper.

DATA AVAILABILITY STATEMENT
All our data are based on Demographic and Health Surveys, which are available at the Measure DHS website after appropriate registration: http://dhsprogram.com/data/available-datasets.cfm.