Trends and predictors of appropriate complementary feeding practices in Nepal: An analysis of national household survey data collected between 2001 and 2014

Abstract There is evidence that suboptimal complementary feeding contributes to poor child growth. However, little is known about time trends and determinants of complementary feeding in Nepal, where the prevalence of child undernutrition remains unacceptably high. The objective of the study was to examine the trends and predictors of suboptimal complementary feeding in Nepali children aged 6–23 months using nationally representative data collected from 2001 to 2014. Data from the 2001, 2006, and 2011 Nepal Demographic and Health Surveys and the 2014 Multiple Indicator Cluster Survey were used to estimate the prevalence, trends and predictors of four WHO‐UNICEF complementary feeding indicators: timely introduction of complementary foods (INTRO), minimum meal frequency (MMF), minimum dietary diversity (MDD), and minimum acceptable diet (MAD). We used multilevel logistic regression models to identify independent factors associated with these indicators at the individual, household and community levels. In 2014, the weighted proportion of children meeting INTRO, MMF, MDD, and MAD criteria were 72%, 82%, 36% and 35%, respectively, with modest average annual rate of increase ranging from 1% to 2%. Increasing child age, maternal education, antenatal visits, and community‐level access to health care services independently predicted increasing odds of achieving MMF, MDD, and MAD. Practices also varied by ecological zone and sociocultural group. Complementary feeding practices in Nepal have improved slowly in the past 15 years. Inequities in the risk of inappropriate complementary feeding are evident, calling for programme design and implementation to address poor feeding and malnutrition among the most vulnerable Nepali children.

in the rate of reduction of child stunting from an average annual rate of decrease of 3.4% per year between 2001 and 2011 to 2.6% per year between 2011 and 2014. In addition, the proportion of wasted children has stagnated at 11. 2-11.3% between 2001 and 2014, and Nepal currently ranks 111th out of 130 countries globally on nutrition indicators (International Food Policy Research Institute, 2016).
Among all the causes of stunted growth in children, inadequate child feeding is one of the most proximal and immediate determinants (Black et al., 2013;Stewart, Iannotti, Dewey, Michaelsen, & Onyango, 2013). Suboptimal child feeding, defined by the World Health Organization (World Health Organization, 2010), includes delayed initiation of breastfeeding, shorter duration (<6 months) of exclusive breastfeeding, discontinuation of breastfeeding before 2 years of age, delayed introduction of complementary foods, and inadequate quantity and quality of complementary foods given to children aged 6-23 months. In Nepal, children are predominantly breastfed in the first 6 months of life (75%) and are likely to continue to breastfeed until 2 years of age (87%), although the proportion of children who initiate breastfeeding within 1 hr of birth (49%) and are exclusively breastfed in the first 6 months of life (57%) is lower (Central Bureau of Statistics, 2015). Compared to the breastfeeding indicators, complementary feeding indicators in children aged 6-23 months reflect much poorer practices: It is estimated that only one in three Nepali children is fed with the minimum frequency and dietary diversity (Central Bureau of Statistics, 2015). It is critical to understand the main time trends in child feeding to better explain the observed trends in child's growth outcomes. However, changes over time in Nepali children's diets and in feeding practices have not been examined.
Better understanding the determinants of poor complementary feeding at the individual, household, and community levels can also help to understand the epidemiology of poor child nutrition and also assist in tailoring programme interventions (Aguayo & Menon, 2016).
Using nationally representative data from Nepal, this paper focuses on the time trends and predictors of four core WHO-UNICEF complementary feeding indicators that assess the timely introduction of complementary foods, the frequency of feeding, the diversity of foods, and the overall adequacy of diets among Nepali children aged  The women and children in all four datasets were included based on a two-stage stratified, nationally representative sample of households.
There are 75 administrative districts in Nepal, which are divided into village development committees and then wards. In most cases, the primary sampling unit (PSU) cluster for the surveys was the ward. In rural areas where the ward were small, neighbouring wards were combined into one cluster; whereas in urban areas when the ward size is large, wards were subdivided to form smaller clusters. Due to differences in population census data and PSU definitions between NDHS and NMICS, the number of PSUs selected by systematic sampling

| Complementary feeding practices
According to the WHO definitions (World Health Organization, 2010), we have defined the four complementary feeding indicators as follows: Introduction of solid, semi-solid, or soft foods (INTRO): The proportion of infants 6-8 months of age who received solid, semi-solid, or soft foods in the previous day or night.
Minimum meal frequency (MMF): The proportion of breastfed and non-breastfed children 6-23 months of age who received solid, semi-solid, or soft foods the minimum recommended number of times or more in the previous day or night. For breastfed children, MMF requires at least two solid/semi-solid feeds for children aged 6-8 months and at least three feeds for children aged 9-23 months.

Key messages
• Despite rapid improvement in health and development indicators, slow progress has been made in complementary feeding practices in the past 15 years in Nepal with average annual rate of increase ranged from 1% to 2% per year.
• Disparities in the risk of inappropriate complementary feeding practices are significantly evident at both individual (child age, maternal education, antenatal visits, and sociocultural group) and community-level (ecological zone and community-level access to health care).
• Contextual socio-economic progress that has driven improved access to education and health care services did not automatically address poor child feeding practices in Nepal over time.
For non-breastfed children, MMF is defined as at least four feeds of complementary food or milk between 6 and 23 months of age.

MMF is only defined for breastfed children in the NDHS 2006
and 2011 because frequency of milk feeds was not available for non-breastfed children in these two datasets.

Minimum dietary diversity (MDD):
The proportion of children 6-23 months of age who received foods from four or more food groups in the previous day or night. Seven food groups were defined as (a) grains, roots, and tubers; (b) legumes and nuts; (c) dairy products; (d) flesh foods; (e) eggs; (f) vitamin-A-rich fruits and vegetables; and (g) other fruits and vegetables. In NDHS 2001, the consumption of meat, poultry, fish, shellfish, and eggs were asked in one question, and therefore, MDD could not be created for year 2001. However, to compare food group intake patterns over time, we combined flesh foods and eggs in the other three datasets for analysis. In summary, data was available for analysis of time trends in INTRO and MMF in years 2001, 2006and for MDD and MAD, in year 2006. All eligible children were included for sample description. However, some feeding indicators were only available in breastfed children in certain years (MMF in 2001 and2006;MAD in 2006). To present findings among the same sample, only breastfed children (95-97% of all eligible children) were included for trend analysis.

| Risk factors
Risk factors were selected from individual, household, and community levels based on our conceptual framework (Stewart et al., 2013) and data availability. To analyse risk factors associated with the complementary feeding indicators, we did not include NDHS 2001 because of missing key risk factors (e.g., household wealth) and missing feeding indicators of MDD and MAD. Further, we restricted risk factor analysis in breastfed children, whose data was readily available in year 2006, 2011, and 2014 for all four feeding indicators. Table 1 lists the variables and their definitions used in the risk factor analysis. A brief introduction of variables at different levels is provided here.
At the individual level, the following characteristics describing the child's, mother's, and father's attributes were included: child age, sex, birth order, birth interval, perceived birth weight, vitamin A supplementation, vaccination, child diarrhoea, fever, and/or cough in the past 2 weeks; maternal age, body mass index, smoking status, utilization of reproductive health care, maternal education, exposure to media, women's empowerment (Kishor, 2005), sociocultural groups (Bennett, Dahal, & Govindasamy, 2008); and paternal age, education, and occupation.
At the household level, we considered sex of household head, number of household members, number of children under 5 years of age, types of cooking fuel, water, and sanitation characteristics, and household wealth index quintiles that were derived by NDHS or NMICS for relative wealth comparison in the sample using socio-economic indicators (Rutstein et al., 2004).
At the community level, we include place of residence and two geographic variables: development region (n = 5) delineated from east to west and ecological zone (n = 3; mountain, hill, and terai). We also calculated the following indicators using information of all survey participants within each PSU: (a) the proportion of women with primary education or higher; (b) mean women's empowerment score; (c) proportion with unimproved toilets; (d) proportion with shared toilets; and (e) general access to health services. This last indicator was a rank score based on 8-10 available variables that describe the utilization of maternal and child nutrition and health care services among all respondents in the cluster. A detailed description of indicator construction is available elsewhere (Na, Aguayo, Arimond, & Stewart, 2017).

| Statistical analysis
Using the provided sampling weights and defining strata by geographic region and place of residence, we adjusted for the complex sampling design in NDHS and NMICS to estimate proportions, means and medians that describe the distribution of sample characteristics and complementary feeding indicators at the population level. The Taylor series linearization method was used to estimate confidence intervals around prevalence estimates (Wolter, 2007).
To examine time trends in the entire sample, we have calculated the average annual rate of increase (AARI) for proportions that described the sample characteristics, complementary feeding indicators, and food group consumption. AARI was calculated to measure the geometric progression ratio, at which proportion changes annually over the period between the first and the latest observed year. Linear or logistic regressions adjusting for complex sampling design were used to test the significance of trends in continuous or binary variables over time. The non-parametric tests were performed to test the significance of trends in ordinal variables over year.
To study the time trend in complementary feeding indicators in subgroups of children, we calculated the weighted proportion and 95% confidence intervals of children meeting indicator criteria by selected sample characteristics. To test if rates of change in complementary feeding indicators differed in subgroups of the sample, an interaction term was created between the group and year and was added to the logistic regression model for each feeding indicator. The slopes representing the linearized rates were estimated, plotted, and compared against each other by contrasting the marginal effects. Delta methods were used to determine statistical significance (Cameron & Trivedi, 2005).
To identify factors associated with complementary feeding, we applied multilevel models to account for the multistage structure in NDHS and NMICS. We first applied intercept only models to understand the source of variance at individual/household (Level 1) and community (Level 2) levels. Before proceeding, we applied log likelihood ratio tests to compare if the single-level random intercept  Overall facility delivery Proportion of women within community who gave birth to their youngest child 0-5 years at health facilities Overall health professional delivery Proportion of women within community who have given birth to their youngest child 0-5 years assisted by health professionals Overall caesarean delivery Proportion of women within community who have given birth to their youngest child 0-5 years by caesarean delivery Overall utility of antenatal clinic visits Proportion of women within community who had at least four antenatal clinic visits prior to the birth of their youngest child 0-5 years Overall maternal iron supplementation Proportion of women within community given or who bought iron tablets during pregnancy of their youngest children 0-5 years Rank of access to health care The summed rank of all community-level indicators was created as the composite index of overall access to health care. The summed rank was categorized into quintiles.
a dummy year variable. To understand potential bias introduced in multilevel models, we examined the intra-class correlation in each community-level attribute in each year and in pooled data and compared with the recommended cut-off of ≥0.2 (Kravdal, 2006 3 | RESULTS the proportion steadily increased from 55% to 81% in the same period ( Figure S1, difference in slope p-value < .01, AARI = 3.0%).

| Time trends in sample characteristics
Considering child age ( Figure S2), the proportion of children meeting the MMF criteria increased most rapidly among children aged 6-11 months (from 49% to 82%, AARI = 4.0%) followed by children aged 12-17 months (from 72% to 85%, AARI = 1.3%). Both rates exceeded that of children aged 18-23 months, who stayed at a high proportion around 80-90% (all pairwise slope comparison p-values < .01). Children whose mothers were in the youngest age group (15-24 years old), showed significantly worse MDD and MAD trends over time ( Figure   S3); both proportions dropped from~30% to~20% (AARI = −4.8% for MDD; −5.6% for MAD, pairwise slope comparison against older groups: all p-value < 0.05). This stands in sharp contrast to the generally improving trends among children of older mothers.   Figure S4). The increase in the proportion of children who consumed dairy foods was smaller: from 43% to 47% in children aged 6-11 months (p-trend = 0.33), from 43%

| Time trends in dietary diversity and food group consumption
to 60% in children aged 12 to 17 months (p-trend < 0.01), and from 45% to 61% in children aged 18-23 months (p-trend < 0.001). The      indicators were also observed in Bangladesh (2004Hanif, 2013), which may indicate broader challenges in improving complementary feeding practices across the South Asia region.
Disparities in time trends for complementary feeding practices are evident in certain subgroups of children. While it is common to introduce solid foods at 6 months with a "rice ceremony" (Locks et al., 2015), there were notably different time trends in the introduction of complementary foods among boys and girls. There were much greater rates of improvement over time among boys than among girls. The difference is not likely due to a gender preference because the probability of meeting MMF, MDD, and MAD, for boys and girls has been similar over the past 15 years. Qualitative studies in the 1980s reported that mothers provided complementary foods sooner to boys than to girls in northwestern Nepal, which coincided with mothers' frequent concerns over insufficient breastmilk for boys (Miller, 1997). Given that the time trend has shifted, more emphasis should be placed on ensuring that girls are also receiving a timely introduction of complementary foods.
When examining time trends by child age, we see that children Second, cultural taboos relating to eggs and flesh foods are common, and children are usually fed such foods only after they have teeth, at about 1 year of age (Locks et al., 2015).  (Joshi, Agho, Dibley, Senarath, & Tiwari, 2012;Khanal, Sauer, & Zhao, 2013). Although there has been significant effort to reduce child marriage, leading to a substantial reduction since 1991 (Raj, McDougal, & Rusch, 2012), child marriage is still prevalent in Nepal, occurring in more than half of girls under 18 years old in 2011 (Raj et al., 2012). While a continued effort to reduce child marriage and early childbearing is paramount, directed efforts to empower young mothers with education and messages about appropriate child feeding practices are also necessary. Our study clearly indicates that programmes need to strengthen efforts to prevent early marriage and early childbearing while improving younger mothers' knowledge and practice with respect to complementary feeding.
Other than maternal and child age, our study points to a few additional consistent and independent factors that predict three or more complementary feeding indicators, including the number of antenatal care visits, maternal education, sociocultural group, ecological zone, and access to health care. Both antenatal care visits and maternal education have been previously identified as strong predictors of complementary feeding practices in Nepal Joshi et al., 2012;Pandey, Tiwari, Senarath, Agho, & Dibley, 2010). We also found that community-level access to health care was an important predictor of adequate complementary feeding. We have previously shown predictive power of a similarly composed indicator in Pakistan (Na et al., 2017). When mothers have better access to health services for themselves and their children in their communities, this may also reflect greater access to health and nutrition information.
Greater maternal education and better feeding knowledge alone are not likely to fully address poor complementary feeding practice, however (Chapagain, 2013). The Dalit and "other" category of sociocultural groups were less likely to meet the minimum complementary feeding criteria. The "other" category is a diverse group of smaller ethnic and religious castes. The fact that this group had consistently poorer practices might be a reflection on their minority status within communities and that they may have lower access to services and culturally tailored information. As complementary feeding practices are embedded in economic and sociocultural contexts, the challenge is to improve the quality of complementary foods and feeding practices by empowering and enabling caregivers to access and utilize healthier foods.
In this study, we identified a few community-level indicators that may be useful in identifying at-risk children. The Terai region shows poorer indicators of appropriate complementary feeding in comparison with other ecological regions Khanal et al., 2013;Pandey et al., 2010), despite the fact that this is a more agriculturally productive zone than the remote mountain and hill regions. We noted that including the indicator of community-level health care services essentially "mutes" all household factors in predicting feeding practices, including household wealth, which has frequently been cited as a significant predictor of feeding practices in South Asia . Although community-level socio-economic status was not assessed or examined, this finding has two implications: First, poorer households may cluster within villages, and therefore, targeting and intervening within communities may be more efficient and effective; second, community factors may outweigh household-level socio-economic factors in predicting complementary feeding practices.
There are certain limitations to the study. First, secular trends were built on survey data from two different sources. Although all were nationally representative, surveys were designed and conducted differently with respect to sample selection (e.g., sampling units were defined differently), questionnaire design (e.g., food items included were different), and administration (e.g., survey months, training, and data quality control done by NDHS and NMICS were different).
Second, analysis of factors was based on pooled cross-sectional data, and our results only imply associations rather than causal relationships. Third, the NDHS and NMICS surveys were not designed for analysing determinants of complementary feeding practices; therefore, the factors included in the analysis did not cover the all the key immediate, underlying, and basic risk factors. This lack of information may be even more problematic at the community level, for which we did not have detailed information about the dynamics of political, economic, agricultural, and sociocultural factors over time that deeply influence and shape feeding behaviours.
Despite these limitations, we have applied multiple nationally representative data over the past 15 years. Rigorous statistical methods were applied to appropriately address the multistage structure, collinearity, and bias in estimation in the multilevel data.
We conducted sensitivity analyses and found little difference in the results, which also lends confidence of the robustness of these research findings.
In conclusion, slow progress has been made in complementary feeding practices in the past 15 years, despite progress in numerous other health and development indicators in Nepal. Poor child dietary diversity is affecting 7 in 10 children under two currently and will continue to affect more than 50% of children in 2025 under the current trajectory of progress. Improvement in the access to education, health, and better sanitation did not necessarily translate into improvement in complementary feeding, particularly dietary diversity, which, in turn, did not and will not completely solve the child malnutrition problem. To promote child growth, our study calls for specific programming effort to address poor child dietary diversity in vulnerable and diverse populations in Nepal, with contextual understanding of the drivers of poor child diet and deliverable approaches to improve feeding behaviours.

ACKNOWLEDGMENTS
We appreciate the help from Shari S. Brown, Alessandra T. Mine and Srishti Tewari for the First Foods Project on literature review.