Nutritional status and risk factors for stunting in preschool children in Bhutan

Abstract Childhood malnutrition remains endemic in South Asia, although the burden varies by country. We examined the anthropometric status and risk factors for malnutrition among children aged 0–59 months through the 2015 National Nutrition Survey in Bhutan. We assessed in 1,506 children nutritional status (by z‐scores of height‐for‐age [HAZ], weight‐for‐height [WHZ], and weight‐for‐age [WAZ]), estimating prevalence, adjusted for survey design, of stunting, wasting, underweight, and overweight (<−2 for HAZ, WHZ, and WAZ and >2 for WHZ). Children were also assessed for pedal oedema. We conducted multivariable linear/logistic regression analysis to identify child, maternal, and household risk factors for childhood undernutrition and overweight, excluding children with oedema (1.7%). Mean (SE) HAZ, WHZ, and WAZ were −0.82 (0.13), 0.10 (0.04), and −0.42 (0.05), respectively. Prevalence of stunting, wasting, underweight, and overweight were 21.2%, 2.6%, 7.4%, and 2.6%, respectively. In multivariable regressions, risk of stunting significantly increased by age: 5.3% at <6 months (reference), 16.8% at 6–23 months (OR = 3.06, 95% CI [0.63, 14.8]), and 25.0% at 24–59 months (OR = 5.07, [1.16, 22.2]). Risk of stunting also decreased in a dose–response manner with improved maternal education. None of the examined variables were significantly associated with wasting or overweight. Despite a WHZ distribution comparable with the World Health Organization reference (with ~2.6% vs. an expected 2.5% of children beyond 2 z in each tail), stunting persists in one fifth of preschool Bhutanese children, suggesting that other nutrient deficits or nonnutritional factors may be constraining linear growth for a substantial proportion of children.

Childhood stunting may result in short-and long-term adverse consequences such as increased childhood morbidity and mortality, impaired cognitive development, increased risk of obstetric complications and mortality in women of reproductive age, reduced productivity and earnings in adulthood, and intergenerational health and nutrition effects (Black et al., 2013). However, only a few national survey reports and studies using nationally representative data are available to provide information on maternal and child nutrition in Bhutan (Aguayo, Badgaiyan, & Paintal, 2015;Zangmo, de Onis, & Dorji, 2012).
Against this backdrop, our study provided in-depth information on the nutritional status of Bhutanese children using data from the National Nutrition Survey (NNS) 2015 (Nutrition Program, Department of Public Health, Ministry of Health, 2016). The objective of this study is threefold: first, we describe the nutritional status of children (i.e., z-scores, stunting, wasting and overweight) by geographic and socioeconomic characteristics. Secondly, we identify the child-, maternal-and household-level predictors of nutritional outcomes of interest. Third, we describe time trends of nutritional status across prior surveys from 1986 to 2015.

| Data sources
We used NNS 2015 as the primary source of data for this study. Further, we used NNS 1986-1988Directorate of Health Services, 1989), NNS 1999Namgyal & Yoezer, 1999), NNS 2008, and the Multiple Indicator Cluster Survey (MICS) 2010 to provide additional information about the time trends in childhood malnutrition in the country.

| Study design, sampling, and sample size
The study design, sampling methods, and main findings of the NNS 2015 are presented in detail elsewhere (Nutrition Program, Department of Public Health, Ministry of Health, 2016). Briefly, the country is divided geographically into western, central, and eastern regions and administratively into 20 Dzongkhags (districts). NNS respondents were selected with a national multistage sampling method. Two Dzongkhags were selected from each region with probability proportional to size sampling. Punakha and Lhuntse, estimated before the survey to be the highest and lowest performing Dzongkhags, respectively, were also included. Within each selected Dzongkhag, subareas were selected with probability proportional to size sampling and with a target urban/rural balance reflecting the population distribution in each region. In each area, 12 households were selected using systematic random sampling, for a total of 3,571 households surveyed. Our study used data, including anthropometric measures, for a total of 1,506 children aged 0 to 59 months identified in the surveyed households.

Key messages
• Despite a steady decrease in undernutrition since 1986, stunting persists among one-fifth of under-five children in Bhutan.
• The risk of preschool Bhutanese children being stunted increases with age, which may be partly mitigated by improved maternal education.
• Weight for height of Bhutanese children, however, tracks the WHO child growth reference, suggesting that factors other than energy deficit may be limiting normal linear growth achievement.
• Additional risk factors of stunting need to be understood and considered in the design of nutrition and other interventions intended to reduce stunting in the presence of apparently adequate weight for height status of children.

| Available variables
The NNS 2015 collected data using survey modules relevant to assessing household-, mother-and child-level variables, including children's anthropometry (weight and length at <2 years and weight and height at older ages), which was measured followed standard procedures (Nutrition Program, Department of Public Health, Ministry of Health, 2016). In addition, children were assessed for presence of pitting pedal oedema in both feet as a separate clinical indicator for severe acute malnutrition. Anthropometric z-scores and the prevalence of undernutrition were calculated from child length/height (n = 1,481) and weight (n = 1,490), excluding missing or values outside biologically reasonable ranges, based on the WHO reference values (WHO Multicentre Growth Reference Study Group, 2006).
The prevalence of stunting, wasting, underweight, or overweight was defined as the proportion of children whose height-for-age (HAZ), weight-for-height (WHZ), and weight-for-age (WAZ) scores were more than two standard deviations below the median of the population standard (or above the referent median of WHZ; for overweight; WHO Multicentre Growth Reference Study Group, 2006).
Factors associated with nutritional outcomes were considered at household, maternal and child levels based on UNICEF's conceptual framework of undernutrition (UNICEF, 1997). Household characteristics included household size, land ownership, livestock ownership, type of house wall materials, perceptions of household food insecurity, food consumption score (FCS), household water source, water storage and treatment, sanitation facilities and receipt of any benefits from government programs. Maternal characteristics included mother's age and education level. Child characteristics included sex and age. Characteristics at all levels were grouped into appropriate

| Statistical analysis
Stata version 14 (StataCorp LP, College Station, TX, USA) was used for all statistical analyses. The prevalence of undernutrition and distribution of z-scores were presented by region, rural/urban, household socioeconomic status, maternal education and child age and sex. Data analysis was conducted in two phases. First, univariate linear or logistic regression was conducted to identify variables that were associated with z-scores and degrees of stunting (<−1 HAZ, <−2 HAZ, and <−3 HAZ), wasting (<−2 WLZ), overweight (>2 WLZ) and underweight (<−2 WAZ). Variables associated with each outcome (P < 0.10) were included in multivariable regression models. We also adjusted for child age and sex, household wealth quintile and type of sanitation facility used (i.e., improved vs. not) as contextual covariates in multivariable models for stunting, although they were not significant in the univariate analysis (Stewart, Iannotti, Dewey, Michaelsen, & Onyango, 2013).
A variance inflation factor was checked for each univariate predictor to assess whether the predictors were highly correlated, and all tested variables reported variance inflation factors less than 3.0. Variables that kept significant association (P value < 0.05 or judged by 95% CI for odds ratios) in the multivariable regression were considered as associates of nutritional status indicator z-scores or event of undernutrition or overweight. The sampling design of the original survey was taken into account in all analyses by using "svy" commands to estimate summary statistics at the population level and to evaluate characteristics associated with nutritional status.

| Selected household, maternal and child characteristics
Out of 1,506 children aged 0 to 59 months, 51.8% were female and the mean age was 29.9 (SE: 1.0) months, 41.7% were from the western, 23.7% were from the central, and 34.6% were from the eastern regions (Table 1), in line with the population of each of these regions.
Most households (84.3%) had access to improved drinking water sources, stored water in a container (81.6%), and treated (sanitized) water (90.1%). Seven out of 10 households (68.3%) had improved sanitation facilities. The proportion of households reporting to be food insecure was low (2.3%). The majority of households (93.7%) showed acceptable FCS. Twenty-four percent of all households reported receiving benefits from any type of government programs. Half of mothers of surveyed children had either no education or informal education only (35.4% and 15.4%, respectively), and 35.6% had completed high school or higher education.   Sampling design of the original survey was taken into account in all analyses by using "svy" commands to estimate summary statistics at the population level.

| Clinical oedema
b Household asset score generated with principal component analysis as was done in the original analysis of the National Nutrition Survey data (Nutrition Program, Department of Public Health, Ministry of Health, 2016) but restricted to households with a child <5 years old who participated in the anthropometry assessment. The included variables were type of water source, shared sanitation facilities, toilet facilities, floor materials, roof materials, wall materials, cooking fuel, ownership of land, ownership of livestock, land of orchard, dry land, wet land, number of rooms, ownership of livestock, type of livestock (cattle/buffalo/yaks, pigs, horses, goats/ sheep, poultry), sofa set, electric iron, bukhari, rice cooker, curry cooker, refrigerator, modern stove, water boiler, microwave oven, bicycle, tractor, power tiller, jewellery, motorbike/scooter, sechu gho/kira, family car, other vehicle, washing machine, sewing machine, television, VCR/VCD/DVD, grinding machine, wrist watch and weaving tool.
c Bhutan-specific definition of household access to an improved water source is piped water into the household only. d In the NNS 2015 survey, mothers were asked if she adds anything to the water to make it safer to drink. e Food insecurity was one or more affirmative answers to the series of food security questions, represented as a composite variable of food insecurity that household experienced any out of four cases in the last month: (a) worry about not enough foods, (b) eat only rice/kharang/flour, (c) eat a smaller amount/skip meals at any meal time and (f) eat fewer meals in a day.
f Food consumption score (FCS) was based on household dietary diversity, food frequency and relative nutritional importance of different food groups in the past 7 days (World Food Programme, 2008).
The prevalence of stunting (<−2 HAZ) was 21.2%: higher in females (24.8%) than males (17.4%), increasing with age (from 5.3% at 0-5 months to 25.0% at 24-59 months), higher in the eastern (28.8%) than western (16.3%) or central (18.7%) regions, higher in rural (26.0%) than urban (16.2%) areas (Table 2, Figure 1, and Table S1) and decreasing with maternal education (from~28.1% and 28.4% among children of mothers with no or only informal education to 21.2%, 12.9% and 0.4% among children whose mothers completed primary, high school and college education, respectively; Table S2). The prevalence of stunting also decreased with household wealth, from 34.8% in the lowest to 5.7% in the highest quintile. The prevalence of children with <−1 HAZ and <−3 HAZ was 48.0% and 6.2%, respectively, with each level similarly distributed by region and area as seen with stunting as classically defined (<−2 HAZ; Table S1).

| Risk factors for wasting and overweight
None of the variables examined were significantly associated with continuous WHZ, wasting or overweight in multivariable models (data not shown).

| Risk factors for stunting
In univariate regression analysis, HAZ was positively associated with wealth, improved housing (i.e., cement walls), water treatment,  Rural versus urban residence was associated with a higher odds of <−3 HAZ in univariate analysis, and medium wealth intervals were associated with lower odds than the lowest group (P < 0.10; Table   S5). None of assessed variables were associated with risk of <−3 HAZ on multivariate adjustment.

| Risk factors for underweight
Similar to HAZ, in univariate analysis, indicators of socioeconomic status, improved housing, water treatment, and acceptable food consumption were positively associated, and child age was inversely associated, with WAZ (P < 0.10; Table S6). In multivariable models, however, only child age in months remained a significant risk factor, reflected by β = −0.50 (95% CI [−1.60, 0.61], P = 0.17) at 6-23 months and β = −1.23 (95% CI [1.36, 25.8], P = 0.04) at 24-59 months, relative to 0-5 months. With respect to underweight (<−2 WAZ), rural children were at greater risk than urban, and a dose-response association was evident by levels of formal education of mothers, although effect estimates were not statistically significant (Table S7).

| Time trend of nutritional status
The prevalence of stunting decreased markedly from 60.   Bhutan-specific definition of household access to an improved water source is piped water into the household only. c In the National Nutrition Survey 2015 survey, mothers were asked if she adds anything to the water to make it safer to drink. d Food insecurity was one or more affirmative answers to the series of food security questions, represented as a composite variable of food insecurity that household experienced any out of four cases in the last month: (a) worry about not enough foods, (b) eat only rice/kharang/flour, (c) eat a smaller amount/skip meals at any meal time, and (d) eat fewer meals in a day. e Food consumption score was based on household dietary diversity, food frequency, and relative nutritional importance of different food groups in the past 7 days (World Food Programme, 2008).  Household asset score generated with principal component analysis as was done in the original analysis of the National Nutrition Survey data (Nutrition Program, Department of Public Health, Ministry of Health, 2016) but restricted to households with a child <5 years old who participated in the anthropometry assessment. The included variables were type of water source, shared sanitation facilities, toilet facilities, floor materials, roof materials, wall materials, cooking fuel, ownership of land, ownership of livestock, land of orchard, dry land, wet land, number of rooms, ownership of livestock, type of livestock (cattle/buffalo/yaks, pigs, horses, goats/ sheep, poultry), sofa set, electric iron, bukhari, rice cooker, curry cooker, refrigerator, modern stove, water boiler, microwave oven, bicycle, tractor, power tiller, jewellery, motorbike/scooter, sechu gho/kira, family car, other vehicle, washing machine, sewing machine, television, VCR/VCD/DVD, grinding machine, wrist watch and weaving tool. b Bhutan-specific definition of household access to an improved water source is piped water into the household only. c In the National Nutrition Survey 2015 survey, mothers were asked if she adds anything to the water to make it safer to drink. d Food insecurity was one or more affirmative answers to the series of food security questions, represented as a composite variable of food insecurity that household experienced any out of four cases in the last month: (a) worry about not enough foods, (b) eat only rice/kharang/flour, (c) eat a smaller amount/skip meals at any meal time and (d) eat fewer meals in a day. e Food consumption score was based on household dietary diversity, food frequency, and relative nutritional importance of different food groups in the past 7 days (World Food Programme, 2008).  Bhutan-specific definition of household access to an improved water source is piped water into the household only. c In the National Nutrition Survey 2015 survey, mothers were asked if she adds anything to the water to make it safer to drink. d Food insecurity was one or more affirmative answers to the series of food security questions, represented as a composite variable of food insecurity that household experienced any out of four cases in the last month: (a) worry about not enough foods, (b) eat only rice/kharang/flour, (c) eat a smaller amount/skip meals at any meal time and (d) eat fewer meals in a day. e Food consumption score was based on household dietary diversity, food frequency and relative nutritional importance of different food groups in the past 7 days (World Food Programme, 2008).
a substantial proportion of children are stunted in the presence of a normal WHZ distribution suggests that a significant number of children with a seemingly "healthy" weight may nonetheless lack nutrients or face nonnutritional stresses that constrain linear growth.
Our study showed an expected higher risk of stunting among older than younger children, consistent with a commonly observed deceleration in linear growth of children through 2 years of age, followed by insufficient recovery thereafter in most impoverished settings (de Onis & Branca, 2016). Our age-stratified analysis, however, did not find any stratum-specific risk factors, likely due to small sample size leading to reduced study power to detect differences.
Although not significant, an inverse trend in stunting may have existed with maternal education, as seen elsewhere in the region (Aguayo, Badgaiyan, & Dzed, 2017;Dorsey et al., 2017), possibly related to factors other than feeding practices, which recently were not found to be associated with early childhood stunting in Bhutan (Campbell, Aguayo, et al., 2018).
Analyses of the MICS conducted in 2010 in Bhutan reported risk factors for early childhood undernutrition, focusing on children under 24 months of age (Aguayo et al., 2015(Aguayo et al., , 2017. Identified determinants of stunting included child age, residency in the eastern or western region (relative to Central), low socioeconomic status and suboptimal complementary feeding practices among children under one year of age (Aguayo et al., 2015;Zangmo et al., 2012), and determinants of wasting were child age, residency in the western region and inappropriate complementary feeding practices (Aguayo et al., 2017). Additional factors related to stunting to investigate should include environmental enteric dysfunction (Campbell, Schulze, et al., 2017), potential roles of micronutrient deficiencies (e.g., zinc) and other possible exposures such as aflatoxin, as recently reported to be endemic in neighbouring countries of Nepal and Bangladesh (Groopman et al., 2014).
Dzed and Wangmo have suggested through their ecological analysis that recent national efforts to modernize, increase trade, eradicate extreme poverty and implement expanded health and nutrition programs may be contributing to the reductions in childhood stunting seen in Bhutan (Dzed & Wangmo, 2016), similar to posited effects of poverty alleviation on reductions in stunting in Nepal and Vietnam (Headey & Hoddinott, 2015). For example, based on findings from the Nepal Demographic and Health Survey 2011, increased household wealth, improved food security, optimal breastfeeding practices and residence in the hill versus mountain agroecological zones were associated with a lower risk of stunting in preschool-aged children (Tiwari, Ausman, & Agho, 2014). In Vietnam, MICS 2000 and 2011 findings have identified rising maternal education as a critical explanatory factor for an observed reduction in stunting (Kien et al., 2016). Prevalence rates of wasting (2.6%) and underweight (7.4%) estimated in the current analysis are slightly lower than rates of 4.3% and 9.0%, respectively, reported earlier from the Bhutan NNS  Zangmo et al. (2012), NNS 1968/1988(Directorate of Health Services, 1989 and NNS 1999 (Namgyal & Yoezer, 1999). The prevalence of nutrition indicators in 2015 was estimated, excluding 26 oedematous children out of 1,506 children recruited in the NNS 2015 report observations were excluded in the association analysis to avoid sample size loss. For example, less data was available for antenatal and postnatal care, maternal hygienic and sanitation (hand washing practices and safe disposal of child stools) and child morbidity (n < 1,100), which could have been predictors of stunting.

Previous trend analyses for
In conclusion, a moderate level of stunting persists in preschoolaged children in Bhutan despite low risk of wasting, suggesting a need to further understand risk factors and potential longer term nutritional or nonnutritional interventions that may help lower the burden of depressed linear growth in the country.