Reducing malnutrition in Cambodia. A modeling exercise to prioritize multisectoral interventions

Abstract Undernutrition is a major contributor to child morbidity and mortality and poses a large burden to the economy, in Cambodia. This study explored factors contributing to child stunting and wasting and their regional inequalities among 1,938 Cambodian children aged 6–23.9 months. Data were drawn from a longitudinal study (year 2017) conducted in six districts of two north‐eastern provinces and the capital and used as cross‐sectional. Socio‐demographic and household characteristics, children's feeding practices during the previous 24 hr, and children's length and weight measurements were collected. Gradient boosting models were used to calculate the contribution of determinants to child undernutrition whereas concentration index was used to assess the impact of those determinants on stunting and wasting inequalities among socioeconomic groups. It was found that low‐household wealth could predict 21% to 45% of child stunting and 23% to 36% of wasting across regions. After wealth, source and treatment of drinking water were found the second major predictor for stunting (15%) and wasting (21%). Combining child nutrition and household water, sanitation and hygiene indicators predicted around 30% of child undernutrition, either in the form of stunting or wasting. Mothers' education predicted >30% of stunting in the north‐eastern region. Results highlight that a complex interplay of factors contributes to child stunting and wasting. An integrated, intersectoral, equity‐focused approach that addresses children's dietary quality, household's water, sanitation and hygiene conditions, mother's education, and poverty is likely to yield the highest impact in achieving further gains in nutritional status among Cambodian children.

therapeutic treatment (approximately 94% of the children affected, in 2017) add to the burden of morbidity and mortality in the under-five population (Khara & Dolan, 2014). The economic burden of undernutrition among women and children in the country is estimated at USD 145-266 million annually; 45% of these economic losses are linked to stunting (low length-for-age), and approximately 3% are associated with wasting (low weight-for-length; Moench-Pfanner et al., 2016).
There is increasing recognition that the current nutrition-focused approach will not be sufficient to achieve global targets for improvements in maternal and child nutrition (International Food Policy Research Institute, 2016). Multisectoral engagement around common goals of improved child survival, growth, and development through integrated and complementary approaches is widely considered to yield greater efficiency of investment (Irwin, Siddiqi, & Hertzman, 2007;Samson, Fajth, & Francois, 2016). A sample size of 1,200 children under 3 years of age per site was calculated required to observe a reduction in child stunting from 32% to 26% over a 3-5-year period (with a precision of 3% and a dropout of 20%). A list of all children under 3 years of age living in the 169 villages included in the study was obtained from midwives and village health volunteers. This paper uses data of the households with children between 6 and 23.9 months (n = 1,938).
An electronic tablet-based questionnaire was used to collect information on (a) socioeconomic status, (b) health knowledge, (c) diet, including dietary diversity of children and women and breastfeeding practices, (d) WASH practices, and (e) mother and child anthropometry. Baseline data were collected in April-May 2016, followed by six consequent follow-ups that will run until March 2019. Data for children collected at follow-up visit 2 (October-November 2017) were used for the present study.

| Ethical approval
Ethical approval for the study was obtained from the Cambodia National Ethical Committee for Health Research. Informed consent was obtained from all participants, with consent obtained from parents or guardians for participating children.

| Anthropometric measurements
Gender, weight, length, and middle upper arm circumference (MUAC) were recorded for all children. Field workers measured length using UNICEF length boards with standing plates and moveable head boards accurate to 1 mm. We calculated length-for-age z scores and weight-for-length z scores using World Health Organization (WHO) 2006 standards for children 0-59 months (de Onis, 2006) using WHO Anthro software (version 3.2.2, January 2011). Field workers measured weight using calibrated digital balances (SECA) with 100-g precision.
The device used to measure the MUAC of children was a plastic, coloured, insertion tape (UNICEF supply division in Copenhagen) marked in millimetres, with cut-off points from red to yellow at 115 mm and from yellow to green at 125 mm, incapable of stretching and unresponsive to temperatures. Length-for-age z scores < −2 standard deviation (SD) was

Key messages
• Household wealth status was found to be the strongest independent predictor of child malnutrition (stunting and wasting). Interventions that target bottom wealth inequality may be most effective to reduce the 23% of stunting inequality found between socioeconomic groups.
• Poor water, sanitation, and hygiene (WASH) indicators and a low child feeding index score were found equally predictive for 30-40% of stunting and wasting prevalence. These findings suggest a need for strategic collaboration between the two sectors to further reduce child wasting and stunting, in Cambodia.
• Multisectoral engagement through integrated and complementary approaches is likely to yield greater efficiency to further improve nutritional status in Cambodian children.
defined as stunting whereas weight-for-length z scores < −2 SD and/or MUAC below 125 mm or presence of oedema was defined as wasting.

| Feeding practices
Infant and young child feeding practices were assessed per UNICEF and WHO indicators for appropriate breastfeeding and complementary feeding practices (WHO, 2008), including continued breastfeeding, bottle feeding, meal frequency, and dietary diversity in the previous 24 hr. Continued breastfeeding, as binary variable, denoted if the child was being breastfed until the age of 23.9 months.
Bottle feeding was recorded as a binary variable, denoting if the child received bottle feeding. Dietary diversity was based on consumption of foods from the following groups: (a) grains, roots, tubers, (b) legumes and nuts, (c) vitamin A fruits and vegetables, (d) other fruits and vegetables, (e) meats, (f) eggs, and (g) dairy products (WHO, 2008). Minimum meal frequency is defined as 2 for 6-8 months child breastfed, 3 for 9-23 months child breastfed, and 4 for nonbreastfed child 6-23 months (WHO, 2008). This indicator is intended as a proxy for energy intake from foods other than breast milk. An adapted child feeding index (Ruel & Menon, 2002) combining the four behaviours above (continued breastfeeding, use of bottle, dietary diversity, and meal frequency) was created for 6-8 months, 9-11 months, and 12-23 months by giving scores from 0 to 7 according to the practices claimed (Table S1). The child feeding index was then divided into three groups according to the score obtained: (1-3) low child feeding index score, (4-5) medium child feeding index score, and (6-7) high child feeding index score.

| WASH indicators
WASH variables included type of water used, type of sanitation, and treatment of the water. Although the treatment was a binary variable (adequate/not adequate), the other two were classified into improved or not according to the classification used during national surveys, such as the Cambodian Demographic Health Survey (National Institute of Statistics, 2015).

| Variables included in the analyses
The child's age in months was calculated by subtracting the date of the visit from the date of birth of the child; results were classified in three groups: (6-12 months), (12-18 months), and (18-23.9 months).
Gender was considered as a binary variable. Household wealth index, a composite measure of a household's living standard, was constructed using principal component analysis, as described by Filmer and Pritchett (Filmer & Prichett, 2001). The index includes materials used for housing construction, type of cooking fuel, and ownership of selected assets, such as radio, television, and refrigerator. WASH indicators (source of drinking water and type of sanitation facility) were not included (O'Donnell, Van Doorslaer, Wagstaff, & Lindelow, 2008). Once the index was created, household was divided in quintiles (poorest, poor, middle, richer, and richest). Mother's education was divided in three groups: no education, if the mother had attended no schooling; informal or primary school only (primary); and higher education if otherwise (secondary and above). Finally, access to national health campaign and birth certification were considered as binary variables using yes if the child had respectively received vitamin A supplementation 1 over the last year during the campaign and/or had a birth certificate.

| Statistical analysis
To rank the predictive power of different variables 2 on stunting and wasting, we used Boosted Regression Trees (BRTs) methods. BRT is a tree-based method and differs fundamentally from traditional regression techniques. In broad strokes, boosted regression trees combine the strength of two algorithms methods: regression trees (models that relate a response to their predictors by recursive binary splits) and boosting (a method that combines simple models to give improved predictive performance; leathwick, Elith, Chadderton, Rowe, & Hastie, 2006). This method has important advantages, as it can handle different types of predictor's variables. In addition, it can model complex nonlinear relationships, and its predictive performance is superior to most traditional modeling methods (Elith, Leathwick, & Hastie, 2008). Although BRTs are complex, the results can be summarized through relative influences of the predictor variables on the outcome that allows to rank the determinant included in the model per their predictive power. Using the formulae developed by Friedman (Friedman, 2001), the influence of each determinant is scaled so that the sum adds to 100, with higher numbers indicating stronger influence on the response. Then, the current paper explored the contribution of 10 variables to socioeconomic inequities on stunting and wasting in six districts of Cambodia. The inequality in stunting and wasting was calculated using the concentration index (referred below as CI; Wagstaff, van Doorslaer, & Watanabe, 2003). This approach of measuring inequality has been widely used and recognized as a standard tool (Emamian, Fateh, Gorgani, & Fotouhi, 2013;Kien et al., 2016;Rabbani, Khan, Yusuf, & Adams, 2016). The concentration index can be computed as twice the covariance of the health variable under study and a person's relative rank in terms of economic status, divided by the health variable mean.
The person's relative rank ri = i/N is a fractional rank of individual i in the living standards distribution, with i = 1 for the poorest and i = N for the richest. The value of this index ranges from −1 to +1: Its negative values imply that the variable is concentrated among disadvantaged people whereas its positive value implies the opposite. In the case of no inequality, the concentration index is equal zero.
All analyses were performed using STATA software version 13.1 (College Station, TX: StataCorp) and R software version 3.4.0.
General characteristics of the sample are represented in Table 1. The 1 Vitamin A is integrated with other health activities such as vaccination 2 Low wealth quintile, low mother education, no access to national health campaign, access to nonimproved sanitation facility, no adequate treatment of water, access to nonimproved source of drinking water, low child feeding index score, and no birth registration prevalences of stunting and wasting were significantly different between the selected provinces (p < 0.001): respectively 14.6% and 9.2% in Phnom Penh and 26.5% and 15.2% in the north-eastern districts. The type of sanitation was significantly different between Phnom Penh and the north-eastern region. Almost two third of the children in Phnom Penh were living in household with access to improved sanitation compared with less than one third in the northeastern study locations.
The top contributing predictors of wasting and stunting among children 6-23 months are listed in Tables 2 and 3, respectively. Lowwealth category was the main predictor of wasting in every district as it could predict three out of 10 wasted children. Combining WASH (sanitation, water access, and treatment of water) and child feeding indicators predicted about 32.9% of wasting in Phnom Penh and 41.4% in the north-east. For stunting, similar ranking in the contributing predictors are shown in Table 3. Among all factors considered, the vs. 16.8%). In the north-east, one out of three stunted children could be predicted by the low mother education factor. Table 4 presents results from the decomposition analysis for stunting. It shows how the various characteristics of respondents contributed to the inequality in child stunting in both north-eastern area and Phnom Penh. The different calculation steps can be illustrated through using "improved sanitation facility" for stunting in the north-east. The results of steps 1-3, which are the risky birth interval coefficient, its mean, and its concentration index are shown in Columns 2-4, respectively.
Step (4)  Due to a limited number of wasted children (11.9% of 574 children) in Phnom Penh, the decomposition of concentration was only performed for the north-eastern districts (Table 5). According to the contribution to CI, in the North East, mother with higher education, access to improved sanitation, improved source of drinking water, and/or appropriate treatment of water and high child feeding index score are all concentrated among people of higher economic status. As it was observed for stunting, children from poorer households had a higher probability of being wasted. The largest contributions to inequality in wasting in the north-east were attributable to household economic status (51.2%), access to improved sanitation (26%), and improved source of drinking water (20.5%). Furthermore, mother's education (14.2%) and access to national health campaign (7.2%) showed a considerable contribution to the measured inequality.
Child feeding index was neither an influencer of the inequality for stunting nor for wasting between socioeconomic groups. Access to national health campaign influenced to approximately 7% of the wasting inequalities between socioeconomic groups in the northeastern districts.

| DISCUSSION
This study presents key factors associated with wasting and stunting among Cambodian children aged 6-23 months. Also, it highlights the main contributors to inequalities of these two forms of child undernutrition between socioeconomic groups.
In our study, household wealth status was the strongest predictor of child stunting and wasting. The greater nutrition vulnerability of impoverished populations and the positive intergenerational impact of poverty reduction on nutrition are widely recognized (Aguayo, Badgaiyan, & Paintal, 2015;Kavosi et al., 2014;Pravana et al., 2017;Saxton et al., 2016;Tiwari, Ausman, & Agho, 2014;Torlesse, Cronin, Sebayang, & Nandy, 2016). Several studies suggest that socioeco- Poor WASH indicators and a low child feeding index score were predictors of 30-40% for stunting and wasting prevalence; these findings suggest a need for more integration between nutrition and WASH for reducing child undernutrition, in Cambodia. In 2013, using data from the Demographic Health Surveys from 2000 to 2010, household wealth and sanitation were found the main predictors of stunting (Ikeda et al., 2013). The importance given to nutritional indicators in this latter paper could be discussed as selection was restricted to early initiation of breastfeeding, without considering dietary quality indicators of complementary feeding for children aged 6-23 months. In a multidimensional poverty analysis that used data from Cambodia's 2014 Demographic Health Survey, the authors showed that the poorest children had the highest rates of nutrition deprivation and stunting nationally; however, a significant proportion of children in the highest wealth quintiles were also found to be nutritionally deprived as indicated by a significant prevalence of stunting (Karpati, de Neubourg, . This finding could explain why  A child feeding index (Ruel & Menon, 2002) combining these four behaviours was created for 6-8-month, 9-11-month, and 12-23-month age groups and then divided in three groups (see methodology section).
c No data available for nonimproved drinking water in Phnom Penh as >99% have access to an improved source.
the child feeding index, the composite measure of the appropriateness of IYCF practices, was not significantly associated with stunting and wasting inequality between socioeconomic groups in our study. foods (Ngure et al., 2014) and create a baby friendly environment. This may be a relevant approach for Cambodia where the presence of pigs near the main household dwelling was found to be a risk factor associated with Giardia duodenalis infection in young children and potentially stunting (Caron et al., 2018).
In our study, households in the north-eastern region had much lower access to safe sources of drinking water and improved sanitation facilities, compared with households in Phnom Penh. Increased latrine coverage is generally believed to be effective for reducing  (Ruel & Menon, 2002) combining these four behaviours was created for 6-8-month, 9-11-month, and 12-23-month age groups and then divided in three groups (see methodology section). exposure to faecal pathogens and preventing disease, but some studies believe that this outcome cannot be assumed (Clasen et al., 2014), and other factors may interfere. For example, in Chad, inappropriate cleaning processes of containers used for water transport in the household increased by two, the risk of being wasted (Marshak, Young, Bontrager, & Boyd, 2017). Ingestion of parasites through faecal-contaminated food or water is often associated with increased incidence of diarrhoea (WHO, 2001) and increased incidence of wasting. In our study, the use of a nonimproved water source was a strong contributor to child stunting (14.7%) and wasting (20.6%) inequality between socioeconomic groups in the north-eastern study locations. Addressing microbiology contamination of water at the point of use may be an important interim measure to improve safe drinking water. Recent analysis of water quality in the MyHealth project at the collection and consumption levels showed considerable levels of water quality deterioration at the household level with concerning high levels of Coliform and Escherichia coli at point of use. More than 80% of the water samples for use in children under 5 years of age had more than 10 colony forming unit (CFU) of coliform per 100 ml of water, and more than 30% had more than 10 CFU of E. coli per 100 ml of water at the point of use . Without ensuring water safe quality at collection and consumption, it will be challenging to ensure an impact on the reducing-related morbidity and linear growth. Protecting children from contaminated water, inappropriate hygiene practices, and impropriate sanitation through a spe-  Table 4 and 5, 94% of the 14.2% of mother's education contribution to wasting inequality between socioeconomic groups is concentrated in mothers who reach secondary education and above. Similar findings are observed for stunting with most of the inequality between socioeconomic groups observed in Phnom Penh and the north-east.
Those results highlight the potential positive impact on child wasting and stunting of allowing adolescent girls to reach and complete secondary school. In Indonesia and Bangladesh, a high level of maternal education was associated with protective caregiving behaviours (Semba et al., 2008). Therefore, a higher level of education for adolescent should help promote gender equality and empower those girls, which will ultimately might contribute to the reduction of child malnutrition (Cunningham, Ruel, Ferguson, & Uauy, 2015).
Our findings are representative of the six selected districts form Kratie, Ratanakiri (the north-east), and Phnom Penh only. To generalize these findings to Cambodia, a similar analysis could be done with data from the most recent national survey (2014 CDHS). Also, our model is considering a selection of factors (wealth quintile, mother education, access to national health campaign, sanitation facility and treatment of water, source of drinking water, child's age, gender, child feeding index, and birth registration) and the inclusion of additional indicators, such as maternal nutrition might further improve insights in the contribution of various factors on stunting and wasting.

| CONCLUSION
Our study confirms the complex interplay of sectorial factors associated with undernutrition. Findings allow to formulate many recommendations for policy reinforcement and intervention prioritization, in Cambodia. The potential for intersectorial gains presents the need for strategic approaches and synergies to be built through a process of collaboration and coordination across sectors at the national and subnational levels. Nutrition-specific and nutrition-sensitive interventions are both and together the best leverage for promoting positive nutritional status. A whole-child-focused strategy in Cambodia is required that brings together food, health, water, and every other sector required for a sustainable food and nutrition forthcoming.
The determinants of malnutrition cover a broad variety of biological, social, cultural, economic, and morbidity factors. The influence of these factors can be used to guide the development of strategies of intervention for reducing child malnutrition. More research is required to support a prioritization of basic and underlying determinants to be addressed to tackle poor nutritional status in Cambodian children most efficiently. An integrated, intersectoral, equity-focused approach that addresses children's dietary quality, household's water, sanitation and hygiene conditions, mother's education, and poverty is likely to yield the highest impact in achieving substantial nutrition gains among Cambodian children.

ACKNOWLEDGMENT
We would like to thank the Fisheries Administration of Cambodia and its team for the data collection.

CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest. The opinions and statements in this article are those of the authors and may not reflect official policies or opinions of the organisations they belong to.