Determinants of age-appropriate breastfeeding, dietary diversity, and consumption of animal source foods among Indonesian children.

Abstract Global child feeding practices remain suboptimal. In this study, we assess the determinants of age‐inappropriate breastfeeding, dietary diversity, and consumption of 3+ types of animal source foods (ASFs) using 11,687 observations from combined data from the Indonesian Demographic Health Survey of 2012 and 2017. We used linear and logistic regression after adjusting for the complex sampling design. Child's age and quality of antenatal care (ANC) were associated with all outcomes. Socio‐economic status and labour force participation were positively associated with higher dietary diversity score, ASF consumption, and age‐inappropriate breastfeeding. More ANC visits and having consultation at ANC were associated with more dietary diversity. Higher women's knowledge level was associated with more dietary diversity and consuming more ASF. Compared with western Indonesia, more children in eastern Indonesia were age‐inappropriately breastfed and had lower dietary diversity. The Indonesian government needs to develop programmes to improve child feeding particularly in eastern Indonesia, focusing on improving dietary diversity and ASF consumption in poorer households and on prolonging breastfeeding in richer households. Women's labour force participation should be encouraged, but programmes for working mothers are also needed to support continued breastfeeding and to express breast milk. ANC and postnatal programmes need improved consultation sessions for child feeding.


| INTRODUCTION
Appropriate child feeding practice from birth contributes to improvement in child survival, achieving growth and development potential, prevention of micronutrient deficiencies, future morbidity, and obesity in adulthood (Begin & Aguayo, 2017). The link between stunting and reduced cognitive abilities (de Onis & Branca, 2016), education opportunities, and up to 7% reduction in per capita income (Galasso, Wagstaff, Naudeau, & Sheekar, 2016) has generated a lot of research interests on the role of appropriate child feeding on reducing stunting. Some studies show negative associations between dietary diversity (Mahmudiono, Sumarmi, & Rosenkranz, 2017;Marriott, White, Hadden, Davies, & Wallingford, 2012;Rah et al., 2010) and stunting.
By feeding five or more food groups to under-five children would avert 13% of stunting in this age group (Krasevec, An, Kumapley, Begin, & Frongillo, 2017). Higher dietary diversity is associated with higher calcium adequacy (Muslimatun & Wiradnyani, 2016), which may link dietary diversity with child physical growth. Similarly, higher consumption of animal source food (ASF; Headey, Hirvonen, & Hoddinott, 2018;Krasevec et al., 2017) or higher household expenditure on ASF (Sari et al., 2010) is also associated with a lower rate of stunting. Moreover, ASF consumption is also positively associated with adequacy of protein and micronutrient intakes, especially Vitamin A, calcium, and zinc (Muslimatun & Wiradnyani, 2016), which are not only important for child growth but also for child development.
In addition to stunting, appropriate infant and young child feeding can reduce child mortality. Globally, scaling up exclusive and continued breastfeeding to up to 90% could prevent 823,000 deaths in under-five children annually (Victora et al., 2016). Not breastfeeding is associated with reduced cognition and economic losses of approximately $302 billion annually (Rollins et al., 2016). In Indonesia, the cost of not breastfeeding reached $1,343.70 million from cognitive losses alone (Rollins et al., 2016) and $118.62 million from treating diarrhoea and respiratory disease annually (Siregar, Pitriyan, & Walters, 2018), the highest in south-east Asia. A 10% point increase in exclusive breastfeeding and continued breastfeeding rates up to 1 or 2 years could reduce costs of treatment for child morbidity by $350 million in the United States, United Kingdom, China, and Brazil combined (Rollins et al., 2016). In south-east Asia, the economic benefits of breastfeeding through higher cognition and potentially higher earnings were estimated to be approximately $1.6 billion per year (Walters et al., 2016).
Globally, there were 151 million under-five children stunted in 2017 (United Nations, 2018). Less than 30% of children aged 6-23 months in the world meet the minimum dietary diversity, with only 42% of children in this age group consumed dairy products, 28% consumed ASF, and 17% consumed eggs (J. M. White, Begin, Kumapley, Murray, & Krasevec, 2017). In Indonesia, the prevalence of exclusive breastfeeding in the first 6 months is only 37.3%, and 47% of children  2015). However, Indonesia is currently focusing efforts to reduce stunting through the national strategy to accelerate stunting prevention, which, if fully implemented, will potentially reduce by two million the number of stunted children in this country (Rokx, Subandoro, & Gallagher, 2018). Thus, it is important to understand the determinants of appropriate child feeding in Indonesia to provide evidence to support stunting prevention strategies through provision of adequate nutrition.
Studies have shown that dietary diversity was positively associated with the child's age (Iqbal, Zakar, Zakar, & Fischer, 2017), maternal education, socio-economic status (Marinda, Genschick, Khayeka-Wandabwa, Kiwanuka-Lubinda, & Thilsted, 2018;Patel et al., 2012;Senarath et al., 2012;Solomon, Aderaw, & Tegegne, 2017), mother's working status (Solomon et al., 2017), women's empowerment (Ickes, Wu, Mandel, & Roberts, 2018), and the number of antenatal care (ANC) visits (Patel et al., 2012). There are reports of differences in consumption of ASF by residence and wealth status and child's age (J. M. White et al., 2017). A review also concluded that maternal autonomy might be an important factor in improving child feeding and nutritional status (Carlson, Kordas, & Murray-Kolb, 2015). In Indonesia, exclusive breastfeeding is reported to be associated with socioeconomic status and working parents (Titaley, Loh, Prasetyo, Ariawan, & Shankar, 2014). However, more studies from Indonesia are needed to guide programmes to improve child feeding practices in this country. This study, therefore, assesses a more complex set of determinants of age-inappropriate breastfeeding, dietary diversity, and consumption of ASF in children 0-23 months in Indonesia using nationally representative survey data.

Key messages
• Child's age and quality of ANC were associated with ageinappropriate breastfeeding, dietary diversity, and ASF consumption.
• Socio-economic status and labour force participation were positively associated with age-inappropriate breastfeeding, dietary diversity, and ASF consumption.
• Government programs are needed to improve child feeding in eastern Indonesia and some western provinces, to improve dietary diversity and ASF in poorer households, and to prolong breastfeeding duration in richer households.
• National programs that support continued breastfeeding and expressing breast milk for working mothers are required.
• To improve the quality of ANC programs, consultation sessions to discuss appropriate child feeding practices need strengthening.

| Data
Indonesian Demographic Health Surveys (IDHS) are conducted every 5 years. We used data from the two most recent surveys (2012 and 2017), which we combined for our analyses.
IDHS employs a complex sampling design with stratification by regions and urban/rural areas before sampling the households. The IDHS country reports (available at www.dhsprogram.com) describe in detail the sampling methods used in the surveys. Institutionalised review board in ICF, as the DHS survey implementing agency, and institutionalised review board in host countries approved the survey protocols. All participants provided informed consent before data collection. We obtained the data in this study from the DHS website (www.dhsprogram.com) and the Indonesian Coordinating Board for National Family Planning website (https://sdki.bkkbn.go.id). This study is a secondary analysis of de-identified data, and thus, no further consent was needed.
All women aged 15-49 years in the selected households were eligible for interview by the trained enumerators. In the surveys, women with under-five children were interviewed to collect data about their children. We analysed dietary data of the youngest child aged 0-23 months born in the 5 years preceding the survey, who was alive and lived with their mother.

| Key variables and measurements
Infant feeding practices were assessed using three indicators: ageinappropriate breastfeeding, dietary diversity, and number of types of ASF consumed with the following definitions: 3. Number of types of ASF consumed from a possible six groups from the following list: (a) breast milk, fresh milk, powdered milk, and baby formula; (b) other processed milk product such as yogurt and cheese; (c) eggs; (d) meats (beef, pork, lamb, and chicken); (e) organs (liver, heart, and other organs); and (f) fish or shellfish. On the basis of the distribution of the variety of ASF types by the number of types consumed, a child is considered to have an adequate number of ASF if fed three or more types of ASF in the 24 hr before the interview. This study analysed the odds of consuming 3+ types of ASF.
We tested a list of determinants previously reported to be associated with age-inappropriate breastfeeding, dietary diversity, and consumption of 3+ types of ASF (analytical framework is available on Figure S1): 1. Child factors: gender, child's age (0-5 months, 6-8 months, 9-11 months, 12-17 months, and 18-23 months), and birth order (first child, second child, and third or later child).
2. Demographic factors: mother's age (15-19 years, 20-14 years, 25-29 years, 30-34 years, 35-39 years, and 40+ years), age difference between mother and father (older than father, 0-4 years, 5-7 years, and >7 years), father's education (did not complete primary education or less, completed primary or some secondary school, and completed secondary school or higher), father's occupation (agricultural and nonagricultural), residence (urban or rural), and region (west or east Indonesia). West Indonesia covered all provinces in and around Sumatera and Java island, and remaining provinces are in East Indonesia. The Indonesian government usually includes Kalimantan as a part of west Indonesia. However, because two of four provinces in Kalimantan were among the worst half in age-inappropriate breastfeeding and dietary diversity, we included Kalimantan as east Indonesia in our analysis. We reanalysed the final models using Kalimantan as west Indonesia to assess the difference in the estimates.
3. Household factors: wealth quintiles (lowest, second, third, fourth, and highest), number of means of transport owned (did not own any, owned one, and owned ≥2 means of transport), size of agricultural land owned (did not own land, owned <3 ha, 3-9 ha, and 10+ ha), number of children under five (<2 or ≥2). We calculated the wealth quintiles by combining the two surveys and using principal component analysis of household ownership of an inventory of assets and facilities, which included electricity, radio, television, telephone, mobile phones, refrigerator, bank account, number of members of household per sleeping room, source of water, type of toilets, materials for floor and roof, and type of cooking fuel.
We included the number of means of transport owned as a proxy indicator for access to food markets and ownership of agricultural land as a proxy for food availability.
4. Women's access to and contact with health care providers, including the number of ANC visits attended, the quality of the ANC, ANC consultation, and the provision of information about the source of care for pregnancy complications. For ageinappropriate breastfeeding, we also evaluated assisted delivery as a determinant. We calculated an indicator for ANC quality by

| Statistical analysis
To provide a programmatic evaluation of the child feeding status and progress over time, we graphed the percentage of age-inappropriate breastfeeding, the mean dietary diversity, and the percentage of consumption of 3+ types of ASF by province, overall, and by the year of the survey. We used logistic regression to determine associations between the covariates and age-appropriate breastfeeding and consumption of ASF. The dietary diversity score was normally distributed, and thus, the association between the covariates and dietary diversity was assessed using multiple linear regression. Covariates univariably associated with a P value <.25 were included in the initial multivariable regression models. Final models were derived using backward elimination of variables, and we only retained those variables with P value <.01 in the final model. We used a lower significance level due to the large sample size, and our intention only to detect factors that were strongly associated with the outcome. Survey year and region were kept in the model regardless of their significance. We replaced region with province in the model to show details of the association by province. In all our analyses, we categorised North Kalimantan, a new province in IDHS 2017, with East Kalimantan, as was the case in IDHS 2012.
All analyses were conducted using STATA 15 using the svy commands to adjust for the complex sampling design.

| RESULTS
There were 11,687 children aged 0-23 months in the sample. Most children were aged 12-17 months (26%) and most had a mother who had a husband with a nonagricultural job (65%), half lived in rural areas, 68% belonged to households that did not own agricultural land, and 67% lived at home with <2 under-five children (  Figure S2). Bangka Belitung was also the most improved province with a 20% point increase in dietary diversity score between surveys, and Papua was the worst performing province with a 17% point reduction in dietary diversity between surveys ( Figure S3). Bali was the best performing province with 76% increase in child consumption of ASF, whereas Papua was again the worst performing with 51% reduction of ASF from 2012 to 2017 ( Figure S4).

| Determinants of age-inappropriate breastfeeding
After adjusting for other covariates, there was an association between age-inappropriate breastfeeding and the age of the child. As seen in Table 2, children aged 0-5 months had a 70% greater odds of being age-inappropriately breastfed, whereas children 6-17 months had approximately 50% lower odds of being age-inappropriately breastfed compared with children aged 18-23 months. Children with siblings had approximately 30% lower odds of being age-inappropriately breastfed compared with the first child.
Children living in rural areas had 15% lower odds of being ageinappropriately breastfed compared with children living in urban areas.
Children in the east of Indonesia had 26% higher odds of being ageinappropriately breastfed compared with those in the west of Indonesia (  Figure S5).
There was an association between socio-economic status and age-inappropriate breastfeeding. Interestingly, compared with the lowest quintile, children from households in the higher quintiles, that is, the third, fourth, and highest quintiles had a 20%, 34%, and 63%    Women's empowerment factors were calculated by combining the data of the two surveys and using principal component analysis of 17 indicators following a previous method (Sebayang et al., 2019). Six variables indicated labour force participation: work in the last 12 months, for whom the woman worked, women's occupation type, types of payment, worked all year, and earned more than the husband. Women's disagreement with justification for wife beating was assessed with five reasons: neglecting children, going out without husband's permission, arguing with husband, refusing sex, and burning food. Decision-making power was derived from household decision makers for women's own health care, household purchases, visiting family, and husband's earnings. The women's knowledge component included formal educational level and access to media (newspaper, radio, and television).  Children of mothers who received more types of services during ANC in her last pregnancy had lower odds of being ageinappropriately breastfed.
In terms of women's empowerment, only mother's labour force participation was associated with age-inappropriate breastfeeding.
Children of mothers who had a high level of labour force participation had 62% higher odds of being age-inappropriately breastfed (adjusted OR = 1.62, 95% CI [1.41, 1.87]; Table 2).

| Determinants of dietary diversity
After controlling for child, demographic, and household factors, we found factors associated with an increased score of dietary diversity were the child's age, region, and wealth quintiles ( in Yogyakarta ( Figure S5).
All health care factors were positively associated with the dietary diversity score. One ANC visit was associated with 0.03 more points of dietary diversity score, 4-6 types of ANC services were associated with 0.24 more points of dietary diversity score, and 7-8 types of ANC services were associated with 0.49 more points of dietary diversity score. Children of mothers who had both consultations during ANC and were told about care for complication had 0.22 more points of dietary diversity score compared with those who had neither one.
Women's empowerment, as indicated by the mother's labour force participation and her knowledge level, were significant positive determinants of dietary diversity. Children of mothers with high labour force participation had 0.24 points higher dietary diversity score, whereas children of mothers with medium knowledge had 0.12 more points and those of mothers with high knowledge had 0.26 points higher dietary diversity score (Table 3).

| Determinants of ASF
The child's age was a significant factor for adequate intake of ASF. The older the child, the greater the odds of consuming more types of ASF.
Compared with children aged 6-8 months, children aged 9-11 months had more than three times greater odds of consuming 3+ types of ASF, children 12-17 months had nearly six times greater odds, and children 18-23 months had more than eight times greater odds of consuming 3+ types of ASF (Table 4).  Figure S5).
The significant household determinant was socio-economic status.
Children from the higher quintiles households had greater odds of   Children of women with high labour force participation had 33% higher odds of consuming 3+ types of ASF compared with children of women with low labour force participation, and children of women with a high knowledge level had 43% higher odds of consuming 3+ types of ASF compared with children of women with low knowledge level (Table 4).

| General findings
In this study, age-inappropriate breastfeeding, dietary diversity, and consumption of ASF were associated with the child's age, household socio-economic status, quality of ANC, and labour force participation of mothers. Children aged 0-5 months and those whose mother was of higher socio-economic status and higher labour force participation, had greater odds of age-inappropriate breastfeeding. Higher birth order of the child and living in a rural area were associated with lower age-inappropriate breastfeeding. We did not find any association between having means of transport as a proxy of access to markets with any of the outcomes. Land ownership as a proxy for food availability was only associated with lower ASF consumption. A larger number of ANC visits and whether women received both consultation and information about care for complications were associated with more dietary diversity, and better quality of ANC was associated with both more dietary diversity and variety of consumption of 3+ types of ASF in children. For women's empowerment, apart from women's labour force participation that was associated with all outcomes, high women's knowledge level was also associated with more dietary diversity and greater odds of consuming 3+ types of ASF among children.

| Comparison with other studies
Similar to our findings for age-inappropriate breastfeeding, a study using combined 2002/2003 and 2007 IDHS reported that being richer and having both parents working were associated with lower odds of exclusive breastfeeding (Titaley et al., 2014). In contrast, in Ethiopia, mother's education level and the number of ANC visits were determinants of exclusive breastfeeding (Tariku et al., 2017), although we did not find such an association for age-inappropriate breastfeeding. In agreement with our results, a study in urban Indonesia found a positive association between minimum dietary diversity and socioeconomic status (Santika, Februhartanty, & Ariawan, 2016 women's educational indicators (mother's education and interaction with media), but urban/rural residence was not a factor for dietary diversity in our study. In India and East Africa, child's age, mother's education, household wealth, and the number of ANC visits were positively associated with dietary diversity (Gewa & Leslie, 2015;Patel et al., 2012). Similarly, a study in Ethiopia also found that the mother's education, occupation, and household income were positively associated with minimum dietary diversity (Solomon et al., 2017).

| Possible mechanisms
The highest odds of age-inappropriate breastfeeding were at 0-5 months, indicating that many children were not exclusively breastfed.  Women's empowerment factors were calculated by combining the data of the two surveys and using principal component analysis of 17 indicators following a previous method (Sebayang et al., 2019). Six variables indicated labour force participation: work in the last 12 months, for whom the woman worked, women's occupation type, types of payment, worked all year, and earned more than the husband. Women's disagreement with justification for wife beating was assessed with five reasons: neglecting children, going out without husband's permission, arguing with husband, refusing sex, and burning food. Decision-making power was derived from household decision makers for women's own health care, household purchases, visiting family, and husband's earnings. The women's knowledge component included formal educational level and access to media (newspaper, radio, and television).  in ASF consumption between ages 6-8 and 12-17 months indicate that many mothers may have waited until 18 months to introduce more types of ASF. The delay may be influenced by limited income and livestock for foods (Wong et al., 2018) that may result in the family's choice of prioritising priced foods only to productive adults but could also be influenced by the lack of caregiver's knowledge on the proper age to introduce more varieties of ASF other than milk to children. The pathway of association between socio-economic status and our outcomes may be due to increased means to purchase baby formula in higher socio-economic status families. Mothers in richer households, thus, weaned babies earlier and gave children more diversity of animal and plant source complementary food, whereas mothers in poorer households would delay feeding their babies more ASF, either because of costs or lack of knowledge. A study in urban Indonesia showed a similar trend in which children in lower socio-economic households had higher odds of continued breastfeeding at 1 year and lower odds of early introduction to solid, semisolid, and soft foods at age 6 months (Santika et al., 2016). Similarly, mothers with high labour force participation may have greater means to purchase baby formula and less time to breastfeed or express their milk for later use, hence, wean their babies earlier (Barati et al., 2018). Mothers seemed to be better at age-appropriate breastfeeding when babies were the second or later child, possibly due to limited funds to purchase formula for the last child as the total cost to raise children increases with each additional child. Agricultural families who only have a small size of land may not have enough animal products for consumption and thus prioritise selling the ASF rather than their family consuming it (Wong et al., 2018 (Beatty, Ingwersen, & Null, 2017). This finding indicates the need to revamp the child feeding consultation programme in Indonesia. Also, a study in Indonesia has reported that mothers who received free samples of infant formula in their delivery clinics mentioned that they continued giving formula to their children afterwards (Barati et al., 2018).
This observation raises an important issue of the need to include marketing ethics in the measurement of health care quality.

| Strengths and limitations
In this study, we combined two nationally representative data sets and, thus, had a large sample size to measure the determinants of child feeding in Indonesia. We used a stricter significance level to determine an association in order to reduce the possibility of associations by chance due to the large sample size. However, this study was crosssectional in design, and thus, we cannot infer causal relationships.
The data sets did not include questions about food security and beliefs around child feeding, and thus, we could not assess the possibility that that these factors predict child feeding practices. We also did not have a better measure of the mother's participation at postnatal care, which would have been a better proxy for her contact with health service.

| Policy implications and future research
Our results show that the Indonesian government needs to focus on programmes to improve child feeding in all eastern parts of Indonesia Women's empowerment factors were calculated by combining the data of the two surveys and using principal component analysis of 17 indicators following a previous method (Sebayang et al., 2019). Six variables indicated labour force participation: work in the last 12 months, for whom the woman worked, women's occupation type, types of payment, worked all year, and earned more than the husband. Women's disagreement with justification for wife beating was assessed with five reasons: neglecting children, going out without husband's permission, arguing with husband, refusing sex, and burning food. Decision-making power was derived from household decision makers for women's own health care, household purchases, visiting family, and husband's earnings. The women's knowledge component included formal educational level and access to media (newspaper, radio, and television).
in the worst half of the provinces. The provinces include Aceh, North Sumatera, and Riau for dietary diversity; Riau Islands, Bangka Belitung, Jakarta, Riau, Banten, and East Java for age-inappropriate breastfeeding; and Bengkulu, North Sumatera, Central Java, West Java, and Aceh for the consumption of ASFs. Additionally, programmes need to focus on improving dietary diversity and consumption of ASFs for poorer households and on prolonging breastfeeding for richer households. Mother's participation in the labour force may improve dietary diversity and consumption of ASFs, and a previous study also shows that it improves ANC visits (Sebayang et al., 2019), which may eventually increase women's access to sound child feeding information and support. Women's labour force participation thus needs to be encouraged, but better national programmes are needed to support prolonged breastfeeding and to provide facilities for expressing breast milk for working mothers. Some studies have shown an impact of maternity leave and workplace support programmes on improving exclusive breastfeeding, but studies on the impact on continued breastfeeding are lacking (Rollins et al., 2016). In Indonesia, the quality of consultation in ANC and postnatal care programmes needs improvement and should cover both breastfeeding and complementary feeding practices, food diversity, and the economic benefits of appropriate child feeding. Antenatal counselling programmes are reported to increase exclusive breastfeeding in children under 6 months by 66% and continued breastfeeding in children aged 12-23 months by 15% (Rollins et al., 2016). A recent review of seven studies in south Asia showed that information, education, and counselling interventions improved dietary diversity (Aguayo, 2017). In Indonesia, providing community activation through women's groups, in addition to general nutrition information, improved exclusive breastfeeding in the first 3 months of life by 16% (S. White et al., 2016). Programmes to enhance the delivery of nutrition information through women's groups need promoting in Indonesia.

ACKNOWLEDGMENTS
The authors wish to thank ICF for providing the 2012 IDHS data, and the Indonesian Coordinating Board for National Family Planning

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.

CONTRIBUTIONS
SKS conceptualised the study, conducted the data analysis, and drafted and revised the manuscript. MJD conceptualised the study, contributed to the data analysis, and revised the manuscript. E A contributed to the data analysis and revised the manuscript. FE contributed to conceptualising the study and reviewed the manuscript.
PJK contributed to the data analysis and reviewed the manuscript.
ML contributed to conceptualising the study and reviewed the manuscript.

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article.