Predictors of complementary feeding practices in Afghanistan: Analysis of the 2015 Demographic and Health Survey

Abstract Despite improvements over the past 20 years, high burdens of child mortality and undernutrition still coexist in Afghanistan. Global evidence indicates that complementary feeding (CF) practices predict child survival and nutritional status. Our study aims to describe CF practices in Afghanistan and to discern underlying predictors of CF by analysing data from Afghanistan's 2015 Demographic and Healthy Survey. Multilevel models were constructed comprising potential predictors at individual, household, and community levels and four CF indicators: timely introduction of solid, semi‐solid, or soft foods (INTRO), minimum meal frequency (MMF), minimum dietary diversity (MDD), and minimum acceptable diet (MAD) among breastfed children. INTRO prevalence among children aged 6–8 months was 56%, whereas the prevalence of MMF, MDD, and MAD among children aged 6–23 months was 55%, 23%, and 18%, respectively. Of the seven food groups considered, four were consumed by 20% or fewer children: eggs (20%), legumes and nuts (18%), fruits and vegetables (15%), and flesh foods (14%). Increasing child age and more antenatal care visits were significantly and positively associated with greater odds of meeting all CF indicators. Lower household wealth and lower community‐level access to health care services were associated with lower odds of MDD and MAD. Disparities in achieving recommended CF practices were observed by region. CF practices in Afghanistan are poor and significant socioeconomic inequities in CF are observed across the country. Our study calls for urgent policy and programme attention to improve complementary feeding practices as an intrinsic part of the national development agenda.

and whose nutrition, growth and development determine the development and prosperity opportunities of the country.
Despite conflict and poverty, under-five mortality in Afghanistan declined by almost 30% between 2000 and 2015, from 120 to 81 child deaths per 1,000 livebirths (United Nations, 2017). Similarly, stunting among children aged 0-59 months declined by 30%, from 59% in 2004(World Bank, 2017  Suboptimal feeding contributes to infant and child mortality (Jones et al., 2003). Its effect is mediated by poor nutritional status (Black et al., 2013;Stewart, Iannotti, Dewey, Michaelsen, & Onyango, 2013) and/or infectious diseases, such as pneumonia, diarrhoea, and others, which are the leading causes of under-five mortality in Afghanistan (Akseer et al., 2016). The latest data indicate that only 43% of Afghan infants aged 0-5 months are exclusively breastfed (CSO, Ministry of Public Health, & ICF International, 2017). Complementary feeding practices are even poorer, as only 16% of children aged 6-23 months are fed diets that meet the minimum adequacy in terms of feeding frequency and diet diversity (CSO et al., 2017).  (Varkey, Higgins-Steele, Mashal, Hamid, & Bhutta, 2015), our study aims to discern the immediate and underlying predictors of complementary feeding of children aged 6-23 months in Afghanistan.
We analysed data collected by the 2015 Afghanistan Demographic and Health Survey (AfDHS; CSO et al., 2017), which provided up to date information on child feeding practices from a large nationally representative sample of infants and young children aged 0-23 months.
Further, our study aims to identify priority areas of focus for future policies, strategies programs aiming to improve the quality of complementary foods and feeding practices among Afghan children.

| Data source
The 2015 AfDHS employed a two-stage sampling design: At the first stage, a total of 950 clusters (260 urban and 690 rural) were selected from 25,974 census enumeration areas. Given the existence of inaccessible clusters in each of the 34 provinces due to insecurity, 101 reserve clusters were preselected to replace inaccessible clusters, resulting in a total of 1,051 clusters selected at the first stage. At the second stage, 27 households were selected in each cluster through an equal probability systematic sampling process. Excluding clusters that were identified as insecure (n = 75), the AfDHS was carried out in 976 of 1,051 clusters and data collection took place from June 15, 2015, to February 23, 2016. The AfDHS could not provide estimates for Zabul Province, where data were collected in only seven urban clusters due to fieldwork challenges.
The data analysed in this study includes information from (a) the household questionnaire, which collected demographic and socioeconomic information; and (b) the women's questionnaire, which collected information from ever-married women aged 15-49 years, including infant and young child feeding (IYCF) practices in children born in the previous 3 years and living with the respondent at the time of the survey. The response rates to the household questionnaire and the women's questionnaire were 97.8% and 96.8%, respectively.
For the purpose of our analytic sample, we defined eligible children as the youngest singleton child aged 6-23 months old to avoid potential recall bias, and to avoid including children from the same households. Given these inclusion criteria, our descriptive results are slightly different from what has been reported in the 2015 AfDHS final report (CSO et al., 2017).

| Complementary feeding indicators
The World Health Organization defines four CF indicators at the population level: introduction of solid, semi-solid, or soft foods  Table 1 and are described briefly below as follows: INTRO: The proportion of infants 6-8 months of age who received solid, semi-solid, or soft foods in the previous 24 hr (day or night).

MMF:
The proportion of children aged 6-23 months who received solid, semi-solid, or soft foods the minimum number of times or more in the previous 24 hr (day or night). For still breastfed children, the

Key messages
• Slightly more than half of Afghan children received complementary foods at 6-8 months of age. Only about half of children aged 6-23 months old were fed with the minimum required frequency; less than a quarter were fed a minimum diverse diet; and less than one fifth were fed a minimum acceptable diet.
• Risk factors for poor complementary feeding practices included younger age, poor access to health care services, household poverty, and residence in the Central Highland region.
MMF is defined as two times for infants aged 6-8 months or three times for children aged 9-23 months. For nonbreastfed children aged 6-23 months, the MMF is four times, including milk feeds.

MDD:
The proportion of children aged 6-23 months who received foods from four or more food groups in the previous 24 hr (day or night). Seven standard food groups were defined: (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 addition to MDD, in our study, we also examined the consumption of individual food groups, and we constructed a child dietary diversity score by summing the total number of food groups consumed in the previous 24 hr (day or night); this variable could range from zero (no food group eaten) to seven (all food groups eaten).

MAD:
The proportion of children aged 6-23 months who received the MDD and the MMF in the previous 24 hr (day or night). For breastfed children, MAD is achieved if the child meets both the MMF and MDD criteria. For nonbreastfed children, the child is required to receive at least four food groups excluding dairy products, two milk feeds, and MMF.
Because there were missing data for milk feed frequency among nonbreastfed children (n = 1,509 out of 1,571 nonbreastfed eligible children), MMF and MAD were calculated only for breastfed children.
Given that the majority (94%) of children aged 6-8 months were still breastfed, the INTRO indicator largely represents breastfed children.
For the ease and consistency of results reporting, we have limited our predictor analysis to breastfed children only for all CF indicators.

| Predictor variables
In line with our previous work (Na, Aguayo, Arimond, Narayan, & Stewart, 2018;Na, Aguayo, Arimond, Dahal et al., 2018) We first selected characteristics of the child, the parents, and the household that represented proximal factors that can potentially predict complementary feeding practices. For children, we included the following variables: sex, age, birth order, birth interval, perceived birth weight, recent vitamin A and iron supplementation status, vaccination record, and recent symptoms of diarrhoea, fever and cough. For mothers, we included the following variables: age, smoking status, and use of reproductive health services: delivery at health facility, delivery with skilled birth assistance, caesarean delivery, number of antenatal care visits, timing of postnatal check-up for mother and for child, education level, occupation, exposure to media (newspaper, radio, and TV), and empowerment status (involvement in decisionmaking for large household purchases, freedom to visit family and friends, own health care, and attitude towards domestic violence regarding five scenarios). Using available data, we calculated a composite women's empowerment score for each mother using an established methodology (Jennings et al., 2014;Na, Jennings, Talegawkar, & Ahmed, 2015). For fathers, we included age, education level, and occupation. Household characteristics included in the analysis were: Sex of household head, household size, number of children under five, type of cooking fuel, water source, time to get to water source, toilet conditions and sharing, and household wealth quintile, which were composed by AfDHS according to a standard principal components analysis method (Rutstein & Johnson, 2004).
Using information from all survey participants, we also included a second-level of potential predictors at the community level that represented contextual factors influencing child feeding and care. The unit of community in this analysis was the cluster. These factors included place of residence (rural or urban), geographical region, prevalence of women who completed primary or higher education, average women's empowerment score, and prevalence of unimproved toilets and shared toilets. In addition, using a previously developed scoring scheme (Na, Aguayo, Arimond, & Stewart, 2017), we created a composite indicator to rank general community-level access to health care based on The complementary feeding indictor would be coded as one if the listed criteria were met by a child per the child's age and feeding mode using child's dietary information in the previous day or night. The table is adapted from (Na et al., 2015) b Food group score is calculated based on consumption of seven food groups, grains, roots and tubers, legumes and nuts, dairy products, flesh foods, eggs, vitamin A-rich fruits and vegetables, and other fruits and vegetables. c Milk feeds are consumption of infant formula, milk such as tinned, powdered or fresh animal milk, and yogurt. d Food group score is calculated based on consumption of six food groups, excluding dairy products. 10 indicators related to child vaccination, reproductive health care services, and coverage of maternal and child supplementation using data from all surveyed subjects (children 0-5 years; women 15-49 years) The rank score characterizing community-level access to health care is used in the later analysis.

| Statistical analysis
Sample characteristics, distribution of complementary feeding practices, dietary diversity score, and individual food group intake were adjusted for the multistage stratified sampling design. Confidence intervals around prevalence and mean values were generated using the Taylor series linearization methods (Wolter, 2007). The linear trend in the prevalence by child age was tested by a score test for trend of odds.
To confirm the number of levels included in the regression analy-  (Rutstein & Rojas, 2006). Fourth, a set of sensitivity analyses were performed: (a) to fix child sex, maternal age, and paternal age, household wealth as covariates in the final models; (b) to include nonbreastfed children in the analysis of MDD (Supplemental Table 3); (c) to include type of assistance at delivery in the model; and (d) to rerun the analysis using multilevel bootstrapping and loosening the estimation assumption of the asymptotic normal distribution.
We used STATA/SE 15.0 (StataCorp, College Station, TX) to analyse data and to generate graphs and tables.

| Sample characteristics
The analytical unweighted sample included 7,963 children from 953 clusters. Demographic and socioeconomic characteristics by individual, household, and community levels are presented in Table 2. After weighting on survey design, about 80% of children were being breastfed at the time of the survey. About half of the children (47%) were second to fourth births. More than half (55%) were born within a 24-month birth interval. More than half (61%) were perceived as having an "average" weight at birth. Less than half (45%) had completed all the vaccinations scheduled for their age. About two-thirds (69%) of mothers were between 25 and 34 years. Although 50% of the mothers delivered their children with the assistance of a nurse or a doctor, less than 20% of mothers benefitted from four or more antenatal check-ups. About 80% of mothers did not have any formal education and most of them (87%) did not work outside their homes.
The prevalence of fathers without formal education was 56%, all of the fathers were working, and most of them (73%) worked in nonagricultural sectors. Households comprised on average 10 people with about three children under 5 years of age. The majority (96%) of households had access to water within 1 hr walking distance but in one-third (32%) of households drinking water was provided by unimproved sources. Two-thirds (67%) of the households did not have improved toilet facilities. Most eligible communities (73%) were rural.
Across communities, the median prevalence of community-level access to health care ranged from 0% for child iron supplementation in the last 7 days to 56% for health facility delivery.

| Distribution of CF practices and individual food group intake
The weighted sample sizes for INTRO, MMF, MDD, and MAD were 1,323, 5,486, 6,151, and 5,606, respectively. The prevalence 95% confidence interval (CI) of respondents reporting recommended CF practices is presented for all eligible breastfed children and by age ( Figure 1).   ). Vitamin A-rich fruits and vegetables and dairy products were consumed by 35% (31%, 38%) and 57% (54%, 59%) of all children.

| Predictors of CF indicators
The adjusted odds ratio (OR) and 95% CI for predictors of CF indicators in the multivariable multilevel models are presented in Table 3.
Comparing children 12-17 months and 18-23 months to children aged 6-11 months (reference group), the adjusted odds of meeting the MMF increased by 2.2-2.3 fold, meeting the MDD increased by 3.4-3.9 fold, and meeting the MAD increased by 3.1-3.4 fold.   Compared with the second-to-fourth born children, firstborn children had~20% lower odds of MMF and MAD, whereas children who were born as fifth or later child in the family had~20% lower odds of MAD.
Children who were perceived to be greater than average at birth had ã 50% and~30% increased odds of meeting INTRO and MDD, respectively, than children whose birth weight was perceived to be "average" Poor dietary diversity during complementary feeding puts fastgrowing infants and young children at risk of inadequate intakes for essential micronutrients, with the greatest gaps usually found for iron and zinc (Dewey, 2013). Poor dietary diversity in children has been associated with low height-for-age z scores and stunting in many low-and middle-income countries (Arimond & Ruel, 2004;Mallard et al., 2014;Rah et al., 2010). In Afghanistan, inadequate dietary diversity was associated with 34% increased odds of stunting among children 6-23 months, independent of the influence of breastfeeding, feeding frequency, maternal stature, and other risk factors and covariates (Kim, Mejía-Guevara, Corsi, Aguayo, & Subramanian, 2017).
Our analysis suggests several opportunities for improving complementary feeding outcomes in Afghanistan: First, household poverty and poor maternal education need to be addressed to improve complementary feeding practices, especially dietary diversity. Living in deprived households as indicated by the household wealth index, was an independent risk factor to poor child dietary diversity and the overall adequacy of children's diets, an observation that has been consistently found in other South Asian settings (Na, Aguayo, Arimond, Dahal, et al., 2018;Na, Aguayo, Arimond, & Stewart, 2017). Economic constraints at the household level make improving children's dietary diversity challenging, as most nutrient-rich foods that are lacking in the diets of Afghan children are more expensive and less accessible under conditions of poverty and crisis (Bouis, Eozenou, & Rahman, 2011;Dewey, 2016;Ruel, Garrett, Hawkes, & Cohen, 2010). Maternal education was another socioeconomic factor that predicted meal frequency and dietary diversity. Afghan women have been extremely deprived of educational opportunities, as evidenced by the greater than 80% female illiteracy rate and 80% of women reporting no formal education (CSO et al., 2017). This means that the majority of children are cared for by mothers with limited access to information and are at risk of inappropriate feeding and care.
It calls for education policies and programmes that promote access to equitable primary and secondary education for girls in Afghanistan.
Second, knowledge, attitudes, and practices need to be aligned with children's nutritional needs, where household poverty is not a constraint to appropriate complementary feeding practices. Both timing of complementary food introduction and meal frequency practices are likely less constrained by household wealth, as repeatedly seen in the literature (Joshi, Agho, Dibley, Senarath, & Tiwari, 2012;Kabir et al., 2012;Na, Aguayo, Arimond, Dahal, et al., 2018;Na, Aguayo, Arimond, Narayan, et al., 2018;Na, Aguayo, Arimond, & Stewart, 2017;Senarath, Godakandage, Jayawickrama, Siriwardena, & Dibley, 2012) and are less likely to be associated with general socioeconomic progress over time (Na, Aguayo, Arimond, Dahal, et al., 2018;Na, Aguayo, Arimond, Narayan, et al., 2018 (Akseer et al., 2016). We found that greater individual-level and community-level access to health care was associated with improved complementary feeding practices. However, progress in the availability of health services has not translated into progress on key complementary feeding practices. One possible explanation is that although complementary feeding is included in the basic package of services at policy level, appropriate behaviour-change counselling for improved complementary feeding practices has not yet been widely implemented as part of facility-and community-based health services.
In the recently launched Afghanistan Zero Hunger Strategic Review, a key consultative recommendation to improve child nutrition was to revise the community-based nutrition intervention package to include counselling on infant and young child feeding (Advisory Committee for Consultations, 2017). There may be other opportunities to further incorporate child feeding modules in the training of health workers' training and performance incentive. Summarizing data from 10 randomized controlled trials, a systematic review concluded that such integration was effective in increasing child energy intake, feeding frequency, and dietary diversity (Sunguya et al., 2013). Another potential channel is to integrate child feeding and nutrition education into Fourth, vast disparities in IYCF indicators call for prioritized and diversified regional policies to promote complementary feeding practices. Additional analysis revealed that key predictors of feeding practices varied significantly by region (see Supplemental Table 4 Province and other parts of Afghanistan due to insecurity, and (c) the need to limit predictor analysis to breastfed children given missing information for milk feed frequency for non-breastfed children.
Regardless of these limitations, we analysed comprehensively predictors of complementary feeding practices using nationally representative data. The statistical methods took into account the complex survey design and multi-level structure of the data. Our results remained robust in multiple sensitivity analyses.
In 2015, complementary feeding practices among Afghan children