Maternal diets matter for children's dietary quality: Seasonal dietary diversity and animal‐source foods consumption in rural Timor‐Leste

Abstract Improving the dietary quality of women and children is essential to reduce all forms of malnutrition. In this study, we assessed seasonal child and maternal dietary diversity and consumption of animal‐source foods (ASF), using 1,236 observations from combined data collected among 167 mother–child dyads in rural Timor‐Leste. We used generalized linear and logistic mixed‐effects models to examine the dietary differentials of mothers and children in two agricultural livelihood zones and across the seasons, as well as to identify household and agroecological characteristics associated with children's dietary quality in relation to their mothers'. We found dietary quality to be marginally better in coastal than in mid‐altitude zones. However, women's diets were strikingly poor, and their intake of ASF was lower than among children. Mothers exhibited preferential allocation patterns of specific ASF, dairy products and eggs, to children. The intake of ASF was predicted by seasonality. Flesh foods and red meat were much more likely to be consumed during the dry season, when cultural ceremonies are often performed. We found a positive and strongly significant association between children's dietary indicators—dietary diversity score, minimum dietary diversity and ASF consumption, and those of their mothers'. Maternal dietary quality and educational attainment, more so than agroecological characteristics, were explanatory factors of children's diet. Our study highlights that addressing the dietary quality of children in Timor‐Leste would benefit from improving women's diets through better access to nutritious foods and to secondary education.

Improving the quantity and quality of foods fed to children in their first 5 years of life and those consumed by prepregnant, gestating and lactating women, is critical to reduce malnutrition in all its forms. ASF (dairy, eggs, meat, poultry and fish) are important elements of dietary quality, containing essential amino acids that regulate growth combined with micronutrients such as iron, vitamins A and B, zinc and calcium that support optimal development (Grace et al., 2018;Headey, Hirvonen, & Hoddinott, 2018). Although the relationship between ASF intake and stunting reduction during early childhood has shown mixed results (Shapiro et al., 2019), the use of ASF to mitigate micronutrient deficiencies is a widely acknowledged food-based strategy (Leroy & Frongillo, 2007). How ASF allocation patterns differ however between children and mothers in contexts of scarcity remains poorly understood and is informed by sociocultural and household factors (Gittelsohn & Vastine, 2003).
Household food security for children and women and related care practices, for example, determine child and maternal dietary intake, and when inadequate, are an immediate cause of malnutrition (United Nations Children's Fund [UNICEF], 1990). Seasonality also can mediate food availability and access in LMIC, particularly in rural areas where agriculture-based livelihoods prevail. Yet dietary data are often collected at a single time point, lacking the ability to reflect potential changes in diets occurring through seasons and agricultural cycles (Wong et al., 2018).
In Timor-Leste, a small post-conflict lower middle-income country, the prevalence of malnutrition is persistently high with 45.6% of children under 5 being stunted and 26.6% of women of reproductive age underweight, paired with anaemia levels of 40. 3% Williams, Andersen, & Turner, 2012;Wong et al., 2018). High stunting levels reflect a history of war and internal displacement as well as lack of access to sufficient and nutritious foods (Reinhardt & Fanzo, 2014).
Despite slow progress to improve nutritional outcomes since indepen- achieved the minimum acceptable diet in 2016, capturing both adequate diversity and meal frequency (GDS et al., 2018). Minimum dietary diversity (MDD), a proxy for micronutrient adequacy, was 33.6% for this age group (GDS et al., 2018) and 40.6% for children 24-59 months (MoH, 2015), whereas for women, nationally representative data on their achievement are unavailable. ASF intake is low, only 25.5% of children 6-23 months consumed dairy products (MoH, 2015) for example, compared with the Asian regional average, of 37.7% (Headey et al., 2018).
In a country that faces cyclical food insecurity (da Costa et al., 2013), seasonal assessments of child and maternal diets require further research. Wong, Bagnol, Grieve, Li, and Alders (2016) and Wong et al. (2018) started addressing this gap, and our work builds from it by examining agroecological considerations and a broader child age range. Localized studies have assessed smallholders' diets and identified the rainy months as the most food insecure and less diverse (da Costa et al., 2013;Glazebrook, Lopes, da Costa, & Ximenes, 2007) and higher consumption of wild foods during the dry season (Erskine et al., 2015). Most of these investigations have not measured dietary adequacy using validated indicators. Additionally, the composition of diets and access to ASF might vary according to distinct agricultural profiles due to the semisubsistence nature of Timorese rural livelihoods. Thus, the contextual determinants of dietary diversity and ASF intake among smallholders warrant exploration.
This study examines the dietary quality of children 6-59 months old and their mothers living in rural Timor-Leste. Our aim was to first assess child and maternal dietary diversity and ASF intake differentials in two agricultural livelihood zones; second, to investigate the role of seasonality in women and children's consumption and

Key Messages
• Women's dietary quality and secondary education attainment were positively associated with children's diets, whereas livelihood zone played a lesser role.
• Seasonality predicted the intake of flesh foods and red meat, which were more likely to be consumed during the dry season.
• Mothers displayed preferential animal-source foods (ASF) allocation patterns to children, specifically eggs and dairy products.
• Improving the quality of children's diets in rural Timor-Leste could benefit from ameliorating mothers' diets through better access to nutritious foods and to formal education.
• To promote the production, affordability and consumption of nutrient-dense foods, including ASF, sustained nutrition-sensitive programmes are recommended.  Figure S2). Coastal lowlands have lower crop-livestock diversity and better access to markets, whereas inland mid-altitude areas have more diverse agricultural systems and poorer roads (Williams et al., 2018).
We hypothesized that mid-altitude zones access less seafood and traded goods and that ASF consumption might be lower as a result.
Out of 442 suku nationwide, 108 fall in these two zones, representing 29% of the population and all predominantly rice-growing communities (Williams et al., 2018).
Timor-Leste has a tropical climate with two seasons, wet and dry.
The wet season ranges from November to April, with a second short rainy period between June and August in the south, followed by a dry season with virtually no rains. The lean season occurs when food reserves deplete before the maize and rice harvests ( Figure 1) and is more severe in the rainfed uplands where rice is seldom grown (da Costa et al., 2013). Examined suku have an annual rainfall of 1,200-2,000 mm (Williams et al., 2018), making them comparable despite their disposition north and south of the island ridge ( Figure S1). We estimated a sample size of 200 households equally distributed by livelihood zone to give the study 80% power to detect a conservative minimum difference in dietary diversity between zones and seasons equal to 0.5 food groups (=1/3 SD) with an α = 0.05. A design effect adjustment of 1.5 was included to allow for correlation among clusters (suku) of observations assuming an intracluster correlation equal to 0.01. An estimated 20% respondent dropout rate was added to the sample size, initially calculated at 160 households. Our convenience sampling followed multiple stages: (1) selection of CDNIP suku with an ASF component; (2) identification of suku in two livelihood F I G U R E 1 Diagram of seasons, data collection iterations and tools used zones (coastal/mid-altitude), when multiple suku were eligible those with less than 50 participants were excluded and better access prioritized; and (3) random selection of 50 households per suku. After randomization, households were visited and invited to participate; when non-available/uninterested, a reserve list was used until target completion. Eligible mothers had a child aged 6-48 months, lived with an adult male and were permanent residents. Youngest child, younger than 4 years, was selected, and twins were registered as one participant.

| Data collection
Data were collected four times in 3-to 5-month intervals through a tablet-based household survey, designed in KoBoToolbox (Harvard

Humanitarian
Initiative, http://www.kobotoolbox.org/), and interviewer-administered to mothers. We aimed to capture the dry (September) and wet (January) seasons, and a transition (June) period-dry in the north and light rains in the south. We measured two dry seasons to capture potential weather variations. All iterations included questions on diets (mother and child 24-h recalls), household and agroecological (child health, crops and livestock); one included socio-economic and food security questions, measured through the Food Insecurity Experience Scale ( Figure 1).
Dietary assessments were based on the maternal recollection of all foods and beverages consumed, and given to children, 24 h before the survey. Open recall was followed by the list-based method (Food Following the DDS, ASF were classified in three groups (dairy, flesh foods and eggs) and measured as intake of any ASF consumed yesterday.

| Explanatory variables
The UNICEF (1990)  women and children consumed more ASF-rich diets through bivariate regressions that accounted for the repeated sampling (Table 2).
Following, we explored seasonal variations in maternal and child food groups' consumption (Table 3). Finally, we modelled the relationship between child and maternal dietary indicators after controlling for potential individual, household and agroecological predictors in a multivariable mixed-effects model using the combined dataset (Table 4).
Given the longitudinal nature of the data, we used generalized linear mixed models (GLMM) in all regressions. These models included a random intercept to account for the autocorrelation of observations collected from the same subject over time and clustering of the observations from the same household, incorporating both random and fixed effects. We applied the logistic form of GLMM for binary outcomes. To build the final models, we first performed bivariate GLMM analyses to test associations between children's outcomes (DDS, MDD and ASF consumed yesterday) and potential explanatory predictors (Table S4). Covariates with a significant bivariate association (P < .05) were included in the initial multivariable regression models as a fixed effect. For each outcome, the final model was built by excluding non-significant variables (P > .10) through backward stepwise selection of predictors. We included one dietary diversity indicator per model, due to collinearity as tested using the variance inflation factor (collin command), and two fixed factors, livelihood zone and aspect, which aggregated suku in pairs. Model residual diagnostics were performed for final variables to determine impact of potential outliers on model coefficients. Residual plots across explanatory variables indicated a random residual spread.

| RESULTS
The final sample included 167 mother-child dyads with 618 respective dietary recalls across the seasons, totalling 1,236 observations over 12 months. Thirty-three dyads (16.5%) were excluded due to having less than three observations, we assumed that participants' Note: The table presents outcomes for samples across four time points, totalling 1,236 dietary recalls from 167 mother-child dyads: 618 for mothers, 381 for children aged 24-59 months, and 237 for children 6-23 months old. Frequencies and means are across seasons, to be interpreted as incidences of dietary recalls. Modelling based on mixed-effects GLMM with a random intercept (household) accounting for repeated sampling. Accordingly, odds ratios and coefficient estimates account for multiple observations over time, to be interpreted at the participant level. Abbreviations: ASF, animal-source foods; DDS, dietary diversity score-infant and young children feeding; DDS-W, dietary diversity score-women of reproductive age; GLMM, generalized linear mixed models; MDD, minimum dietary diversity-infant and young children feeding; MDD-W, minimum dietary diversity-women of reproductive age; SD, standard deviation. a OR = odds ratio, unless β = coefficient if specified. Mid-altitude zone is the reference value. CI, confidence interval. b P values are a test of association between participants' dietary quality indicator and livelihood zone, applying a mixed-effects logistic model for binary outcomes (OR) and a generalized linear mixed-effects model for continuous outcomes (β). c To enable mothers' and child DDS comparability, the seven-food groups indicator is presented for mothers. d DDS and DDS-W cannot be compared as they use different food grouping parameters. e ASF groups: dairy products (excluding sweet condensed milk), flesh foods (organs, red meat, poultry and fish/shellfish) and eggs.
T A B L E 3 Seasonal food groups and animal-source foods consumption and dietary diversity scores of women and children 6-59 months old in rural Timor-Leste, samples across four time points 2017-2018 Child age group was kept in the multivariable models regardless of its significance due to being a child characteristic of interest in this study.
e Livelihood zone and aspect aggregate suku in distinct pairs, used as fixed factors in the models instead of suku.

| Dietary quality indicators by livelihood zone
On aggregating dietary recalls across the four data points, we found that almost all child and maternal dietary indicators scored higher in quality in coastal zones, yet only some associations were statistically significant ( Table 2). The DDS derived from aggregated recalls for coastal children 24-59 months, whose mean was 2.9 ± 1.1, was a third of a food group higher than in mid-altitude areas (95% CI: 0.05, 0.59; P = .018). This pattern was also observed for children 6-59 months (β = 0.28; P = .019) but not for the younger cohort, who overall consumed less diverse diets. Although the dietary recalls of older children were 2.18 times more likely to report four or more groups in coastal areas, this difference was not statistically significant (P = .065). More recalls collected for children achieved MDD at least once in the coast, 45.8%, compared with 40.5% in mid-altitude zones, whereas for mothers the ratio was much lower, 27.7% and 20.2%, respectively ( Figure S3). An extremely low proportion of maternal dietary recalls (less than 10%) met MDD-W. In mid-altitude zones, recalls of children and mothers were more likely to show two or less food groups (P < .030), an indicator of poor micronutrient intake.
A larger proportion of children's dietary recalls (45.6%) included ASF than women's (37.5%), consistent across zones. Overall, recalls among children 24-59 months reported more dairy products and eggs and slightly more red meat and poultry compared with their mothers'; whereas those from children 6-23 months recorded more dairy and eggs ( Figure S4). Recalls of coastal children were 1.90 times more likely to include ASF in the last 24-h (95% CI: 1.18, 3.07; P = .008). Data on frequency intake of ASF types (Table S2) showed significantly higher coefficients for dietary recalls among coastal children and women (β = 0.48, P < .010). More maternal recalls did not include any ASF in the past week (17.5%) compared with those for older children (13.1%).
In coastal zones, recalls of children had at least twice greater odds to include eggs (OR = 2.11; P = .034), fish (OR = 2.56; P < .001) and dairy (OR = 2.95; P < .001), whereas the likelihood of including red meat was 33% lower (OR = 0.67; P = .063) than that of mid-altitude children.

| Seasonal variation of diets
Dietary composition was seasonal, particularly ASF intake among children, and specific ASF and most vegetables and fruits among mothers (Table 3). Flesh foods showed a notable seasonal trend with greater consumption in the dry seasons when around 40% of women and children consumed flesh foods, highly significant for both (P < .001). Child and maternal intake of red meat halved during the wet and transition seasons (10.3-13.3%). The odds of children consuming flesh foods were lowest in the transition period (June), 0.36 (95% CI: 0.21, 0.63; P < .001) times or 64% lower than in the first dry season (data not shown). Fish consumption was associated with seasonality, highest in the wet (16.9-19.9%) and lowest in June (6.3-7.6%), whereas poultry intake was low (less than 7%). Dairy product consumption was marginal for mothers (less than 3.7%), whereas among children, intake declined sharply from 21.0% to 2.6% (P < .001). Children's egg consumption was around double that of their mothers and highest in the wet (16.9%; P = .005). Maternal intake of dark green leafy vegetables topped in the wet (83.8%) and dropped in the second dry season (59.9%; P < .001), also found for other vitamin A-rich fruits and vegetables, and other fruits (P = .036).
Children's intake of fruits and vegetables was independent of season.
These patterns were found similarly in northern and southern suku (Table S3). Nonetheless, as children grew older in age, their mean food groups' consumption increased gradually ( Figure S5). DDS did not show a distinctive reduction in the wet season, as expected. 24-59 months old (β = 0.20; P = .002), having consumed an ASF group the day before (β = 0.94; P < .001), with a mother that attended primary school (β = 0.20; P = .018) or secondary or above (β = 0.21; P = .017), and access to improved sanitation (β = 0.16; P = .037), were significantly associated with higher DDS. Second, the odds of achieving MDD were 9.80 times higher among children whose mother achieved MDD-W (P < .001). Significant predictors controlled for were child consumption of ASF (OR = 22.77; 95% CI: 9.77, 53.05; P < .001), mother's secondary education or more (OR = 4.09; P = .002) and season through a negative association and lowest in the second dry season (OR = 0.25; P < .001).

| Characteristics associated with child dietary diversity and ASF intake
The predictor with the largest effect size for a child achieving MDD was consuming an ASF the day before. Third, mother's intake of ASF was strongly associated with a child also consuming ASF (OR = 21.83; P = .001). Children were significantly less likely to consume an ASF the day before when living in a household with more than seven members, lowering their odds by 43% (OR = 0.57; P = .021). Ownership of buffalo or chickens were not associated with ASF intake (Table S4). Final models included WI, north-south aspect and livelihood zone as predictors, which despite statistically significant in the bivariate analyses, these lost significance after controlling for other covariates.   (Spencer, Sanders, & Judge, 2018). Our results found higher odds to consume eggs, fish and dairy products among coastal children, whereas red meat was more prevalent in mid-altitude zones. Surprisingly, livestock ownership was not statistically associated with children's ASF consumption, perhaps due to the importance of livestock for income, rituals, and status (Bettencourt, Tilman, Narciso, Carvalho, & Henriques, 2015). Lesser ASF intake in mid-altitude suku could reflect lower market access and asset levels-a proxy for household income, as described in the zones characterization (Williams et al., 2018).

| Strengths, limitations and implications
Key strengths of this study were the volume of dietary recalls obtained over 1 year and data collection quality through the recruitment of local and trained enumerators. We faced several limitations.
The longitudinal design incurred participant dropout, which we assumed to be independent of specific characteristics, yet we lack certainty. Our sample size and geography were not representative of all rural populations, limiting the results' generalizability. Dietary data were self-reported, thus prone to recall error. It is plausible that foods recommended by the intervention, including orange-flesh sweet potatoes, moringa (Moringa oleifera) and eggs, were overreported due to participant bias and socially desirable responses. This bias would be consistent across suku.
Our results show that policies and programmes aimed at improving infant and young child feeding practices in Timor-Leste would gain from supporting mother's access to nutrient-dense foods and secondary education. Nutrition-sensitive agriculture programmes could have an important role to play to improve the production of nutritious foods, including ASF, particularly in inland areas. Smallholders would benefit from widespread market access to affordable eggs and nonsugary dairy products. Efforts are required to continue promoting optimal feeding practices for children 6-23 months, more nutritionally

| CONCLUSION
In this study, we investigated the dietary quality of mothers and children 6-59 months old in rural Timor-Leste and explored agroecological factors including seasonality. We found that, first, dietary diversity and ASF intake were marginally better in coastal livelihood zones. Yet women's dietary achievements were outstandingly poor-only a fraction was likely to meet micronutrient adequacy, and their consumption of ASF was lower than among children. Second, that seasonality predicted the intake of flesh foods and red meat, both much more likely to be consumed during the dry season. Lower quality diets were more acute during the transition season in June, suggesting distinct nutrition insecurity patterns. Further, women exhibited preferential allocation patterns of specific ASF to children, dairy products and eggs. Third, we found a positive and strongly significant association between children's dietary indicators, DDS, MDD and ASF consumption, and those of their mothers' . Maternal dietary quality and education achievement, more so than agroecological characteristics, were explanatory factors of children's diet among our sample. These findings contribute to the understanding of rural diets' contextual determinants in LMIC and provide insights on ASF allocation patterns in a unique sociocultural setting.
Overall, our study highlights that addressing the dietary quality of children in Timor-Leste, as has been shown in other resource-poor settings, would benefit from improving women's diets through better access to nutritious foods and to secondary education.