Study design and participants
The present study used data from the Growing Up in Singapore Towards healthy Outcomes (GUSTO) cohort – an ongoing mother-offspring study in Singapore [25]. The GUSTO study recruited pregnant women (aged ≥ 18 years) in their first trimester from 2 major public maternity units in Singapore (National University Hospital and KK Women’s and Children’s Hospital) between June 2009 and September 2010. These women are Singapore citizens or permanent residents of Chinese, Malay or Indian ethnicity with homogenous parental ethnic background. Detailed description of the study design and recruitment criteria have been published previously [25].
Children of these mothers were followed up from birth. The present study included children with dietary data from at least two follow-up time-points at 18 months, 5 and 7 years of age for the analysis of dietary pattern trajectories (Fig. 1). Additionally, maternal feeding practices assessed at 15 months, 3, and 5 years of age were incorporated into the study. All procedures in GUSTO have received ethical approval from the Institutional Review Boards at the 2 public maternity units (CIRB 2018/2767; DSRB D/2009/00021, B/2014/00406) where the women were recruited. Written informed consent was acquired from the participants at study recruitment. The Strengthening the Reporting of Observational Studies in Epidemiology-Nutritional Epidemiology (STROBE-nut) checklist was used for the reporting of this study (Additional File 1) [26].
Flowchart of participants included in dietary pattern trajectories in the Growing Up in Singapore Towards healthy Outcomes (GUSTO) study
Assessment of dietary intakes
Validated food frequency questionnaires (FFQ) were used to assess dietary intakes of children at 18 months, 5 and 7 years of age. The FFQ was first developed to assess dietary intake of 18-month-old children in GUSTO. The list of food items in the FFQ was created based on data from 24-hour recalls or 3-day diet records collected from the GUSTO children at 12 months and with reference to the validated Southampton Women’s Survey 12 Month Infancy Questionnaire [27]. Inputs from paediatric dietitians were also sought to include foods commonly consumed at 18 months of age. This FFQ was subsequently modified to closely reflect dietary habits of 5 and 7-year-old children. Details on the development and validation of the FFQ used at each time-point have been previously published [28,29,30].
The FFQs at 18 months, 5 and 7 years of age comprised of 94, 112 and 120 items, respectively, covering food groups such as breads, spreads, cereals, rice and rice dishes, noodles and pasta, vegetables and legumes, fruits, poultry and meat, fish and seafood, eggs, savoury and sweet snacks, desserts, fast food, dairy products, and beverages. Mothers (or caregivers familiar with the child’s diet) were asked to indicate their child’s frequency of consuming each food item in the previous month, and the average amount consumed using household measurements. To assist in portion size estimation, images of standard household utensils (e.g. bowls, spoons, and a standard plate), and portion sizes of several food items (e.g. dessertspoons of meat or vegetables, pieces of fruits, a bowl of cereals, a slice of cake) were provided.
Derivation of dietary patterns
At each time-point (18 months, 5 and 7 years of age), the FFQ items were collapsed into 42 food groups based on similar nutritional composition and culinary use (Additional File 2). Principal component analysis (PCA) with varimax rotation was conducted separately at each time-points to derive dietary patterns. The number of factors chosen to retain was based on eigenvalues > 1, the break point of the scree plot, and the interpretability of the factors [31]. Two dietary patterns were derived at each time-point and their combined percentage of total variance ranged from 8 to 22% (Additional File 3). The Kaiser-Meyer-Olkin values were 0.72, 0.68, and 0.73 for 18 months, 5 and 7 years of age, respectively. Detailed information on the factor loadings for each food groups included in the PCA are summarised in Additional File 3. Apart from some slight differences, the identified dietary patterns were largely similar and consistent in the factor loadings for most of the food groups as well as their directions, at all ages. Overall, one pattern was characterised by fish and seafood, vegetables, poultry meat, nuts, seeds and legumes, thus labelled as the ‘healthy’ dietary pattern; the other pattern was characterised by low-fibre bread, processed meat, fast foods, potatoes, fishes and poultry meat that were deep-fried and curried and chocolate and sugary confectioneries, thus labelled as the ‘less healthy’ dietary pattern.
Each child received a z-score (z-score mean = 0 and SD = 1) for the ‘healthy’ and ‘less healthy’ dietary patterns, which indicated the extent the child’s dietary intake adhered to these identified dietary patterns, correspondingly, relative to the rest of children in this study. A dietary pattern z-score of less than zero, equivalent to zero, and more than zero indicate lower adherence, average adherence and higher adherence to the identified dietary patterns, respectively.
Modelling of dietary patterns trajectories
The z-scores for both ‘healthy’ and ‘less healthy’ dietary patterns were employed in group-based trajectory analysis to identify trajectory groups for each dietary pattern from 18 months to 7 years of age in all children. Group-based trajectory analysis is a statistical method that assigns individuals into groups with similar patterns of a longitudinal measure and therefore the variability in trajectory shapes for each DP in this study i.e., ‘healthy’ or ‘less healthy’ can be assessed. An underlying data distribution of censored normal model was used to identify the trajectory groups for the 2 derived dietary patterns [32]. The number of trajectory groups were determined based on the average of Bayesian Information Criteria (BIC), Akaike Information Criteria (AIC) proportion of estimated trajectory groups (the smallest group includes at least 5% of participants) and statistics of model fit [33]. Children were classified into each of the dietary pattern trajectory groups based on the maximum estimated probability of belonging to each group [34]. Two trajectory groups were found to be the best model fit based on their interpretability, proportion of estimated belongings and showed sufficiently high average posterior probability of belonging to each group (0.75–0.90), and therefore included in this study (Additional File 4). Correspondingly, two trajectory groups for each of the two dietary patterns (‘healthy’ and ‘less healthy’) from 18 months to 7 years of age, respectively, were included in this study.
Assessment of maternal feeding practices
Maternal feeding practices were assessed using the Infant Feeding Questionnaire (IFQ), Preschooler Feeding Questionnaire (PFQ) and Comprehensive Feeding Practices Questionnaire (CFPQ) when the children were 15 months, 3 years and 5 years of age, respectively [35, 36]. The assessment of feeding practices at these ages were deemed suitable as this is when table foods, food preferences and exposures are much established during childhood. The questionnaires were self-administered by mothers in English (the official working language used in Singapore). Mothers who required the questionnaires to be translated to other languages such as Mandarin, Malay or Tamil were excluded.
The IFQ was used to assess maternal feeding beliefs and practices for the first time when their children were 15-months of age [37]. This 28-item self-report measure employed a 5-point Likert scale ranging from ‘never’ to ‘always’ or from ‘disagree a lot’ to ‘agree a lot’. Factor analysis of the original IFQ revealed 7 subscales: concern about infant undereating or becoming underweight, concern about infant’s hunger, awareness of infant’s hunger and satiety cues, concern about infant overeating or becoming overweight, feeding infant on a schedule, using food to calm infant’s fussiness and social interaction with the infant during feeding. The use of IFQ in this cohort was previously published elsewhere [35].
The PFQ was designed for parents of children aged between 2 and 5 years of age, and assesses parents’ feeding practices as well as beliefs and concerns about their child’s weight [37]. The PFQ comprises of 35 items measuring 8 subscales: difficulty in child feeding, concern about child being overweight, using food to calm the child, child’s control of feeding interactions (refers to understanding the extent to which a child is involved in or influences their feeding experiences), age-inappropriate feeding, pushing the child to eat more, concerns about child overeating and being under or overweight, structure during feeding interactions. Mothers rated their response on a 5-point Likert scale ranging from “never” to “always” or “disagree” to “agree”. Further details on the PFQ and its validation have been previously published [36, 37].
The CFPQ was designed for parents of children aged 18 months to 8 years [38]. The CFPQ comprises 49 items measuring 12 subscales: child control, emotion regulation, encourage balance and variety, environment, food as reward, involvement, modelling, monitoring, pressure to eat, restriction for health, restriction for weight control, and teaching about nutrition. Similar to the PFQ, mothers rated their response on a 5-point Likert scale ranging from “never” to “always” or “disagree” to “agree”. The CFPQ is a well-known measure that has been extensively used and validated to assess parental feeding practices with preschool aged children. It has also been validated for use in a Malaysian population [39], which has very similar ethnic profile, cultural and dietary practices to Singapore. Further details on the CFPQ and its validation have been previously published [36, 38].
Derivation of maternal feeding practices patterns
To identify patterns of maternal feeding practices, PCA with varimax rotation was performed separately at 15 months, 3 years, and 5 years of age. The analysis incorporated the 7 subscales from the IFQ, 8 subscales from the PFQ, and 12 subscales from the CFPQ. Similar to derivation of dietary patterns, the number of factors chosen to retain was based on eigenvalues > 1, the break point of the scree plot, and the interpretability of the factors [31].
PCA conducted on maternal feeding practices at 15 months of age did not capture similar feeding practices patterns as observed at 3 and 5 years of age (Additional File 5). This is likely because at 15 months, infants are still largely dependent on caregivers for selecting and preparing foods / meals. They only become more independent and require more autonomy as they grow older. This is aligned with Vaughn et al., who recommends that child age and development stage are important factors to consider when examining relationships between feeding practices and child diet [15]. Instead, the scores of the individual subscales from the maternal feeding practices at 15 months of age were presented under Table 1 to illustrate their relationships with the children adhering to different DPs.
At both 3 and 5 years of age, two patterns of maternal feeding practices were derived, which we labelled as ‘structured with autonomy support’ and ‘coercive control’, respectively (Additional File 5). The Kaiser-Meyer-Olkin values were 0.61 and 0.72 for ages 3 and 5, respectively. The percentage of variations for all maternal feeding practices items ranged from 18 to 24% and 15–25% for ‘structured with autonomy support’ and ‘coercive control’ patterns, respectively from 3 to 5 years of age (Additional File 5).
At 3 years of age, the ‘structured with autonomy support’ pattern was characterised by greater structure during feeding interactions (0.772), pushing the child to eat more (0.439) and less concern about child overeating or being overweight (-0.558); while the ‘coercive control’ pattern was characterised by concern about child being underweight (0.651), as well as practices of age-inappropriate feeding (0.645), child’s control of feeding interactions (0.611), using food to calm the child (0.604) and difficulty in child feeding (0.588).
At 5 years of age, the ‘structured with autonomy support’ pattern was characterised by practices of modelling (0.732), balance and variety (0.723), teaching about nutrition (0.690), healthy environment (0.644), monitoring (0.485) and involvement (0.337); while the ‘coercive control’ pattern was mainly characterised by practices of giving food as reward (0.762), emotion regulation (0.666), restriction for health (0.502), pressure to eat (0.383) and restriction for weight control (0.375).
Each mother received a z-score for each pattern of maternal feeding practices and was subsequently categorized into high (z-score ≥ 0) and low adherence (z-score < 0) to each pattern for the ease of interpretation in the subsequent analyses.
Maternal characteristics
Data on maternal age, ethnicity, highest educational attainment and household income were collected at the recruitment visit. At 26–28 weeks’ gestation, mothers’ self-reported cigarette smoking and environmental tobacco smoke (ETS) exposure during pregnancy were obtained via interviewer-administered questionnaires. Maternal dietary intake during pregnancy was also assessed at 26–28 weeks’ gestation using a 24-hour recall administered by trained clinical staffs. Overall diet quality was assessed using the Healthy Eating Index for pregnant women in Singapore (HEI-SGP) in which a higher score indicated a better diet quality [40].
Postnatal or child characteristics
Information on infant sex was retrieved from hospital delivery records. Mothers’ breastfeeding practices (duration of any breastfeeding) were obtained via interviewer-administered questionnaires during postnatal visits at week 3, week 6 and at every 3-month interval from 3 to 12 months of age. Infant birth weight was measured to the nearest gram using a calibrated scale (SECA 334 weighing scale; SECA Corp). Recumbent birth length was measured from the top of the head to the soles of the feet using an infant mat (SECA 210 mobile measuring mat; SECA Corp) to the nearest 0.1 cm. At ages 5, 7 and 8 years, children’s weight (kg) was measured in light clothing using a SECA 803 weighting scale, and height (cm) with a stadiometer (SECA 213, Hamburg, Germany). Age- and sex-specific BMI z scores were calculated based on the WHO standards and references [41].
Statistical analysis
Descriptive statistics included mean and standard deviation for continuous variables, and number and percentage for categorical variables. Separate multivariable logistic regression analyses were performed to determine the associations between the ‘healthy’ and ‘less healthy’ dietary pattern trajectory groups and patterns of maternal feeding practices at ages 3 and 5 years. Crude and adjusted odds ratios (OR) and their 95% confidence intervals (CIs) were reported, accordingly. Potential covariates in these multivariable analyses included mother’s age, ethnicity, highest education and their diet quality score during pregnancy. Modelling of dietary patterns trajectories were performed using STATA/SE version 15.0 (StataCorp, 2017) and all other analyses were performed using SPSS version 26 (IBM Corp., 2019). The significant level for all statistical analyses was set at P < 0.05.
Add Comment