The Western Australian Pregnancy Cohort (Raine) Study has been described in detail elsewhere (34). In brief, the Raine Study is a multigenerational cohort study which began with the recruitment of 2900 pregnant women (Generation 1) through the public antenatal clinic and local private clinics in Perth, Western Australia between 1989 and 1991, for an initial study to investigate the effects of repeated ultrasounds on fetal growth (27). Their 2868 births (Generation 2) have been followed up since birth and approximately biennially thereafter. To date, comprehensive dietary assessments have been undertaken at follow ups conducted at 14, 17, 20 and 22 years of age.
This analysis uses dietary data collected at 14, 17, 20 and 22 y and covariates measured at 14 y, including body mass index (BMI), physical activity and fitness level, sedentary behaviour, parental socio-economic status and family functioning status in Generation 2 of the Raine Study.
At 14 and 17 y of age a semi-quantitative food frequency questionnaire (FFQ) designed by the Commonwealth Scientific and Industrial Research Organisation (CSIRO) Australia was administered to assess usual food and nutrient intakes (35). The FFQ collected information on usual frequency of consumption and serving sizes (in household units) of 227 food and beverage items. Nutrient intakes estimated by the CSIRO FFQ have been shown to be comparable with those estimated using a 3-day diet record at 14 y in this cohort (36).
As the CSIRO FFQ was not available for use at the 20 and 22 y follow ups, usual dietary intake at 20 and 22 y was estimated using the 74-item semi-quantitative Dietary Questionnaire for Epidemiological Studies (DQESV2) FFQ, developed by the Cancer Council of Victoria (CCV), Australia (37). The DQESV2 FFQ has been found to be reproducible and suitable for ranking respondents according to their estimated nutrient intakes and comparable to the CSIRO FFQ (38, 39). The DQESV2 collects information on the usual frequency of consumption of food and drinks, and usual serve sizes using pictorial examples.
Unlike the CSIRO FFQ, the DQESV2 FFQ did not include a comprehensive list of beverages therefore, a semi-quantitative beverage questionnaire was administered at 20 and 22 y. The DQESV2 FFQ included fruit juice, flavoured milk, milk and alcohol intakes. The semi-quantitative beverage questionnaire listed other beverages that were included in the CSIRO FFQ including: water, ‘fizzy drink’, diet ‘fizzy drink’, energy drink, diet energy drink, tea, herbal tea, green tea, instant coffee, and ground coffee. Estimates of usual serve size were collected (with examples of typical serve sizes provided) along with the usual frequency of consumption.
The FFQs were completed with assistance from a parent or caregiver at 14 y (2003-2006) and by the study respondents at 17 y (2006-2009), 20 y (2010-2012) and 22 y (2012-2014). All FFQs were checked by a research nurse and missing and unclear responses were corrected with the study respondent at the time of their physical assessment. Australian Food Composition Tables were used to estimate usual nutrient intakes and total energy intake (21).
To conduct dietary pattern analyses, all food and beverage items listed in the FFQs were assigned to 38 pre-defined major food groups based on nutrient profile and culinary usage (30). Although the two FFQs were highly similar, a small number of food groups in the CSIRO FFQ were not captured by the DQESV2 FFQ, including: meat-based mixed dishes, milk-based dishes, soups, sauces, and dried fruit.
The Raine Study ‘Healthy’ and ‘Western’ DPs have been described in detail previously (30). In brief, all 38 major food groups were entered into a factor analysis and varimax rotation applied, to achieve uncorrelated factors or dietary patterns, using PROC FACTOR in SAS (SAS Institute, Cary, North Carolina, USA). The resulting factor solution identified two major dietary patterns that explained the greatest amount of total variance in food group intakes (21.5% in total) (40). The total variance of the food intakes explained by the individual dietary patterns was 13% for ‘Western’ dietary pattern and 8.5% for ‘Healthy’ dietary pattern (40). Each participant received a z-score for the ‘Healthy’ and the ‘Western’ dietary pattern (calculated using PROC SCORE, SAS), which indicated how close their reported dietary intake corresponded with the pattern, relative to the rest of the study sample (z-score mean=0; SD=1).
For each dietary pattern, the factor solution generated factor loadings for each food group, indicating their ‘weighting’ in each dietary pattern (30). Foods with a factor loading greater than an absolute value of 0.30 were considered the most influential in each pattern. At 14 y, the ‘Healthy’ dietary pattern was characterised by high intakes of wholegrain cereals, fresh fruit, legumes, steamed, grilled or canned fish and all vegetables, except potatoes (30). The ‘Western’ dietary pattern consisted of positive factor loadings for takeaway foods, red meats, processed meats, full-fat dairy products, fried potato, refined grains, soft drinks, confectionery, and crisps (30). This pattern showed strong correlations with intake of energy, total fat, saturated fat, cholesterol and refined sugar (32). ‘Healthy’ and ‘Western’ dietary pattern z-scores estimated using the CSIRO FFQ have been shown to be comparable to those estimated using a 3-day diet record in this cohort, at 14 y (36).
The exploratory factor analysis described above was repeated using food group intakes estimated at 17, 20 and 22 y. This identified two major dietary patterns consistent with the ‘Healthy’ and ‘Western’ dietary patterns identified at 14 y, at each age. Apart from some minor variations, the factor loadings were similar across 14, 17, 20 and 22 y (Supplementary Table 1). As these both dietary patterns were consistent over time at the population level, a longitudinal analysis of z-scores for the ‘Healthy’ and ‘Western’ dietary patterns was deemed appropriate. However, to score individuals for exactly the same dietary patterns over time, applied dietary pattern scores were estimated. These were calculated by applying the factor scoring coefficients for each food group identified at 14 y, to dietary intakes at 17, 20 and 22 y (confirmatory analysis) using PROC SCORE in SAS. Those five food groups not captured by DQESV2 FFQ (meat-based mixed dishes, milk-based dishes, soups, sauces and dried fruit) did not load strongly on either pattern at 14 or 17 y (see Supplementary Table 1) and were therefore excluded from the confirmatory factor analyses. Despite this difference, exploratory and applied dietary pattern scores were highly correlated (r>0.94).
Height was measured using a Holtain stadiometer without shoes and weight was recorded using a Wedderburn digital chair scale with light clothing to calculate BMI (kg/m2) at 14 y. Self-reported physical activity levels were estimated by asking the respondents to report the number of times they exercised enough to sweat, when they were not at school (excluding compulsory school physical education sessions). Respondents were asked to choose one option from five categories ranging from exercising once a month or less, through to exercising every day. These data were used to create an ordinal variable: low (exercising once per month); medium (exercising 1-3 times per week); and high (exercising more than 4 times per week). Self-reported physical activity levels have been shown to be highly correlated with an aerobic fitness data measured objectively using a Physical Working Capacity (PWC-170) on ergometer bicycle, in this cohort (30). However, only self-reported physical activity was used in this analysis, as more respondents completed this questionnaire than the ergometer bicycle test.
Maternal factors during pre-pregnancy and pregnancy, such as maternal body weight may have a direct or indirect (e.g. as a lifestyle marker) influence on the development of childhood obesity and other metabolic risks (41). Information on maternal self-reported pre-pregnancy weight (kg) was collected at enrolment into the study (~18 weeks of gestation). Height (cm) was measured at the first physical assessment (16-20 weeks of gestation). Maternal age at child birth was recorded upon birth of the study child.
Since several studies have suggested that parental socio-economic factors are closely associated with children’s dietary intake, we investigated the potential role of socio-economic factors in determining an individual’s dietary pattern trajectory (42-45). A standardised questionnaire was used to obtain information on parental socioeconomic status (SES) at 14 y. This included categories of maternal education represented by highest school year (<10 y, 10-12 y and >12 y), family income in Australian dollars (≤35k, >35-50k, >50-70k, >70-104k and >104k), family structure (two-parent or single-parent, with de facto parents considered a two-parent family structure), parental smoking status (yes or no) and family functioning (General Functioning Scale (GFS) from the McMaster Family Assessment Device (46)). The GFS comprised questions on family communication, affective responsiveness and behaviour control, with higher scores representing better family functioning. The GFS scores were classified into quartiles (‘0’= quartile 1(scores ≤ 25), ‘1’=quartile 2 (scores between 26 and 28), ‘2’=quartile 3 (scores between 29 and 33) and ‘3’=quartile 4 (scores between 34 and 39).
Summary statistics for the eligible study sample at 14 years of age (n=2424) and individuals included in the trajectory analyses i.e. individuals with at least two outcome measures (n=1402) were summarised and compared using Chi-square and t-test for categorical and continuous variables, respectively (34). Summary statistics are provided for ‘Healthy’ and ‘Western’ dietary pattern scores at each time point, with means and standard deviations, and the medians and ranges as well as missing numbers (Supplementary Table 2).
Group-based trajectory modelling was used to assess variation in the developmental course of z-scores for both the dietary patterns and separately for males and females, using PROC TRAJ in SAS V9.4. Z-scores for ‘Healthy’ and ‘Western’ dietary patterns from each respondent were used as dependent variables and their ages at each follow-up were applied as independent variables (47). Groups were determined based on interpretability and an information criterion approach and the degree of the polynomial trajectory was determined by statistical significance. Based on the results of the trajectory modelling, two groups or trajectories for each dietary pattern (and sex) had the best model fit.
Multivariable logistic regression, with group membership as the outcome (Group 1 vs. Group 2 of trajectories), was carried out to examine the relationships between dietary pattern trajectory and: maternal education level; maternal age at birth; maternal pre-pregnancy BMI; whether the parents smoked, whether the father lives with the family, family income, family functioning status; and sex, BMI and self-reported physical activity. Multivariate odds ratios (ORs) and 95% confidence intervals (CIs) are presented with p-values for those variables significant in the model. All analysis was carried out using SAS V9.4.