This is secondary analyses of data from the ROLO (Randomised Control Trial of a low Glycaemic Diet in Pregnancy) longitudinal birth cohort study. The primary study was a randomised control trial (RCT) of a low glycaemic diet in pregnancy to prevent the recurrence of foetal macrosomia (birth weight >4 kg) [16]. Full detailed methodology and findings are published elsewhere [16].

In brief, women on their second pregnancy, who had previously given birth to an infant weighing >4 kg were recruited (n = 800). Those included were < 18 weeks gestation, with a singleton pregnancy, had no previous history of gestational diabetes, and were over 18 years old. Women were randomised to either the intervention group which received dietary advice on a low glycaemic diet, or the control group who received routine antenatal care. Results showed no significant difference in birth weight between the control and intervention groups. Women in the intervention group did, however, reduce their gestational weight gain and improve their glycaemic control [16]. The mother and child dyads from the primary study have been followed up at regular timepoints, including when the children were 5 and 9–11 years old. The current analyses included 306 mother-child dyads from the 5 year and 224 from the 9–11-year follow-up. A total of 167 children had matched data on appetitive traits for both timepoints. Ethical approval was provided by ethics committees at the National Maternity Hospital, Dublin, Our Lady’s Children’s Hospital, Crumlin, Dublin, Ireland (Ethics reference number: GEN/279/12) and the office of Research Ethics Committee, University College Dublin (LS-15-06-Geraghty-McAuliffe).

At the 5 and 9–11-year follow-up, maternal and child weight was measured using a calibrated stand-on digital weighing scale (SECA 813 GmbH & co. Kg. Hamburg, Germany) to the nearest 0.1 kg. Participants were measured in light clothing without shoes. Standing height was measured, without shoes, with head aligned in the Frankfort plain, using a free-standing stadiometer (SECA 217 GmbH & co. Kg. Hamburg, Germany) and measurements were recorded to the nearest 0.1 cm. Body mass index (BMI) was calculated as kilogram per metre squared (kg/m2). Children’s weight, height, and BMI values were converted to age-and sex-specific z-scores according to the 1990 UK reference data using Excel LMS Growth Macro [17, 18]. BMI z-scores were categorised as ‘Underweight’ (BMI z-score <−2.0), ‘Healthy weight’ (BMI z-score >−2.0 and ≤1.0), ‘Overweight’ (BMI z-score >+1 SD) or Obesity (BMI z-score >+2 SD), using the World Health Organisation criteria for children aged 5 to 19 years [19]. All measurements were taken as per study protocol and were carried out by trained researchers.

At both timepoints, mothers were asked to complete the Children’s Eating Behaviour Questionnaire (CEBQ). The CEBQ [20] is a parent reported, validated psychometric tool, which was developed to capture individual differences in eating styles that may contribute to both underweight and overweight in children [20]. The CEBQ has been shown to have good internal and test-retest reliability and has been validated against observational measures of eating behaviour [5, 21], corresponding well to children’s energy intake. Each item response is graded on a 5-point Likert scale (‘never to always’), with five items within the CEBQ being reverse scored, due to opposite phrasing. Each question relates to one of eight appetitive traits which can be classed as either ‘food approach’ or ‘food avoidant’. The ‘food approach’ domain represents the degree to which a child has a more avid appetite and greater interest in food and includes ‘Food Responsiveness’, ‘Enjoys Food’, ‘Emotional Overeating’ and ‘Desire to Drink’. The ‘food avoidant’ domain represents the degree to which a child has a smaller appetite and is less interested in food and includes ‘Satiety Responsiveness’, ‘Slowness Eating’ ‘Emotional Undereating’ and ‘Food Fussiness’. Each subscale is summed to give a total score, and this is divided by the number of items within the subscale, to give a mean score of the sum. A higher score indicates the child is more likely to express this appetitive trait. A more detailed description of each subscale is presented in a previous publication [22]. Cronbach alpha was completed on each CEBQ subscale item for the 5 and 9–11 year old follow-up. For the 5-year-old follow-up internal reliability coefficients (Cronbach’s α) ranged from 0.695 to 0.928 and in the 9–11 year old follow-up internal reliability coefficients ranged from 0.710 to 0.917, thus all questions were included in the analysis.

To assess children’s dietary intake, mothers completed a food frequency questionnaire (FFQ) on behalf of their child at both timepoints. This measured their habitual intake over the previous year. The FFQ used at the 5 year and 9–11 year follow-up was a 53-item questionnaire based on the Growing up in Ireland (a nationally representative large longitudinal study of children and young people in Ireland), 5-year, caregiver main questionnaire [23]. The FFQ is divided into different categories including breads/pastas, cake/confectionary, dairy, sugar, sweetened beverages, fruit, vegetables, spreads, and meat/fish and asked mothers to report how frequently (e.g. ‘at least once per week’, ‘most days’, ‘once a day’) their child consumed each food product in the past year. For the 5 year follow-up, portion sizes for each item of the FFQ was allocated using the Food Standards Agency food portions sizes book [24]. McCance and Widdowson’s The Composition of Foods Seventh Edition, Integrated dataset [25] was used to assess habitual dietary intakes. For the 9–11 year old follow-up, portion sizes were allocated using the Irish Food Sources Database for Children aged 9–12 years old [26]. McCance and Widdowson, Composition of Foods, Integrated Dataset 2021, [27] was used for nutrient analysis.

Children’s dietary quality at 5 and 9–11 years old was assessed using the Healthy Eating Index-2015 (HEI-2015) [28]. The HEI is a measure for assessing whether a set of foods aligns with the Dietary Guidelines for Americans (DGA) [29]. The latest version, the HEI-2015 reflects the 2015–2020 DGA [30]. The HEI uses a density-based methodology (with the exception of fatty acids, which is a ratio of unsaturated to saturated fatty acids), therefore, foods are assessed as per 1000kcals, rather than absolute amounts. The HEI contains 13 components. Nine ‘Adequacy’ components which address the adequacy of the diet in terms of intakes of necessary food groups. This includes total fruits, whole fruits, total vegetables, greens and beans, total protein containing foods, seafood and plant proteins, whole grains, dairy, and fatty acids (ratio of poly- and monounsaturated fatty acids to saturated fatty acids). The other four foods are ‘Moderation’ components, addressing negative elements of dietary intake; refined grains, sodium, added sugars, and saturated fats. The scores are summed to yield a total score ranging from zero to 100, with a higher score indicating greater adherence to the DGA. The following graded approach can be used to help interpret HEI scores; Overall scores of 90 to 100, or component scores that are 90 to 100% of maximum score: A; Overall scores of 80 to 89, or component scores that are 80 to 89% of maximum score: B; Overall scores of 70 to 79, or component scores that are 70 to 79% of maximum score: C; Overall scores of 60 to 69, or component scores that are 60 to 69% of maximum score: D; and Overall scores of 0 to 59, or component scores that are 0 to 59% of maximum score: F [30]. However, the numerical score is more meaningful and grades should only be used alongside the numerical score [30]. The HEI scoring system uses cups as the method of measurement. To convert cups to gram equivalents the food patterns ingredient database [31] was used. Foods from the FFQ’s were mapped to the different food categories used within the HEI. The equivalent values for the HEI component ‘Added Sugars’ were determined slightly differently to the other nine components (which aligned well with Irish dietary intakes). As amounts of ‘Added Sugars’ can vary considerably between US foods and their equivalents on the market in Europe, McCance and Widdowson composition of foods database [27] was used to estimate the amount of sugar added, using maltose and sucrose. Calculation of the HEI was completed using bespoke scripts written in Python programming language.

Mothers provided self-reported estimates of their child’s pubertal status using standardised Tanner staging figures (from Stage 1 to 5) relative to their secondary sex characteristics [32]. Boys were assessed by their corresponding stage of pubic hair distribution, while girls were assessed by their corresponding stage of pubic hair distribution and breast development. Preadolescence was defined as Tanner Stage 1 for pubic hair distribution and breast development.

Breastfeeding exposure was recorded at the 6 months, 2 and 5 year follow-up. At these timepoints, mothers reported if they had ever breastfed. At the 2-year and 5-year follow-up, mothers reported the age (weeks) their infant commenced solid food. A variable was created to indicate if their child had started solid food as per European [33] and national guidelines .

Continuous data were tested for normality using the Kolmogorov–Smirnov test and visual inspection of histograms. Normally distributed variables were reported as mean and standard deviation (SD). Non-parametric variables were reported as median and interquartile range (IQR 25th–75th). Categorical variables were reported as n (%). Paired t-tests examined differences between appetitive traits at both timepoints, for those with matched CEBQ data (n = 167). To account for missing data, further analysis using linear mixed-effects modelling was completed to examine change in appetitive traits from 5 to 9–11-years old (n = 224). Time was used as the predictor, and child sex, breastfeeding exposure, maternal BMI at the 9–11-year-old follow-up and original RCT group were used as the fixed effects and covariate variables.

Pearson’s correlations were completed to examine relationships between child appetitive traits and HEI and total energy intake at 5 and 9–11-years-old. Those with a correlation p-value of <0.05 were further analysed using multiple linear regression. ‘Total energy’ was not evenly distributed and was log^10 transformed for correlation and regression analysis. Prior to linear regression analyses, assumptions were tested. Models were adjusted for child age at the timepoint being analysed, child sex, total energy (kcals) intake, breastfeeding exposure, whether complimentary feeding was introduced as per European and national timing recommendations or not and original RCT allocation group. Statistical analyses were completed using IBM Statistical Package for Social Sciences (SPSS) for Windows, version 24.0 (Armonk, NY: IBM, Corp).