2.1 Study population and design
Study subjects were participants of the Cork BASELINE (Babies after SCOPE: Evaluating the Longitudinal Impact using Neurological and Nutritional Endpoints) Birth Cohort Study 18, a mother–infant prospective birth cohort study based in Cork, Ireland. It was initiated in 2008 as a follow-up to the SCOPE (Screening for Pregnancy Endpoints) Ireland study, a major multi-centre prospective pregnancy study involving primiparous low-risk women. 1,537 SCOPE participants consented for their infants to participate in the BASELINE study and during a second stream of recruitment, a further 646 infants were recruited after delivery from the postnatal wards of Cork University Maternity Hospital, with primiparous low risk women having a singleton pregnancy being the main inclusion criterion. Paediatric follow-up with in-person assessments were conducted at birth, 2, 6, and 12 months and at 2 and 5 years. Data on the child’s early-life environment, diet, health and development were recorded at each assessment. In a subgroup of children (n=591), body composition and bone mineral density (total body and lumbar spine) were measured at 5 years using DXA (GE Healthcare Lunar iDXATM). In total, 2,172 infants were recruited. 1,229 of the children had data at the five year assessment and was used for the purpose of this analysis; thus representing a 43% attrition rate.
2.2 Children’s lifestyle factors included in LCA
Eating behaviour at 5 years of age was assessed using the CEBQ. The CEBQ consists of 35 items scored on a 5-point Likert scale from 1 ‘never’ to 5 ‘always’. Items are assigned to eight subscales: Emotional Overeating (EOE), Food Responsiveness (FR), Enjoyment of Food (EF), Desire to Drink (DD), Emotional Undereating (EUE), Satiety Responsiveness (SR), Food Fussiness (FF) and Slowness in Eating (SE). Subscales represent two dimension; “food approach” (EOE, EF, FR, DD) and “food avoidance” (EUE, SR, FF, SE). All items are listed in Supplementary Table 1. The mean score was calculated for each subscale and expressed as z-scores. Physical activity and TV watching were assessed from the responses to questions regarding the time spent on average participating in active play and sitting still watching TV on weekdays and weekends. Dichotomous variables were created for active play <1 hour per day and ≥1 hour per day and TV watching <2 hours per day and ≥2 hours per day based on national and international guidelines and World Health Organization (WHO) recommendations on sedentary behaviour19. A dichotomous variable was created for sleep duration (<10 hours or ≥10 hours) as10-14 hours sleep has been recommended by the American Academy of Sleep Medicine as appropriate for children aged 5 years20.
2.3 Sociodemographic and maternal characteristics and early feeding
Maternal educational attainment, socioeconomic status (SES), marital status, smoking status, and activity levels were assessed at the 5 year assessment. SES was determined using the New Zealand Socioeconomic Index (SEI)21, with a variable created for SEI<24. A dichotomous variable was created for participation in physical activity other than walking and for TV watching with categories <2 and ≥2 hours per day. Maternal body mass index (BMI) was obtained according to standard operating procedures at the 5 year assessment and categorised as under/normal weight, overweight, or obese. Breastfeeding status was obtained at 2 months and categorised as any breastfeeding vs no breastfeeding. Timing of introduction of solids was assessed at 6 month and a dichotomous variable was created for introduction to solids <18 weeks.
2.4 Child anthropometric measurements
Weight, length, and waist circumference at 5 years were obtained according to standard operating procedures. Naked weight was measured using digital scales correct to the nearest 0.1 kg. Standing height was measured using a wall mounted stadiometer. Body composition (including fat mass and lean mass) and bone mineral density (total body and lumbar spine) was measured using DXA (GE Healthcare Lunar iDXATM). These measurements were used to calculate WHtR, BMI, FMI, and LMI (calculated as body weight, fat, and lean mass divided by square of height). Weight status was assessed according to the International Obesity Task Force BMI cut-offs, with the cut-offs for 5.5 years of age used22.
2.5 Statistical analysis
To identify clusters of children, we conducted LCA using variables concerning eating behaviour, physical activity, sedentary behaviour, and sleep as outlined above. We used the maximum likelihood robust estimator to account for missing data by full information maximum likelihood (FIML). This process approximates missing data by estimating a likelihood function for each individual based on variables that are present, such that all the available data points are used23. The optimal number of latent classes was identified based on six model-fit indices: Akaike information criterion (AIC), sample-size adjusted Bayesian information criterion (BIC), adjusted Bootstrap likelihood ratio test (BLRT), Lo-Mendell Rubin test (LMRT), entropy, and interpretability of the trajectories. Lower AIC and BIC values indicate a better model fit, while the BLRT and LMRT provide a p-value indicating whether a model with one less trajectory group (k-1 model) should be rejected in favour of a model with k trajectories 24. Entropy is a statistic that ranges from 0 to 1 with high values (>0.8) indicating that individuals are classified with confidence 25. Distinct classes were coded as a categorical variable (with k number of categories) and were named based on their visual appearance.
Associations between sociodemographic and maternal characteristics (educational attainment, socioeconomic index, marital status, smoking status, physical activity other than walking, TV watching, and weight status), child sex, and early feeding (breastfeeding at 2 months and timing of introduction to solids) and latent class membership were examined using multinomial logistic regression, with class membership the outcome of interest and the most commonly occurring class chosen as the reference category. Associations between class membership and weight status, FMI, LMI, and WHtR were examined using logistic and linear regression, with class membership as the independent variable. These models were further adjusted for maternal characteristics, sex, and early feeding. Analysis was conducted using Mplus version 826 and Stata version 1427.