Infant BMI trajectories as early risk markers of poor psychosocial health in preadolescence | BMC Public Health

Study design and population

This research utilized data from the Healthy Growth Study, which is a large cross-sectional study comprising n = 2665 schoolchildren aged 9–13 years old, attending the fifth and sixth grades of primary schools located in municipalities within Attica, Aitoloakarnania, Thessaloniki and Iraklio in Greece [14]. Recruitment was conducted between 2007 and 2009. The Healthy Growth Study protocol was approved by the Greek Ministry of National Education and the study was conducted in accordance with Declaration of Helsinki. All procedures involving human subjects were approved by the Ethics Committee of Harokopio University of Athens (16/19.12.2006). Recruitment procedures and study methods have previously been described in detail [14]. Briefly, sampling of the schools was random, multi-stage and stratified by parents’ educational level and total population of students attending schools within these municipalities. Written informed consent was obtained from all parents before data were collected. All participating children underwent a physical examination by an experienced paediatrician and children’s health status was assessed using a standardised checklist. Information regarding children’s medical history was collected from their parents via a standardised questionnaire and paediatric records were obtained to acquire data relating to perinatal and postnatal growth. Infants born very preterm (< 32 weeks gestation), and infants born with a very low birth weight (< 1500 g) were excluded from this analysis (n = 35).

BMI and weight status

Weight and height/length data, measured monthly during the first year of life, were retrospectively obtained from children’s paediatric health records. BMI (weight[kg]/Height[m]²) from one to 12 months of age was calculated for inclusion in trajectory analysis. In childhood, height and weight were measured by a trained researcher. BMI was calculated and was categorised as underweight, normal weight, overweight and obese in accordance with the International Obesity Task Force cut-off points [15].

Emotional functioning

Emotional functioning was self-reported via the Dartmouth Cooperative Functional Health Assessment charts/World Organization of Family Doctors (COOPS/WONCA), which measures various aspects of functional status [16, 17]. The COOPS/WONCA was initially translated from English into Greek, followed by a verbatim back-translation into English by two independent translators. The original English version, the translated Greek version, and the back-translated English version were then compared following translation validation methods similar to those outlined by Sperber [18]. A high level of agreement across the three versions was observed in terms of accuracy, consistency and meaning. Emotional functioning was assessed by the following question: “During the past two weeks, how much were you pre-occupied with emotional problems such as feeling anxious, depressed, irritable or downhearted and sad?’. Response options included ‘1 = not at all’, ‘2 = slightly’, ‘3 = moderately’, ‘4 = quite a bit’ and ‘5 = extremely.’ In accordance with Hoare et al. [19], we defined emotional functioning as ‘good’ if participants responded 1 or 2, and ‘poor’ if participants responded ≥ 3. The COOPS/WONCA charts have been translated into over 19 languages [16], and psychometrically validated in various populations [20, 21]. However, specific validation for the Greek language and population has not been conducted. Despite this, the charts are widely used by researchers and clinicians in Greece to assess children’s emotional functioning, owing to their ease of use and simplicity.

Self-esteem

Self-esteem was assessed using the Greek version of the 10-item Rosenberg Self Esteem Scale (RSES), a widely used and validated self-report instrument for Greek school children [22, 23]. Each item is scored on a four-point Likert scale and summed to create a total score ranging from 0 to 30, with higher scores indicating greater self-esteem. Although there is no standardized method of classification, a RSES score < 15 is generally accepted within the literature as identifying individuals with low self-esteem. Further research has shown the sensitivity, specificity, positive and negative predictive values of this threshold to be acceptable for use [24]. Therefore, we defined children as having “low” self-esteem if they scored < 15 and “normal” self-esteem if they scored ≥ 15.

Body image and dieting practices

Body image and dieting behaviour was assessed using the Children’s Eating Attitudes Test (ChEAT), which is a validated self-report instrument for measuring one’s level of disordered eating behaviour [25]. This questionnaire consists of 14 items regarding body image perception, dieting practices, food preoccupation, food awareness and social pressure to eat. Each item is scored on a six-point Likert scale, with response options ranging from 1 = never to 6 = always. The current study analysed scores from the body image and dieting subscales only, which range from 3–18, with higher scores indicating greater frequency of body image concern and dieting behaviour. The ChEAT was translated and back translated from English into Greek using the same methods employed for the COOPS/WONCA. The translation was validated in accordance with Sperber [18], achieving a high level of agreement with regard to accuracy, consistency and meaning.

Restrained eating

Restrained eating refers to the intentional restriction of dietary intake to control weight and was assessed using the Greek version of the Dutch Eating Behaviour Questionnaire (DEBQ) for children aged 7–12 years [26]. The DEBQ includes 10 items that measure dietary restraint, such as “how often do you refuse food or drink offered because you are concerned about your weight?” and “when you have eaten too much, do you eat less than usual the following days?”. Each item is scored on a five-point Likert scale, with response options ranging from 1 = never to 5 = very often. Item scores were summed to create a restrained eating score ranging from 10–50, with higher scores indicating greater dietary restraint.

Other variables

Birth weight (grams), gestational age (weeks) and the duration of exclusive breast feeding (months) were obtained from paediatric health records. Maternal years of education were reported by parents and were stratified into three categories: less than 9 years (equivalent to junior high school degree); 9–12 years (equivalent to having a high school degree); and more than 12 years (equivalent to having a college, university, or post-graduate education). Tanner stage, an index of biological maturation, was determined by female paediatricians, who thoroughly inspected breast development in girls and genital development in both girls and boys.

Statistical analysis

BMI trajectory modelling

We used group-based trajectory modelling (GBTM) to delineate and characterize patterns of BMI among participants during the first 12 months of life. Unlike traditional models that typically assume a single, uniform growth trajectory for the entire population, GBTM captures the variability in growth patterns by identifying distinct subgroups with similar trajectories of change. This makes GBTM particularly well-suited for accommodating the inherent heterogeneity in BMI development.

Trajectories were modelled using raw BMI scores rather than standardized z-scores due to the limitations in using z-scores based on cross-sectional reference data to assess longitudinal changes [27,28,29,30,31,32]. Specifically, changes in BMI z-scores may not accurately reflect physiological growth because they are influenced by the distribution of BMI in the reference population rather than the child’s actual growth trajectory. Additionally, within-child variability in BMI z-scores is dependent on baseline adiposity, complicating the comparison of growth among infants with varying baseline BMIs [27, 29,30,31,32]. For example, a small increase in BMI can lead to a large change in z-score for an infant with a low baseline BMI, while the same increase results in a much smaller z-score change for an infant with a high baseline BMI. Accordingly, it has been proposed that raw BMI values provide a better indication of longitudinal changes, particularly when analyzing short term-variability and periods of dynamic growth [27, 31]. Therefore, we modeled trajectories during the first year of life using raw BMI values.

After excluding participants with missing or implausible anthropometric data during infancy (n = 774), and those with less than two measures of infant BMI (n = 60), a total of n = 1778 participants remained for inclusion in trajectory analyses. Anthropometric data was considered implausible if the weight-for-age, height-for-age or BMI-for-age z-score of a participant was greater than 5 or -5 at any given time point. Trajectories were modeled in STATA version 16.0 (StataCorp, College Station, Texas) using a plug-in developed by Jones and Nagin [33]. Using a stepwise approach, we ran models with one to five classes, fitting a quintic (fifth order) polynomial function for each class. If the order of the polynomial was not significant, we reduced the polynomial order until significance was reached (p < 0.05). The optimal model was selected based on the following fit indices: Bayesian Information Criterion (BIC), Akaike’s Information Criterion (AIC), group posterior probability, entropy, trajectory sample size, and interpretability. Further details regarding model selection are provided in the online supplement.

Descriptive and association analyses

Participant characteristics were compared across trajectory groups using the Kruskal-Wallis test with Dunn’s multiple comparisons or the chi-square test as appropriate. Continuous variables are presented as median and interquartile range (IQR) and categorical variables are presented as numbers (n) and percentages (%). Emotional functioning and self-esteem were treated as binary outcome and body image dissatisfaction, dieting behaviour and restrained eating outcomes were categorized into tertiles of increasing severity.

Binary and ordinal logistic regression models were used to determine the effects of infant BMI trajectories on emotional functioning, self-esteem, dieting, body image concern and restrained eating in childhood. Adjustments were made for age, sex, birth weight, gestational age, duration of breastfeeding, Tanner stage and maternal education. Sex differences were explored using interaction terms and stratified analyses. Odds Ratio (OR) and 95% confidence intervals (95% CIs) are presented.

Further supplementary analyses estimated cross-sectional associations between childhood weight status and psychosocial outcomes. All analyses were performed in STATA version 16.0 (StataCorp, College Station, Texas) and statistical significance was defined at a p-value <0.05 for main effects and <0.1 for interaction effects.