This cross-sectional observational study aimed to assess anthropometric measurements among the schoolgoing population in Vigo. The study was conducted during May–June 2009.

The study population included both male and female schoolgoers aged 11 through 17 years in the Vigo metropolitan area. The total schoolgoing population in Vigo was 10,747, with 60% attending state-subsidized schools and 40% public schools. The breakdown by school year was 2741 in the first year of Compulsory Secondary Education, 2789 in the second year, 2735 in the third year, and 2482 in the fourth year.

The sample size calculation was based on an estimated overweight prevalence of 17%, a 95% confidence interval, the ability to detect a difference of 3%, and a total schoolgoing population of 10,747 students. A total of 577 participants were to be recruited, as determined by a sample size calculator [18].

Cluster randomization was performed with schools as the sampling units, ensuring representation from each school year. In the sample, efforts were made to have each course represent 25% of the total.

The sampling procedure involved randomly selecting the initial school from a compiled list of both public and state-subsidized schools. Afterwards, consecutive schools were chosen until the predetermined sample size was achieved. In instances where a school declined participation, the subsequent school listed was then selected to ensure the continuation of the sampling process.

The study encompassed various variables for comprehensive assessment: age (years), sex (male and female), nationality/country of origin, and height (cm); tricipital, bicipital, subscapular, suprailiac, abdominal, pectoral, thigh, and leg skinfolds (mm); radial bistyloid, humeral biepicondyle, femoral biepicondyle bone diameters (cm); waist, hip, contracted arm, relaxed arm, head, wrist, and leg girths (cm); and impedance measurement. Protocols detailing the measurement procedures are available in Appendix 1, and the instruments used are listed in Appendix 2.

The classification of overweight and obesity was determined based on the bioimpedance measurement criteria of Mueller et al., using the 85th percentile as the cut-off point for each age. This study refers to the 5th to 95th percentiles of body fat percentage, derived from bioelectrical impedance, of a cohort of 678 children of different races (black and non-black) who were followed for 4 years [19, 20].

Before initiating fieldwork, researchers underwent training to standardize the procedure for obtaining anthropometric measurements. Selected schools were briefed about the study, and information leaflets and informed-consent forms were distributed to pupils. The study criteria were explained to school authorities, teachers, and parents/guardians.

Anthropometric data were collected by a research team during scheduled visits to schools. Measures were directly entered into an EXCEL spreadsheet, with each pupil assigned a unique code for anonymity. Pupils without signed consent forms or those with illnesses affecting anthropometric values were excluded. Information about the measurements was provided to families upon request.

A comprehensive statistical analysis was conducted, involving data cleaning, debugging, and the removal of implausible values. For qualitative variables, absolute frequencies and percentages were presented, while for quantitative variables, normality was assessed, and mean and standard deviation (SD) or median and 25th and 75th percentiles were reported as appropriate.

In the bivariate analysis, linear regression models adjusted for age and sex were employed for each parameter in the study. Graphical representations illustrated changes in anthropometric measures with increasing age for each gender (Fig. 1, supplementary file).

Anthropometric measures analyzed by age and sex. A Linear model considering height by age and sex. B Linear model considering weight by age and sex. C Linear model considering lean mass by age and sex. D Linear model considering fat mass by age and sex. E Linear model considering body water by age and sex. F Linear model considering humerus diameter by age and sex. G Linear model considering radio diameter by age and sex. H Linear model considering femur diameter by age and sex. I Linear model considering cephalic perimeter by age and sex. J Linear model considering contracted arm perimeter by age and sex. K Linear model considering arm perimeter by age and sex. L Linear model considering wrist perimeter by age and sex. M Linear model considering waist perimeter by age and sex. N Linear model considering hip perimeter by age and sex. O Linear model considering leg perimeter by age and sex. P Linear model considering pectoral skinfold by age and sex. Q Linear model considering bicipital skinfold by age and sex. R Linear model considering abdominal skinfold by age and sex. S Linear model considering suprailiac skinfold by age and sex. T Linear model considering thigh skinfold by age and sex. U Linear model considering leg skinfold by age and sex. V Linear model considering subescapular skinfold by age and sex. W Linear model considering tricipital skinfold by age and sex

To establish the predictive model, overweight (yes/no) was considered the outcome variable, according to the predefined criteria presented by Mueller et al., for each age/sex group and white population. A generalized additive model (GAM) logistic model was utilized to select predictors. Overweight served as the result and anthropometric measures, age, and sex acted as predictors. The optimal combination of variables was determined from the cross-validation technique, considering the area under the ROC curve (AUC) as the evaluation metric. This technique involves splitting the data set, such that the model is trained on several subsets of the data and evaluated on the remaining subset. This procedure is repeated several times, each time with a different combination of training and test sets. The AUC evaluation metric is calculated for each iteration. Finally, the results are averaged to obtain an overall evaluation of the model’s performance. This approach helps reduce the risk of overfitting and provides a more robust estimate of model performance on unseen data. In addition, it allows identifying the optimal combination of variables that maximizes the predictive capacity of the model, thus improving its generalization to new data sets. The R code is provided in Appendix 3.

The same methodology was applied to other routinely used pediatric care indices, such as BMI, adjusting for age and sex.

The diagnostic utility of the developed models was compared using the pROC software package [21]. For each model, the AUC with confidence intervals was calculated, along with sensitivity, specificity, positive and negative predictive values, true positive and true negative values, false positive and false negative values, accuracy, and positive and negative likelihood.

All analyses were conducted using R Studio statistical software package version 4.1.3 [22].