Compiled data for MULT Allometric Body Mass Index Growth charts

Sociodemographic and anthropometric data from birth to 22 years old were collected from the population surveys:Young Lives: an International Study of Childhood Poverty (YL), Millennium Cohort Study (MCS), Epidemiological Health Investigation of Teenagers in Porto (EPITeen), and Longitudinal Study of Nutritional Assessment of Adolescents (ELANA), which together accounted for 32,162 individuals. However, individuals with missing data (age, weight, height, or socioeconomic indicator), follow-up losses, biologically implausible values, classified as height-for-age z-score below -6 SD or above +6 SD, or BMI-for-age z-score below -5 SD or above +5 SD, measurement errors (children whose height decreased over time), and individuals not meeting the following eligibility criteria were excluded: a) assessed in all three follow-ups for ELANA HS or at least four follow-ups in the other surveys; b) ethnicity classified into one of five categories: Caucasian, Indian, Black, Asian (East or Southeast Asian), or Native Peruvian. Additionally, For the construction of the ABMI growth charts, with the aim of having a representative sample of a healthy population without the influence of a high number of overweight individuals, the following exclusions were applied:a) BMI values with fewer than 10 observations per age in months were excluded.b) Weight measurements considered as outliers, calculated as weight-for-age z-scores below -2 SD or above +2 SD based on WHO reference values for children under five years old and based on the weight of the sample itself for those over five years old.c) BMI measurements considered as outliers, calculated as values below -2 SD or above +2 SD from the expected individual BMI. The expected BMI was calculated for each individual’s BMI value through a linear mixed-effects model.As studies suggest that there is no allometry for children under five years of age, only anthropometric measurements from individuals aged five years and older were selected for the construction of the ABMI-for-age growth charts. In the end, 17,447 individuals remained in the dataset.The construction of the ABMI reference, stratified by sex, were constructed using statistical software including R and RefCurv, which features a Python-based graphical interface and an R-based statistical engine. The LMS method, pioneered by Cole and Green in 1992, was employed to estimate these reference values. This method utilizes three key curves: L (skewness), M (median), and S (coefficient of variation). Smoothing parameters for the L, M, and S curves were determined using Generalized Additive Models for Location, Scale, and Shape (GAMLSS), introduced by Rigby and Stasinopoulos in 2005. GAMLSS employs a regression framework where all parameters of the conditional distribution of the response variable are modeled using explanatory variables. To stabilize variance and approximate normality of the data, the Box-Cox transformation—a statistical technique—was applied within the Cole Green Box-Cox framework. Growth charts for both sexes were then adjusted using penalized splines. The Bayesian Information Criterion (BIC) was employed to select optimal degrees of freedom for the ABMI growth charts, thereby avoiding overfitting. "Worm plots," introduced by Van Buuren and Fredriks in 2001, were utilized to assess appropriate levels of smoothing for height parameters and to evaluate model fit. Regarding cutoff points, four proposed options were estimated by applying BMI values of 17 kg/m², 25 kg/m², and 30 kg/m² to ages 17, 18, 19, and 20 years, delineating percentiles for underweight, overweight, and obesity.