TY - JOUR
T1 - Predicting obesity at adolescence from an early age in a Dutch observational cohort study
T2 - the development and internal validation of a multivariable prediction model
AU - Huizing, Arjan Henryk Jonathan
AU - Welten, Marieke
AU - van der Pal, Sylvia
AU - Schönbeck, Yvonne
AU - van Empelen, Pepijn
AU - Gaillard, Romy
AU - Jaddoe, Vincent W.V.
AU - van Buuren, Stef
N1 - Publisher Copyright:
© The Author(s) 2025.
PY - 2025/6/7
Y1 - 2025/6/7
N2 - Background: – Identifying children with a high risk of developing future obesity could enable timely targeted prevention strategies. The study’s objective was to develop prediction models that could detect if young children at very early age, from birth to age six, have an increased risk of being obese in early adolescence. Methods: – We analyzed a subset of data (N = 4,309) from the Generation R study, a population-based prospective cohort study of pregnant women and their children from fetal life to young adulthood in the Netherlands. Parental, household, and birth/child characteristics were considered as predictors. We developed separate models for children at age zero (three months), two, four, and six years that predict obesity at age 10 to 14 years. Per age we fitted an optimal prediction model (full model) and a more practical model with less predictors (restricted model). For the development of the prediction models we used regularized regression models with a least absolute shrinkage and selection operator (LASSO) penalty to avoid overfitting. Results: – Parental body mass index (BMI), parental education level, latest child BMI measurements, ethnicity of the child, breakfast consumption, cholesterol, and low-density lipoprotein (LDL) of the child were included as predictors in all models when considered as candidate predictor. The models for all age groups performed well (lowest area under the curve (AUC) 0.872 for the age 0 restricted model), with the highest performance for the 6-year model (AUC 0.954 and 0.949, full and restricted model). Sensitivity and specificity of models varied between ages with ranges 0.80–0.90 (full model); 0.79–0.89 (restricted model) and 0.80–0.88 (full model); 0.79–0.87 (restricted model). Conclusions: – These obesity prediction models seem promising and could be used as valuable tools for early detection of children at increased risk of being obese at adolescence, even at an early age.
AB - Background: – Identifying children with a high risk of developing future obesity could enable timely targeted prevention strategies. The study’s objective was to develop prediction models that could detect if young children at very early age, from birth to age six, have an increased risk of being obese in early adolescence. Methods: – We analyzed a subset of data (N = 4,309) from the Generation R study, a population-based prospective cohort study of pregnant women and their children from fetal life to young adulthood in the Netherlands. Parental, household, and birth/child characteristics were considered as predictors. We developed separate models for children at age zero (three months), two, four, and six years that predict obesity at age 10 to 14 years. Per age we fitted an optimal prediction model (full model) and a more practical model with less predictors (restricted model). For the development of the prediction models we used regularized regression models with a least absolute shrinkage and selection operator (LASSO) penalty to avoid overfitting. Results: – Parental body mass index (BMI), parental education level, latest child BMI measurements, ethnicity of the child, breakfast consumption, cholesterol, and low-density lipoprotein (LDL) of the child were included as predictors in all models when considered as candidate predictor. The models for all age groups performed well (lowest area under the curve (AUC) 0.872 for the age 0 restricted model), with the highest performance for the 6-year model (AUC 0.954 and 0.949, full and restricted model). Sensitivity and specificity of models varied between ages with ranges 0.80–0.90 (full model); 0.79–0.89 (restricted model) and 0.80–0.88 (full model); 0.79–0.87 (restricted model). Conclusions: – These obesity prediction models seem promising and could be used as valuable tools for early detection of children at increased risk of being obese at adolescence, even at an early age.
KW - Early age
KW - Generation R study
KW - Obesity
KW - Prediction model
UR - http://www.scopus.com/inward/record.url?scp=105007450265&partnerID=8YFLogxK
U2 - 10.1186/s12887-025-05661-1
DO - 10.1186/s12887-025-05661-1
M3 - Article
AN - SCOPUS:105007450265
SN - 1471-2431
VL - 25
JO - BMC Pediatrics
JF - BMC Pediatrics
IS - 1
M1 - 465
ER -