Prediction of overweight/obesity risk in adolescence using static and dynamic multi-exposure models

Abstract

Abstract Background Studying the risk factors for obesity development and timely identification of individuals at increased risk are key for obesity prevention. We aimed to assess the associations of multiple prenatal, infancy and childhood exposures with overweight/obesity at 13 years (y) and develop static and dynamic multiple-exposure prediction models. Methods In 4232 mother-child pairs from the Generation XXI birth cohort, we assessed, during pregnancy/infancy and childhood, 55 sociodemographic factors, anthropometric measures, lifestyle behaviors, and clinical exposures related to the mother and the child and categorized body mass index z-scores as normal weight and overweight/obesity at 13y. Static models for pregnancy/infancy, 4, 7 and 10y, and a dynamic model were fitted using Elastic Net and logistic regression and were compared in terms of predictive performance. Results Mothers with lower educational level and household income, no partner, overweight/obesity before and after pregnancy, excessive gestational weight gain, pregnancy complications and who were smokers during pregnancy had children with a higher risk of overweight/obesity at 13y. Children with higher BMI during infancy, overweight/obesity during childhood, an early introduction of solid foods (before 4 months), a higher consumption of soft drinks at 4y and females had a higher risk of overweight/obesity at 13y. We observed the lowest predictive performance for the pregnancy/infancy model (prediction error of 36%) and the highest predictive performance for the 10y and dynamic models (prediction errors of 17%). Conclusions We identified the most important predictors of overweight/obesity in adolescence, most of them modifiable, and able to be targeted in prevention strategies. Predicting overweight/obesity risk in adolescence might be suboptimal if only considering pregnancy/infancy data and improves as closer to the age of the outcome, with no additional benefit in using a complex dynamic model. Key messages • Prevention of overweight/obesity should start during prenatal life. • Children at higher risk of overweight/obesity may be identified years earlier using prediction models in clinical practice.