Predicting obesity in adolescents from birth to age 6 years

Aims –Identification of children with an increased risk of future overweight or obesity enables targeted preventive strategies. The objective of this study is to develop a screening algorithm that can detect young children with a high risk on overweight or obesity at early adolescence (age 10 – 14 years). Materials and methods – We analyzed a subset of data from Generation R (N=3157), a population-based prospective cohort study. The following types of predictive information were considered: 1) Maternal BMI; 2) Maternal age; 3) Paternal BMI; 4) Pregnancy characteristics; 5) Child characteristics; 6) SES-indicators; 7) Family and child’s lifestyle; 8) Child lipid measurements. BMI Z-scores are predicted at age 0.25 years (birth model), 2.5 years (two-year model), 4.5 years (four-year model) and 6.5 years (six-year model). Two models were fitted, one generic model and one restricted to variables currently available in Dutch child care organizations. The data was split into a training and a testing set. We compared a variety of methods on our training data, and found that regularized regression model with lasso consistently performed well. We used multiple imputation to address missing values and fitted our final model on all imputed data sets and averaged predictions. Results – Our best performing model to predict overweight or obesity at 10 – 14 years of age was the model with variables at age 6 (out-of-sample R2=0.568). The most important variables in our models were: child BMI, BMI of both parents, ethnicity of the child and the age at which the child was introduced to eating fruit and vegetables. Conclusion – Our models perform well in predicting overweight or obesity at 10 – 14 years of age. These models can be used at earlier ages by preventive child health care to predict and prevent obesity. During the presentation we will show a live demo of a prediction tool.