The development of a dietary assessment tool to predict future obesity risk in young people

Worldwide prevalence of childhood and adolescent obesity continues to rise. It warrants prevention, but finite resources dictate targeted interventions. This research developed and evaluated an obesity risk algorithm, translated into a questionnaire and risk score to identify childhood communities at higher risk of obesity by early adolescence. A systematic review of children’s diet and adiposity outcomes found evidence for 24 potential predictors of future obesity. 20 predictors, including food and drink intakes and other factors at 10+ years, were matched to variables in a dataset from a UK birth cohort (Avon Longitudinal Study of Parents and Children). The data (n = 5,486) was randomly split, 75% for derivation of the algorithm and 25% for internal validation. Purposeful selection of covariates determined a predictive logistic regression model for adolescent obesity at 13+ years. Predictive metrics were run. Risk scores were based on β coefficients of the final model in the combined dataset. Evidence from 14 longitudinal childhood cohorts showed that foods and drinks which contributed to energy dense dietary patterns, plus some eating habits, health behaviours and familial factors, were associated with adverse adiposity outcomes. The final model had 9 predictive variables: Intake of vegetables, milk, dairy foods and snacks/treats, sugar sweetened beverage frequency, early puberty, mother’s overweight, child’s body satisfaction and active travel to school. In the derivation sample the model had good overall predictive ability (Brier score = 0.04), acceptable discrimination (AUROC = 0.76) and showed potential usefulness (PPV = 10%). Metrics were similar in the validation sample, showing reproducibility. The Children’s Obesity Risk Assessment (CORA) is the first predictive model of childhood obesity known to include detailed measures of diet. The model and risk score require external validation to demonstrate transportability to different populations. A discriminating and well calibrated model could help target obesity prevention interventions more effectively.