Keywords: BMI correction, bias, reporting errors, obesity
Errors in reported height and weight raise concerns about body mass index (BMI) and obesity estimated from self or proxy reports. Researchers have used BMI corrections models in adult samples. We evaluated the accuracy of BMI correction models among children. The Military Teenagers' Environment, Exercise and Nutrition Study collected child and parent reports, as well as objective measurements, of height and weight for 475 youth. Two correction approaches were evaluate: separate models for height and weight versus one model for BMI. Each approach considered models including child, parent or both reports, stratified by gender. Prediction accuracy was computed using leave-one-out validation. Models were compared using RMSE, sensitivity and specificity. Models that included both reports provided best fit relative to a model using either set of reports. BMI estimates from separate correction of height and weight had lowest prediction error, while cross-validated RMSEs favored a model that included only parent reports. In sum, linear correction models generated BMI estimates that improved upon child or parent reports. Future work will evaluate nonparametric or nonlinear correction model