Accuracy of BMI correction using multiple reports in children.

Area of Science:

• Pediatric obesity research
• Biostatistics
• Public health

Background:

• Inaccurate self- or proxy-reported height and weight can skew body mass index (BMI) and obesity estimates in children.
• Previous BMI correction models primarily used adult samples.

Purpose of the Study:

• To compare the accuracy of BMI correction models in children using child reports, parent reports, or both.
• To evaluate whether separate prediction models for height and weight are more accurate than a single BMI model.

Main Methods:

• Analyzed height and weight data from 475 children (child reports, parent reports, objective measurements).
• Evaluated two approaches: separate linear models for height/weight vs. a single linear model for BMI.
• Used leave-one-out validation to compute prediction accuracy, comparing models using root mean squared error (RMSE) and sensitivity/specificity for obesity indicators.

Main Results:

• Models incorporating both child and parent reports showed the best fit (adjusted R(2) 67.1-88.3%).
• Separate prediction models for height and weight resulted in the least BMI prediction error compared to single BMI models or uncorrected reports.
• Cross-validated RMSE indicated parent-only reports were preferred for males and females over child-only or combined reports.

Conclusions:

• Objective measurements are superior for BMI accuracy, but correction models offer a practical solution when infeasible.
• Combining multiple reports (child and parent) in a validation subsample can create cost-effective BMI correction models.
• Corrected BMI estimates are closer to objective measurements than uncorrected self- or proxy-reports.