Relative BMI: A machine-driven model for adolescent obesity severity classification


Creative Commons License

Mohamed A., Kara Gülay B., Cowan P., Velasquez-Mieyer P.

9th International Conference on Obesity and Chronic Diseases, Massachusetts, Amerika Birleşik Devletleri, 5 - 07 Kasım 2025, ss.7-8, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Doi Numarası: 10.17756/jocd.2025-suppl1
  • Basıldığı Şehir: Massachusetts
  • Basıldığı Ülke: Amerika Birleşik Devletleri
  • Sayfa Sayıları: ss.7-8
  • Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

Background: Body mass index (BMI) percentiles are widely used to assess obesity in adolescents, but they often misclassify

adiposity, particularly in moderate to high-risk groups.

Objective: To derive and validate sex-specific relative BMI (rBMI) cut-offs using decision tree models to improve adolescent

adiposity classification compared to BMI percentile and receiver operating characteristic (ROC)-based rBMI methods.

Methods: We analyzed data from 567 U.S. adolescents (ages 11 - 19) with DXA-derived body fat percentage (BF%) as the

reference standard. Classification and regression tree models were used to generate sex-specific rBMI cut-offs for four adiposity

categories: normal weight, overweight, obese, and severely obese. Model performance was evaluated using accuracy, kappa, Matthew’s

correlation coefficient (MCC), sensitivity, specificity, balanced accuracy, and F1-score. Performance was benchmarked

against BMI percentile and ROC-based rBMI cut-offs.

Results: The tree-derived rBMI model achieved the highest overall accuracy (0.720), kappa (0.625), and MCC (0.627). It outperformed

alternative methods across most adiposity categories, with substantial improvements in classifying overweight (F1 =0.572), obese (F1 = 0.636), and severely obese (F1 = 0.808) adolescents. The model showed better alignment with BF% defined

categories than BMI percentile and ROC-based rBMI, which tended to underestimate high-risk individuals.

Conclusion: Tree-derived rBMI cut-offs provide a statistically and clinically robust method for classifying adolescent adiposity.

This approach enhances early identification of at-risk youth and can be integrated into pediatric screening to guide targeted

interventions.