Truace
Subscribe

External validation and application of machine learning models for obesity in Chinese children and adolescents per Chinese and WHO growth standards

Background Childhood and adolescent obesity is a growing global public health challenge, underscoring the need for reliable tools to identify individuals at high risk. We aimed to develop, validate, and deploy machine learning (ML) models for predicting obesity in children and adolescents aged 3-18 years using multi-domain risk factors per Chinese and World Health Organization (WHO) growth standards, with an online tool for clinical and public health translation. Methods We conducted a population-based cross-sec…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2058· cite & verify
External validation and application of machine learning models for obesity in Chinese children and adolescents per Chinese and WHO growth standards
The child-welfare special : A suggested method of reaching rural communities ... U. S. Department of labor. Children's bureau by United States. Children's Bureau Bradley, Frances Sage, 1862-1949. Public domain

In brief

Researchers trained four machine learning models on survey data from 22,555 children and adolescents in Beijing to predict obesity, defined separately under Chinese and WHO growth standards. They then tested the best models on 5,343 participants from Tangshan, a separate population.

The work suggests that a short list of routinely available factors, including parental BMI, screen time and birthweight, can support risk identification and that the tools can be used online. Predictive accuracy was only moderate, so how much the tools would help in real clinical or public health practice remains to be shown.

Main points

  1. Training and internal validation used 22,555 participants in Beijing (2022, 2024); external validation used 5,343 participants in Tangshan (2020).
  2. Of 38 features across 5 domains, LASSO regression retained 5 for the Chinese standard and 12 for the WHO standard.
  3. Age, maternal BMI, paternal BMI, screen time and birthweight were consistent predictors under both standards.
  4. The neural network (AUROC 0.735) was best under the Chinese standard and XGBoost (AUROC 0.736) under the WHO standard.
  5. External validation AUROCs fell to 0.699 and 0.707, respectively.

The gain

Models built on data from Beijing and tested in Tangshan identified obesity in children aged 3-18 using only 5 to 12 features, and were released as online tools for clinical and public health use.

The rundown

The study was a population-based cross-sectional design covering ages 3-18. Obesity was defined with both Chinese and WHO growth standards, allowing the models to be compared across the two definitions.

Four ML models were compared using 28 performance metrics. SHapley Additive exPlanations were used to make the predictions more interpretable, and least absolute shrinkage and selection operator regression selected the key predictors from the 38 candidate features.

Authors report that the final models performed modestly but reliably. Performance dropped slightly in the Tangshan external dataset, and online prediction tools were deployed for translation into clinical and public health settings.

Sources

  1. Peer-reviewedJournal of Global Health2026-10-19

The debate