An interpretable machine-learning model identifies obesity-related metabolic complications in children attending a tertiary obesity clinic

Abstract

Childhood obesity-related metabolic complications are common in Northwest China, yet region-specific evidence and risk-stratification tools are lacking. In a single-visit cross-sectional study of 131 children (3–18 years) with overweight or obesity attending a Gansu maternal and child-care hospital, obesity-related metabolic complications (ORMC) were present in 31.3% (dyslipidaemia 16.0%, hyperuricaemia 9.2%, elevated blood pressure 7.6%). An interpretable model combining LASSO-selected adiposity measures (body mass index, neck circumference, body fat percentage) with penalised logistic regression achieved moderate, internally cross-validated discrimination (AUC 0.72, 95% CI 0.62-0.82) that awaits external validation, comparable to a random forest (0.73). Screen time was the strongest modifiable behavioural correlate. Routine metabolic screening and simple body-composition measures should be embedded in paediatric obesity clinics, with screen-time reduction as a counselling priority.