Predicting Weight Outcomes From Obesity Medications in a Paediatric Population.

INTRODUCTION: There are limited paediatric studies comparing outcomes from different obesity medications (OMs) in real-world settings. The aim of this study is to describe real-world variability in outcomes and develop models to predict outcomes from OMs. METHODS: We examined electronic health record (EHR) data of patients (12-21 years) with obesity and without diabetes dispensed an OM from 2003 to 2025 in a paediatric healthcare system. We determined percent change in BMI percentile above the 95th percentile (%BMI) and for patients dispensed the medication ≥ 6 months (to allow for a therapeutic dose) used a ≥ 5% reduction in %BMI as the outcome for Catboost gradient-boosting machine learning. We included patient (socio-demographics, baseline %BMI, comorbidities, medications) and treatment (dose, duration, adherence, specialty visits) factors as predictors. RESULTS: A total of 595 patients were included (20% Hispanic, 27% Black, 49% public insurance, 62% with severe obesity). For patients dispensed Liraglutide, Phentermine, Phentermine/Topiramate and Semaglutide > 12 months (comparable duration to clinical trials), average percentage change in %BMI was 8.76, 10.90, 9.34 and 12.87, and 64%-80% had a ≥ 5% final %BMI reduction. Predictive models demonstrated AUROC ≥ 0.75. OM duration, baseline %BMI and number of specialty visits were associated. CONCLUSIONS: The OMs studied had similar outcomes, with Semaglutide demonstrating slightly better outcomes, but not all patients had successful outcomes. Predictive models can inform clinical decision-making about OMs based on individual characteristics. Future studies validating models prospectively are needed.