Presentation (CTI)

OBJECTIVES: This study aims to quantify how different levels of weight loss relate to clinical outcomes across obesity-related comorbidities, using large-scale real-world data and modern causal inference techniques.
METHODS: We analyzed linked electronic health records and claims from approximately 2 million U.S. adults with obesity (at least one measurement of BMI ≥30 kg/m² in 10 years) across multiple chronic indications, including respiratory and musculoskeletal conditions. A target trial emulation (TTE) framework was used to define treatment strategies, eligibility, and follow-up while addressing confounding and immortal time bias. Causal effects of weight-loss categories (0-5%, 5-10%, 10-15%, 15-20%, ≥20%) on condition-specific outcomes were estimated using double machine learning (DML) with cross-fitting. G-computation was used to decompose total effects into weight-loss-mediated and modality-specific components. Sensitivity analyses evaluated robustness under varying treatment adherence definitions, insurance-coverage constraints.
RESULTS: Across the evaluated indications, we characterized the relationship between weight loss and clinical outcomes, identifying patterns that varied by condition, outcome type, and patient characteristics. The results show which disease indications are set to benefit most from new obesity treatments, and what the target weight loss for such indications should be
CONCLUSIONS: This study highlights how TTE and DML can be integrated to estimate dose-response relationships from noisy, time-varying real-world data, providing methodological insight applicable across therapeutic areas.