Data-driven prioritization of high-risk individuals for weight loss interventions

Area of Science:

• Obesity Medicine
• Cardiovascular Health
• Machine Learning in Healthcare

Background:

• New obesity medications show promise but lack tools for targeted patient selection based on complication risk.
• Current methods, primarily Body Mass Index (BMI), do not fully capture individual risk for obesity-related comorbidities.

Purpose of the Study:

• To develop and validate a machine learning-based risk prediction model (OBSCORE) for identifying individuals at high risk of obesity-related complications.
• To provide a framework that complements BMI for prioritizing treatment in individuals with overweight or obesity.

Main Methods:

• Utilized a machine learning framework on a large population-based sample (~200,000 individuals with BMI > 27 kg/m²) to identify key risk features.
• Developed the integrated OBSCORE model to stratify individuals into distinct risk groups for 18 obesity complications over a 10-year period.
• Validated the model's generalizability across diverse ancestral populations and within a clinical trial (SURMOUNT-1).

Main Results:

• Identified 20 informative features beyond BMI that predict future onset of 18 obesity complications.
• OBSCORE successfully stratified individuals into risk groups, demonstrating differential 10-year incidence rates for outcomes like cardiovascular mortality (e.g., 5.7% in highest risk vs. 0.1% in lowest).
• The model showed generalizability in independent populations, and in the SURMOUNT-1 trial, predicted risks decreased post-treatment with tirzepatide, with similar weight loss across risk strata.

Conclusions:

• OBSCORE offers a robust framework for prioritizing high-risk individuals with overweight or obesity for intervention.
• The model enhances clinical decision-making by providing a more nuanced risk assessment than BMI alone.
• OBSCORE's validation across populations and in a clinical trial setting supports its utility in real-world obesity management.