Machine learning algorithms for predicting glycemic control and weight loss outcomes in GLP-1 receptor agonist users

Area of Science:

• Biomedical informatics
• Pharmacogenomics
• Metabolic disease research

Background:

• Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are effective for type 2 diabetes and obesity.
• Significant variability exists in patient responses to GLP-1 RAs.
• Predicting individual treatment success remains a challenge.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting glycemic control and weight loss outcomes in patients initiating GLP-1 RA therapy.
• To identify key patient features influencing treatment response.
• To leverage real-world data for personalized medicine approaches.

Main Methods:

• Retrospective cohort study using the All of Us Research Program data.
• Developed and validated multiple ML models (e.g., Random Forest, XGBoost) using 10-fold cross-validation.
• Employed SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Ensemble ML models (RF, XGBoost) showed high accuracy (≈0.89-0.90) and discrimination (AUC ≈0.94) for predicting weight loss.
• ML models achieved modest accuracy (≈0.73) and discrimination (AUC ≈0.79) for glycemic control.
• Key predictors for weight loss included baseline BMI and body weight; for glycemic control, they included diabetes duration, baseline HbA1c, and concurrent medication use (sulfonylureas, insulin).

Conclusions:

• Machine learning, especially tree-based ensemble methods, shows significant potential in predicting GLP-1 RA treatment response.
• Explainable AI integrated with real-world data can personalize GLP-1 RA therapy selection.
• Identification of patient subgroups most likely to benefit from GLP-1 RAs can be enhanced.