Bayesian Hierarchical Models With Calibrated Mixtures of g-priors for Assessing Treatment Effect Moderation in...

Area of Science:

• Biomedical Science
• Statistics
• Psychiatry

Background:

• Assessing treatment effect moderation is crucial for personalized medicine.
• Individual participant-level data meta-analysis (IPD-MA) is a robust method but faces challenges like high between-study variability.
• Existing Bayesian shrinkage methods are suboptimal for IPD-MA due to inflexible priors.

Purpose of the Study:

• To propose novel Bayesian priors for IPD-MA to enhance moderation effect estimation.
• To address limitations of traditional methods in heterogeneous study settings.
• To improve efficiency and reduce risks in personalized intervention assessments.

Main Methods:

• Developed calibrated mixtures of g-priors tailored for IPD-MA.
• Incorporated study-level calibration and moderator-level shrinkage for flexible prior specification.
• Conducted simulation studies to compare performance against existing Bayesian shrinkage methods.

Main Results:

• Calibrated mixtures of g-priors demonstrated equivalent or superior performance in estimating moderation effects.
• The proposed methods showed particular benefits in scenarios with high between-study variability, model sparsity, and weak moderation.
• Effectiveness illustrated in a real-world application for major depressive disorder treatments.

Conclusions:

• Calibrated mixtures of g-priors offer an effective and flexible Bayesian approach for moderation analysis in IPD-MA.
• These methods improve upon existing techniques, especially in challenging real-world data scenarios.
• The approach provides valuable insights for personalized treatment strategies in biomedical research.