Precision Mental Health: Predicting Heterogeneous Treatment Effects for Depression through Data Integration

Area of Science:

• Clinical Psychology
• Biostatistics
• Pharmacology

Background:

• Individualized treatment selection for depression is complex due to numerous therapeutic options.
• Integrating data from multiple randomized controlled trials (RCTs) can reveal treatment effect heterogeneity.
• Extrapolating findings from RCTs to real-world patient populations is essential for clinical practice.

Purpose of the Study:

• To develop and validate a novel two-stage meta-analytic method for predicting conditional average treatment effects (CATEs) in target patient populations.
• To enhance individualized treatment allocation by leveraging CATE distributions across multiple RCTs.
• To generate prediction intervals for CATEs in new settings, accounting for both within-study and between-study variability.

Main Methods:

• A two-stage meta-analytic approach was developed to predict CATEs.
• First-stage models incorporated parametric regression, causal forests, or Bayesian additive regression trees (BART).
• The method was validated via simulations and applied to RCTs comparing duloxetine and vortioxetine for depression.

Main Results:

• The method successfully generated 95% prediction intervals for CATEs in target patient profiles.
• Analysis of depression RCTs showed limited evidence of effect heterogeneity, except for potential age-related differences.
• CATE prediction intervals effectively captured broader uncertainty compared to traditional study-specific confidence intervals.

Conclusions:

• The proposed meta-analytic method enhances personalized treatment allocation for depression by predicting CATEs in target populations.
• Prediction intervals provide a more comprehensive measure of uncertainty, crucial for clinical decision-making.
• The approach offers a valuable tool for integrating evidence across RCTs to inform individual patient care.