Mediation analysis with principal stratification

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Psychiatric Research

Background:

• Mediation analysis in randomized clinical trials (RCTs) assesses treatment effects via intermediate variables.
• Standard methods assume sequential ignorability, which may not always hold.
• Principal stratification (PS) offers an alternative framework for causal inference.

Purpose of the Study:

• To apply and evaluate the principal stratification (PS) approach for estimating direct treatment effects in RCTs.
• To assess the impact of the sequential ignorability assumption in mediation analysis.
• To examine the sensitivity of PS estimates to variance heterogeneity and model mis-specification.

Main Methods:

• Utilized the principal stratification (PS) framework to define and estimate direct effects.
• Employed a Bayesian estimation procedure for the PS model.
• Conducted simulation studies to assess robustness and sensitivity.
• Applied the methodology to two real-world psychiatric clinical trial examples.

Main Results:

• The PS approach successfully estimated direct treatment effects without the sequential ignorability assumption.
• Results demonstrated robustness to the assumption of homogeneous variances across principal strata.
• Simulations indicated that the magnitude of direct effects estimated by PS is sensitive to model mis-specification.

Conclusions:

• The principal stratification (PS) approach provides a viable alternative for mediation analysis in RCTs when sequential ignorability is questionable.
• Careful model specification is crucial for reliable estimation of direct treatment effects using the PS method.
• The PS approach offers valuable insights into treatment mechanisms in psychiatric research.