Inference in group sequential designs with causal mechanisms: implications for power and mediation analysis

Area of Science:

• Biostatistics
• Clinical Trial Design
• Causal Inference

Background:

• Group sequential designs allow early trial stopping with statistical rigor.
• Mediation analysis properties are unknown in group sequential designs, despite common use in psychological trials.

Purpose of the Study:

• Evaluate analysis techniques for mediation in group sequential designs.
• Assess bias of estimators for total, direct, and indirect effects.
• Compare maximum likelihood estimators under early stopping scenarios.

Main Methods:

• Simulation study of a group sequential design with one interim analysis.
• Analysis of continuous outcomes with a continuous mediating variable.
• Examination of null hypothesis rejection rates and estimator bias.

Main Results:

• Reduced power in group sequential designs when sample size ignores causal mechanisms.
• Maximum likelihood estimator is unbiased only for the mediator-outcome path.
• Conditional maximum likelihood estimator shows less bias for total and direct effects.

Conclusions:

• Mediation analysis requires specific consideration in group sequential designs.
• Sample size calculations must account for causal mechanism variability.
• Multiple estimators may be needed, acknowledging potential bias in indirect effect estimation.