Power and Sample Size Determination for Multilevel Mediation in Three-Level Cluster-Randomized Trials.

Area of Science:

• Biostatistics
• Health Services Research
• Psychometrics

Background:

• Mediation analysis is crucial for understanding treatment pathways and mechanisms of action.
• Existing guidance for planning mediation studies with complex hierarchical or clustered structures is limited.
• Evaluating mediation effects alongside total effects in experimental designs is increasingly common.

Purpose of the Study:

• To provide methods for computing statistical power to detect mediation effects in three-level cluster-randomized designs.
• To address the need for guidance in planning mediation studies with multi-tiered hierarchical structures.

Main Methods:

• Development of power computation methods for three-level cluster-randomized designs.
• Assessment of the proposed methods through simulation studies.
• Application of methods using the R package PowerUpR and its Shiny application for clinic-randomized studies.

Main Results:

• The study provides practical tools for power calculations in complex mediation designs.
• Demonstration of application in a three-level clinic-randomized study context.
• The R package PowerUpR facilitates the implementation of these power computation methods.

Conclusions:

• The developed methods and tools support researchers in planning robust mediation studies with clustered data.
• This work addresses a critical gap in the methodology for evaluating mediation in complex hierarchical designs.
• Facilitates more accurate and efficient study design for mediation research in nested data structures.