Power and sample size calculations for discrete bounded outcome scores

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Methods

Background:

• Bounded outcome scores (BOS) present challenges for traditional statistical analyses in randomized trials.
• Standard methods struggle with the J- or U-shaped distributions common in BOS.
• Calculating statistical power for BOS is complex due to the failure of common location-shift alternatives.

Purpose of the Study:

• To develop and evaluate a parametric method for power and sample size calculations in randomized trials with BOS.
• To address the limitations of existing methods when analyzing BOS adjusted for covariates.
• To provide a robust framework for power calculations with complex BOS distributions.

Main Methods:

• A parametric approach assuming a logit-normal distribution for the true BOS, allowing covariate adjustment.
• A two-step power calculation procedure: conditional power based on covariates, then marginal power via Monte Carlo integration.
• Simulation studies to assess the method's performance and application to real-world data (ECASS-1 stroke study).

Main Results:

• The proposed parametric method effectively calculates power and sample size for BOS.
• The two-step procedure accurately estimates marginal power by averaging conditional power.
• The method demonstrates applicability and robustness in simulations and a clinical study.

Conclusions:

• The developed parametric approach offers a reliable solution for power and sample size calculations with bounded outcome scores.
• This method enhances the statistical rigor of randomized trials involving complex outcome data.
• The findings are crucial for optimizing clinical trial design and resource allocation.