Power and sample size calculation for stepped-wedge designs with discrete outcomes

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Stepped-wedge designs (SWD) are increasingly utilized in healthcare systems to evaluate process-of-care changes.
• Accurate sample size calculation is critical for ensuring adequate statistical power in SWD.
• Existing methods primarily focus on normal outcomes, with limited options for non-normal outcomes like binary endpoints common in healthcare.

Purpose of the Study:

• To propose a novel power calculation formula for stepped-wedge designs (SWD) applicable to both normal and non-normal outcomes.
• To extend sample size calculation methodologies for SWD within the framework of generalized linear mixed models.
• To provide a practical tool for researchers to accurately determine sample size and power for SWD.

Main Methods:

• The proposed method utilizes the Laplace approximation, as detailed by Breslow and Clayton, to derive the covariance matrix of estimated parameters.
• This approach is integrated into generalized linear mixed models to accommodate a wider range of outcome types.
• The formula is designed for efficient computation, enabling rapid sample size adjustments.

Main Results:

• The performance of the proposed power calculation method is validated through simulations and compared against simulation-based sample size calculations.
• The method's utility is demonstrated through practical applications in studies on STI treatment and radiologic imaging reporting.
• An R package, "swCRTdesign", is provided to facilitate the implementation of these sample size and power calculations for multilevel SWD.

Conclusions:

• The developed method offers a computationally efficient approach to sample size and power calculations for SWD.
• It supports dynamic updates and facilitates the exploration of various design options and assumptions.
• This advancement aids in generating more definitive evidence from healthcare intervention studies using SWD, particularly with non-normal outcomes.