Analysis of cluster randomised stepped wedge trials with repeated cross-sectional samples

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Stepped wedge cluster randomized trials (SW-CRTs) are increasingly utilized for policy and service delivery interventions.
• A significant gap exists in trial literature regarding analytical approaches for SW-CRTs.
• Many published SW-CRTs suffer from methodological flaws, notably failing to adjust for secular trends during analysis.

Purpose of the Study:

• To highlight assumptions of the basic SW-CRT analytical model by Hussey and Hughes.
• To explore modifications accommodating deviations from model assumptions using random and fixed effects.
• To illustrate the importance of adjusting for secular trends in SW-CRT analysis through a case study.

Main Methods:

• Examined assumptions of the basic SW-CRT analytical model.
• Proposed simple model modifications using random and fixed effects to address assumption deviations.
• Assessed implications of modifications on intracluster correlation coefficients.
• Conducted a case study to demonstrate the impact of secular trend adjustment.

Main Results:

• The basic SW-CRT model assumes a common secular trend and constant treatment effect.
• Modifications allow for varying secular trends and treatment effects across clusters or time.
• A case study revealed a reversal of the apparent treatment effect after adjusting for secular trends, contrasting with unadjusted estimates.

Conclusions:

• SW-CRT analysis must account for secular trends to prevent biased treatment effect estimates due to time confounding.
• The standard Hussey and Hughes model relies on critical assumptions that may not hold.
• Careful consideration of appropriate analytical models is essential for SW-CRTs, with provided Stata code for implementation.