This document presents a detailed overview of the research article “Stepped wedge cluster randomised trials: a review of the statistical methodology used and available” by D. Barker, P. McElduff, C. D’Este, and M. J. Campbell, published in 2016. This review is particularly significant as it addresses the statistical methodology, including the appropriateness of sample size methods, in stepped wedge designs (SWDs), an area that previous reviews had not focused on.
The authors conducted a comprehensive literature review across Medline, Embase, PsycINFO, CINAHL, and Cochrane databases to identify methodological guides and randomized controlled trials (RCTs) that employed the stepped wedge design. Their investigation sought to understand currently recommended statistical methods for analysis and sample size/power estimation in SWDs, as well as their practical application.
Key findings reveal a notable increase in the use of SWDs, with this review identifying 102 such trials, a substantial increase from previous reviews. Despite the growing popularity, the review highlights significant methodological concerns:
- A striking number of studies employed the SWD with few clusters, with nearly half (45%, n = 46) having fewer than 10 clusters and 64% (n = 65) sampling fewer than 20. The median number of clusters was 12, ranging from 2 to 506.
- While most trials (77%, n = 77) used either Generalised Estimating Equation (GEE) or mixed model approaches for analysis, only a minority (22%, n = 22) estimated sample size in a way that appropriately accounted for the stepped wedge design.
- The review also found that for cohort stepped wedge designs, there was only one paper considering analysis and one considering sample size methods, indicating a significant gap. This is particularly relevant as 46 trials identified were cohort designs.
- A concerning issue is the misapplication of sample size methods: 11 studies identified as cohort SWDs used methods applicable only to cross-sectional SWDs. This underscores the critical need for researchers to distinguish between cohort and cross-sectional SWDs in sample size calculations.
- The potential confounding effect of time was explored and adjusted for in 60% (n = 61) of the studies, but its impact on statistical power is an area requiring further research.
The authors conclude that statistical methodology for SWDs is lagging behind current practice, leading to concerns that many studies may be underpowered or biased due to an insufficient number of clusters for the chosen analysis methods. They emphasize the urgent need for research on the minimum number of clusters required for both types of SWDs and on the impact of adjusting for time on study power, particularly for smaller trials. The article highlights that the perceived power advantage of SWDs over parallel cluster RCTs may not hold true when all design complexities, such as time adjustment, are fully accounted for. This review serves as a critical appraisal of current practices and a call for more robust methodological development in stepped wedge cluster randomized trials.
APA Reference:
Barker, D., McElduff, P., D’Este, C., & Campbell, M. J. (2016). Stepped wedge cluster randomised trials: a review of the statistical methodology used and available. BMC Medical Research Methodology, 16(1), 69. https://doi.org/10.1186/s12874-016-0176-5
