CRTpowerdist: An R package to calculate attained power and construct the power distribution for cross-sectional stepped-wedge and parallel cluster randomized trials

Yongdong Ouyang; Liang Xu; Mohammad Ehsanul Karim; Paul Gustafson; Hubert Wong

doi:10.1016/j.cmpb.2021.106255

CRTpowerdist: An R package to calculate attained power and construct the power distribution for cross-sectional stepped-wedge and parallel cluster randomized trials

Comput Methods Programs Biomed. 2021 Sep:208:106255. doi: 10.1016/j.cmpb.2021.106255. Epub 2021 Jun 25.

Authors

Yongdong Ouyang¹, Liang Xu², Mohammad Ehsanul Karim², Paul Gustafson³, Hubert Wong²

Affiliations

¹ School of Population and Public Health, University of British Columbia, 2206 E Mall, Vancouver, BC Canada V6T 1Z3; Centre for Health Evaluation & Outcome Sciences, 588 - 1081 Burrard Street, St. Paul's Hospital Vancouver, BC Canada V6Z 1Y6. Electronic address: douyang@cheos.ubc.ca.
² School of Population and Public Health, University of British Columbia, 2206 E Mall, Vancouver, BC Canada V6T 1Z3; Centre for Health Evaluation & Outcome Sciences, 588 - 1081 Burrard Street, St. Paul's Hospital Vancouver, BC Canada V6Z 1Y6.
³ Department of Statistics, University of British Columbia, 3182 Earth Sciences Building, 2207 Main Mall Vancouver, BC Canada V6T 1Z4.

PMID: 34260969
DOI: 10.1016/j.cmpb.2021.106255

Abstract

Background: The attained power, calculated conditional on the realized allocation, of a clinical trial may differ from the expected power, obtained pre-randomization through averaging over all potential allocations that could be generated by the randomization algorithm (RA). For example, a two-arm trial using a RA that is expected to allocate 20 participants to each arm will attain less than the expected power if by chance it allocates 25 and 15 participants to the arms. Cluster randomized trials with unequal cluster sizes have elevated risk of realizing an allocation that yields an attained power much lower than the expected power when modest numbers of clusters are randomized.

Method: We developed the R package CRTpowerdist, which implements both simulations and approximate analytic formulae to calculate the attained powers associated with different realized allocations and constructs the pre-randomization power distribution associated with the RA to facilitate assessing the risk of obtaining inadequate power. The package covers unequal cluster-size, cross-sectional stepped-wedge and parallel cluster randomized trials, with or without stratification. Allowed outcome types are: continuous (Gaussian), binary (Binomial) and count (Poisson). The analytic formulae-based calculations are also implemented in a Shiny app.

Results: The functionality of the CRTpowerdist is illustrated for each type of trial design. The examples show how to obtain the attained power, the power distribution, and the risk of low attained power, using both simulation and analytic formulae.

Conclusion: For cluster randomized trials with unequal cluster sizes, the CRTpowerdist package can assist users in identifying an appropriate randomization algorithm by enabling the user to assess the risk that a randomization algorithm will lead to an allocation with inadequate attained power. The Shiny app makes these assessments accessible to researchers who are unable or do not wish to use the CRTpowerdist package.

Keywords: Attained power; Cluster randomized trial; Power distribution; Stepped-wedge.

MeSH terms

Cluster Analysis
Computer Simulation
Cross-Sectional Studies
Humans
Randomized Controlled Trials as Topic
Research Design*
Sample Size