Abstract
Purpose: :
To compare study power associated with use of the Bonferroni, Hochberg, and Closed Testing methods of controlling family-wide Type I error (FWE) when making pairwise comparisons of means in the Standard Care versus COrticosteroid for REtinal Vein Occlusion (SCORE) Study.
Methods: :
The SCORE Study consists of two independent 3-armed randomized controlled clinical trials to investigate the safety and efficacy of standard care versus intravitreal triamcinolone acetonide for macular edema associated with central and branch retinal vein occlusion. We simulated outcomes of the SCORE Study under the study's design hypotheses, and used logistic regression to compare predicted performance when the Bonferroni, Hochberg, or Closed Testing methods were employed.
Results: :
Closed testing was more powerful than Hochberg's method by several definitions of multiple-test power. Under the SCORE design hypotheses, Closed testing improved power to reject a single false hypothesis by about 5 percentage points, and power to reject both false hypotheses by about 8 percentage points. In general, power improvement is roughly equal to that of Hochberg's method over the Bonferroni procedure. Simulations over a wider parameter space confirmed this superiority for normal, logistic, and Poisson regression, although the closed benefit over Hochberg begins to disappear when 5 or more arms are considered.
Conclusions: :
The SCORE Study modified its statistical analysis plan to use Closed Testing in place of Hochberg's method of controlling FWE. Clinical trials with 4 or fewer arms could consider using the Closed Testing method in preference to the Bonferroni or Hochberg approach for strong control of FWE.
Clinical Trial: :
www.clinicaltrials.gov NCT00105027
Keywords: clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology