Clinical Trials: Superiority-Testing

Abstract

One of the flaws of modern statistics is that it can produce statistically significant results even if treatment effects are very small. The objective of this work was to provide examples of studies that have been published as unequivocally positive, although the treatment effects were substantially smaller than expected; and 2 to introduce superiority-testing as a novel statistical approach avoiding the risk of statistically significant but clinically irrelevant results. We from recent volumes of the Lancet six original articles of controlled clinical trials that were reported as being positive studies, although they did not meet their expected power. The studies produced only 53 to 83% of the statistical power expected, while the new treatments produced only 46 to 86% of the magnitude of response expected. Instead of a p-value of 0.05 as cut-off criterion for demonstrating superiority a stricter criterion seems to be needed. For that purpose, similar to equivalence-testing and non-inferiority-testing, prior boundaries of superiority have to be defined in the protocol. If the 95% interval of the study turns out to be entirely within these boundaries, then superiority is accepted. Nowadays, too many borderline significant studies are being reported as convincingly positive studies. This is misleading, as it produces overestimated expectations from new treatments. Superiority-testing, as introduced in this paper, is a simple method to avoid this problem.

Key words: :

• Clinical Trials
• Statistical Power
• Type II Error
• Sample Size
• P-value
• Z-value
• Non-Inferiority-Testing
• Equivalence-Testing
• Superiority-Testing