Specific statistical considerations relevantto the design and analysis of gingivitis trials demonstrating

Simulation studies were conducted to address specific statistical issues which arise in the design and analysis of gingivitis studies whose principal aim is the demonstration of superiority or equivalence of one product to another. The effects of measurement scale, using differences or ratios of group means, particular statistical test produces and specific rules demonstrating superiority or equivalence were investigated. An alternative concept to equivalence—denoted “least as good”—was also defined and evaluated. For a wide class of possible distributions of gingivitis scores, characterized by specific gamma distributions, the student-t test applied to means of subject GI gingivitis scores proved to be the most powerful of the test produces considered, having statistical properties quite similar to the randomization or permutation test procedure. Transformations of subject GI mean gingivitis scores did not produce an advantage in demonstrating either superiority or equivalence, and in some cases made it more difficult. Little difference was observed in test results when using the difference in group means as compared with using the ratio of group means for demonstrating either equivalence or superiority. The clinically significant rule produced the lowest false-positive rates for products slightly better than the active control, and similar false-positive and -negative rates as the statistically significant rule for products clearly superior to the active control. Demonstration of product equivalence will require more subjects per group than demonstrating product superiority, the size of this difference being a function of the definition of superiority that is accepted. Showing that the 90% confidence interval for 100*R is completely contained within the [90%, 110%] interval is the preferred method of demonstrating equivalence today, although much more research needs to be done to improve methods for demonstrating product equivalence. The “least as good” alternative to “equivalence” makes it easier to demonstrate “equivalence” for products slightly better than the active control product, but both experience great difficulty in demonstrating equivalence for lest products not quite as good as the active control.