Abstract:
|
Ratio of means and difference of means are often used for continuous or ordinal data in average bioequivalence (ABE) or non-inferiority (NI) to evaluate whether the TEST mean is equivalent or non-inferior to the reference mean by a certain margin. Particularly, for generic drugs, ratio of means is widely used due to its invariance to scale differences. However, caution is needed when using mean ratio for continuous or ordinal data in NI or ABE tests. Mean ratio is invariant to scale shift, but not to location shift. The power of a NI or ABE test is determined by the CV of the data (Hauschke et al 1999). Therefore, how continuous data and ordinal data are defined can make a significant difference to the power of an ABE or NI test. We performed simulation for normally distributed, skewed normally distributed, and ordinal data based on multinomial distributions. Power is dramatically changed when continuous or ordinal data are defined differently. Mean difference, however, is invariant to the shift of location. Conclusions: Caution is needed when defining continuous or ordinal data and when choosing ratio of means or difference of means for a NI or ABE test.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.