Abstract:
|
The equivalence test in analytical similarity assessment uses a margin of . In the current practice, , although estimated using study data, is considered as a fixed constant. The impacts on type I error rate inflation and power reduction were pointed out by many investigators. In 2015 and 2016, various approaches were proposed to address the issue. These approaches include Wald test using restricted maximum likelihood estimate (RMLE) of , generalized pivotal quantity (GPQ), and exact test based approaches. Through simulation comparisons, it is demonstrated that all approaches risk the possibility of type I inflation especially when the variance is different in a study with imbalanced sample sizes. On the other hand, Wald test with RMLE of could have seriously low type I error rate when the sample size is small. We propose to use the exact distribution of percentile estimation with RMLE of the variance of the percentile estimate to improve the type I error rate and power from Wald test approach. The results and discussion through simulation studies will be presented.
|