Loading...
Comparing the equivalence testing with using two one-sided test, square root of f distribution and 2-DF for shift-scale-equivalence test
Author(s)
Date Issued
June 1, 2019
Type
Conference Paper
Abstract
In pharmaceutical and medical studies, we would like to show any formulations or two treatments are equivalent. For example, Westergren ESR and STATplus ESR are two popular measurements of sedimentation rate, which are used to monitor disease severity in patients with rheumatoid arthritis and other inflammatory rheumatologic conditions. Westergren ESR is a well-known measurement that was developed by R. S.Fahraeus and A.V.A. Westergren in 1921, while STATplus ESR is an innovative measurement to accelerate turnaround time. Compared with Westergren ESR, the result from STATplus ESR is easier to understand. Since these two measurements can be used to test the same study, it is necessary to know if they can be switched. Typically, a new measurement process is compared with an existing measurement process. Paired data of these two measurements occur because they are used on the same subject. Usually, paired t- test is appropriate for paired data, but it does not fit well for some situations because paired t-test can only be used to check significant differences from paired data. If the paired data have positive or negative association, the result from the paired t-test might be the same. For example, one paired dataset has positive correlation, and the other one paired dataset has negative correlation. But paired t-tests give the same conclusion because they have the same differences .Moreover, the paired t-test might have low power for scale-type relationships. In this paper, we propose a test that has reasonable power for both shift and scale-type relationships, which is based on shift- scale type relationships. We consider an equivalence testing for hypothesis. It is an approach to swap the hypotheses so that statistical equivalence of the two measurements is the alternative hypothesis and bears the burden of proof. We conclude “equivalence” only if there is evidence to support the claim that the magnitude of disagreement between the two measurements lie within specified limits.
Citation
Proceedings European Council for Modelling and Simulation Ecms, 33(1), 400-406, 2019