Premium
Pairwise non‐parametric non‐inferiority tests in 3 × 3 cross‐over trials: should we adjust for period?
Author(s) -
Duchateau Luc,
Janssen Paul
Publication year - 2005
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.2048
Subject(s) - parametric statistics , pairwise comparison , statistics , placebo , mathematics , medicine , nonparametric statistics , semiparametric model , econometrics , alternative medicine , pathology
Abstract Before an active compound can be registered as a drug, it needs to be demonstrated that it does not cause particular unwanted compound specific side effects. Demonstrating non‐inferiority of an active drug with placebo with respect to an unwanted side effect is often based on pharmacodynamic 3 × 3 cross‐over trials, with the third treatment arm consisting of a positive control. The main comparison then is the pairwise comparison between active drug and placebo. In this paper, two different non‐parametric methods with adjustment for period effect are compared with the non‐parametric non‐adjusted test and the parametric period‐adjusted test with respect to size and power. The non‐parametric test with period adjustment based on rank alignment has generally the largest power, but its size exceeds the nominal significance level. The non‐parametric test with period adjustment based on stratification has low power for trials with a small number of subjects. The non‐parametric test without period adjustment is a valid alternative in such cases, but its power decreases substantially in the presence of period effects. In the case of unbalanced designs, however, only the non‐parametric test with period adjustment based on stratification can be used. Copyright © 2005 John Wiley & Sons, Ltd.