Non‐parametric assessment of non‐inferiority with censored data

We suggest non‐parametric tests for showing non‐inferiority of a new treatment compared to a standard therapy when data are censored. To this end the difference and the odds ratio curves of the entire survivor functions over a certain time period are considered. Two asymptotic approaches for solving these testing problems are investigated, which are based on bootstrap approximations. The performance of the test procedures is investigated in a simulation study, and some guidance on which test to use in specific situations is derived. The proposed methods are applied to a trial in which two thrombolytic agents for the treatment on acute myocardial infarction were compared, and to a study on irradiation therapies for advanced non‐small‐cell lung cancer. Non‐inferiority over a large time period of the study can be shown in both cases. Copyright © 2005 John Wiley & Sons, Ltd.