Saddlepoint $$p$$ p -values and confidence intervals for the class of linear rank tests for censored data under generalized randomized block design

Abstract

© 2015, Springer-Verlag Berlin Heidelberg. One of the commonly used classes of tests for testing treatment effects for censored data is the linear rank class. The underlying distribution of this class is determined by the randomization design used to collect the data. Many randomization designs are used in clinical trials. The randomized block design is an important design that reduces selection bias and accidental bias. In this paper, a double saddlepoint approximation for the exact underlying randomization distribution for the linear rank class under generalized randomized block design is presented. Extensive simulation studies are used to assess the performance of the saddlepoint approximation. This approximation shows a great improvement in accuracy over the asymptotic normal approximation. This accuracy enables us to calculate almost exact confidence intervals for the treatment effect.