ABSTRACT
Rank-based test procedures in censored survival data differ considerably in their sensitivity to various alternatives to the hypothesis of equality of underlying distributions. Procedures based on the simultaneous use of multiple statistics provide an appealing approach to obtaining more global sensitivity while maintaining the sensitivity to alternatives of interest. In this article, the joint distribution of weighted logrank statistics (to assess overall differences in time-to-event distributions) and a standardized difference in Kaplan-Meier estimates (to assess differences at a specific prespecified time post randomization) is obtained and this result is used to formulate statistical test procedures based on the simultaneous use of these two types of statistics. Simulations are used to assess small sample properties of this approach and its usefulness is illustrated in important recent oncology clinical trials.
Notes
Notes: S i (t) = exp(−t/3.6); C i (t) = 1 − t/b, 0 ≤ t ≤ ∞, i = 1, 2. ρ∘, α∘∗ and α∘∗ ∗ are Monte Carlo estimates of ρ,α∗, and α∗∗, respectively.
Note: ρ∘, α∘∗ and α∘∗ ∗ are Monte Carlo estimates of ρ, α∗ and α∗∗, respectively.