Abstract:
|
Between-group summary measures based on restricted mean survival time (RMST) provide robust and clinically interpretable information about the treatment effect. Statistical tests based on the RMST may be useful options in comparative clinical trials, because we can have quantitative information about the treatment effect (e.g., 0.95 confidence interval) that corresponds to the test result. Although a choice of the truncation time (tau) for RMST is challenging in practice, this nice property of coherency still stands whichever tau may be used. Motivated by this, we developed a pre-specified versatile test based on RMST, where tau is selected data-dependently. We confirmed the validity of the new test by numerical studies, which also demonstrated that the new test is dramatically more powerful than logrank, Wilcoxon, and RMST-based tests with a fixed tau under patterns of difference in survival function seen in recent cancer clinical trials. The new test is inferior to the logrank test under proportional hazards differences by theory. However, the new test would be preferable when investigators expect a similar pattern of difference seen in those cancer clinical trials.
|