ABSTRACT
Delayed separation in survival curves has been observed in immuno-oncology clinical trials. Under this situation, the classic log-rank test may confront high power loss. In this paper, we consider a Zmax test, which is the maximum of the log-rank test and a Fleming–Harrington test. Simulation studies indicate that the Zmax test not only controls the Type I error rate but also maintains good power under different delayed effect models. The asymptotic properties of the Zmax test are also established, which further supports its robustness. We apply the Zmax test to two data sets reported in recent immuno-oncology clinical trials, in which Zmax has exhibited remarkable improvement over the conventional log-rank test.
Acknowledgments
The paper is substantially improved thanks to the helpful comments of an associate editor and the referees. Research has been supported by National Science Foundation grant DMS 1812258.