Confidence Sets-based Procedures for Identifying the Minimum Effective Dose with Right-censored Lifetime Data

In this paper, we consider identifying the minimum effective dose (MED) in a doseresponse study with randomly right-censored lifetime data, where the MED is the lowest dose level with a median lifetime larger than that of the zero-dose control by a certain threshold value. The MED is identified based on the stepwise confidence sets for the ratios of median lifetimes of each non-zero dose group and the zero-dose control. Parametric confidence sets are proposed when lifetimes in different dose groups are distributed as generalized gamma distributions. Nonparametric confidence sets are also suggested when distributions of the lifetimes remain unknown. The comparative results of a Monte Carlo error rate and power/bias study for a variety of survival and censoring distributions are then presented and discussed. The application of the proposed procedures is finally illustrated for identifying the MED of the diethylstilbestrol in the treatment of prostate cancer.