Estimating survival benefit in castrate metastatic prostate cancer: decision making in proceeding to a definitive...

In designing a Phase II trial, the acceptable clinical activity region for a new therapy is often developed using data from historically treated patients. This region incorrectly ignores the variability of this estimate, because the efficacy of the prior treatment lies somewhere around the estimate. The size of this interval is dependent on the sample size used. This report illustrates the use of a published method that accounts for this uncertainty and aids in the decision to proceed to a definitive trial.

A historical data set of low-risk patients with progressive castrate metastatic prostate cancer and a group of similar patients treated in a Phase II chemotherapy trial were used. The 1-year Kaplan-Meier estimate of survival was obtained for both. This approach uses the 75% upper confidence bound of the 1-year survival probability from the historical data set to define the lower limit of acceptable clinical activity. Use of this bound makes the approach more conservative, and hence the decision to proceed to a Phase III trial more difficult.

In the low-risk historical patients, the 1-year Kaplan-Meier estimate of survival was 66.4% (75% upper confidence bound 71.0%). In the Phase II patients, the 1-year Kaplan-Meier estimate of survival was 89.5% (95% lower confidence bound 78.2%).

A hypothesis test using the 75% upper confidence bound to define the lower limit of acceptable clinical activity demonstrates that the 1-year survival probability on Taxol/estramustine/carboplatin is greater than that of the historical population, and hence should be taken into a definitive trial. The design provides investigators increased confidence in making this decision.