A trial of a new therapy is to be compared to results from a previous trial of patients treated with a standard therapy. For a given sample size for the trial of the new therapy, we desire the power, against a specific alternative hypothesis, for the hypothesis test of the null hypothesis that the therapies are equivalent. Alternatively, the sample size required for the trial of the new therapy is needed for a target power. We explain why a popular method for doing these calculations is wrong, and discuss alternative methods in the context of normal outcomes, binary outcomes, and time-to-event outcomes.
Biometric Research Branch, National Cancer Institute, Bethesda, MD 20892, USA. korne@ctep.nci.nih.gov
Stat Med. 2006 Sep 15;25(17):2922-31.
AbstractA trial of a new therapy is to be compared to results from a previous trial of patients treated with a standard therapy. For a given sample size for the trial of the new therapy, we desire the power, against a specific alternative hypothesis, for the hypothesis test of the null hypothesis that the therapies are equivalent. Alternatively, the sample size required for the trial of the new therapy is needed for a target power. We explain why a popular method for doing these calculations is wrong, and discuss alternative methods in the context of normal outcomes, binary outcomes, and time-to-event outcomes.