ABSTRACT
Suppose you wish to conduct a clinical trial to compare the survival of leukemia patients under two regimens (one a standard or control therapy, and the other experimental). From past experience, suppose the 5-year survival under the standard therapy is 60%, and you would like your trial to be large enough to have a high probability of obtaining a statistically significant difference (i.e., high enough power) when the experimental therapy represents a 10% improvement to 70% survival at 5 years. Due to sampling error, even if the true target population rates are 60% and 70% at 5 years for the standard and experimental therapies respectively, the observed 5-year rates will be different. An undersized study may well miss this clinically important improvement. Conversely, an oversized study may waste resources and could delay both publication of the conclusions and the opening of a successor trial.
