Exact power calculations when the dependent variable is a single proportion and the number of events is small

This paper presents methods for calculating the exact power to reject the null hypothesis that an observed event rate is significantly lower than a clinically relevant alternative event rate or that the confidence interval is less than a specified range when the hypothesis will be tested using exact binomial confidence intervals. SAS programs that perform the calculations are presented and the results are compared with power calculations using the normal approximation. The method can be extended to studies involving paired comparisons but, in that case, it does not appear to provide a significant advantage over normal approximation methods.