1 A Simple Way to Program Power Calculations

We propose simulating large sets of statistical power calculations and selecting from a long list of study designs those with appropriate power. Typical commercial programs fill in all but one parameter and solve for the remaining parameter. We avoid solving equations and thereby obtain power for atypical designs. First, for simplicity, we compute the power for the test comparing two proportions without assuming either approximate normality or a common variance. Then, we extend the log-rank test power calculation to account for various patterns of on-study censoring, study duration, and patient recruitment period. We generate a new approximation for the censoring adjustment. Insodoing, we set out a easy way for others to modify our program for other related power calculations