Group Sequential, Adaptive, and Interim Analysis
Calculate the conditional probability of achieving the planned final-analysis success criterion given the interim Z-statistic, information fraction, and an assumed future treatment effect. Uses the Brownian-motion B-value framework underlying standard conditional-power calculations in sequential clinical-trial monitoring.
This calculator uses the Brownian-motion formulation of conditional power for a sequential clinical trial. At information fraction t, the interim Z-statistic is converted to the B-value b = Z(t)√t. Conditional on this interim B-value, the future increment from t to 1 has variance 1 − t.
Here, c is the final upper critical Z-value and θ is the assumed standardized drift for the remainder of the trial. The calculator reports three commonly useful assumptions: the original planned effect, the current-trend effect implied by the interim data, and the null effect.
Published material gives an illustrative trial with planned power of 85%, an interim information fraction of 50%, and an interim Z-statistic of 0.3. Using a final critical value of 1.96, the planned-effect drift is c + z0.85 = 1.96 + 1.03643 = 2.99643. The corresponding conditional power under the planned effect is approximately 36.20%. Under the current-trend assumption, the conditional power is approximately 1.49%.
Conditional power is a conditional probability, not a replacement for the prespecified group-sequential decision rule. It is typically used to describe the chance of eventual success under an explicit assumption about the future treatment effect. Different effect assumptions can produce materially different conditional-power values from the same interim data.
Jennison, C. & Turnbull, B.W. (2000). Group Sequential Methods with Applications to Clinical Trials. Chapman & Hall/CRC, Boca Raton. The text develops conditional power for interim analyses and group-sequential trials.
Proschan, M.A., Lan, K.K.G., & Wittes, J.T. (2006). Statistical Monitoring of Clinical Trials: A Unified Approach. Springer, New York. Chapter 4 develops conditional, unconditional, and predictive power using the Brownian-motion/B-value framework.
the relevant methodological literature identifies conditional-power procedures for multiple endpoint types, including one- and two-sample means, proportions, and logrank tests, and describes conditional power as the probability that the final result will be significant given the interim data. documentation likewise describes conditional power as the probability of rejecting the null at a subsequent look given the current test statistic and assumed parameter values.