Performance evaluation and operating characteristics of commonly used two stage adaptive designs and extension of the sample size calculation method to a Poisson endpoint

In group sequential clinical trials, there are several sample size re-estimation methods that allow for change of sample size conditioned on effect size observed at the interim analysis. Our simulation studies compared two commonly used sample size re-estimation methods, J. S. Denne and Cui, Hung and Wang (CHW), based on their operating characteristics. Based on this research, the J. S. Denne method is slightly better in terms of overall power and is comparable to the CHW method in terms of average sample size and type I error rate. To test the robustness of these methods, we explored the impact of incorrect variance assumption on the operating characteristics. Based on simulations, we found an overall decrease in power in both the methods.;In many clinical trials, the clinical endpoint, especially MRI endpoints in neurology trials, follows a Poisson distribution. An extension to J. S. Denne method for a Poisson endpoint was developed using the normal approximation to the Poisson distribution. The desired overall power was achieved and the type I error rate was maintained.;Finally, we propose a method that uses the exact distribution of the difference between two Poisson variables to calculate sample size at the protocol design stage. When the difference between the two Poisson rates is more than 1.3 units, the number of subjects and events needed at the desired power and type I error rate is slightly less than that computed by simulation based on the normal approximation method. The exact sample size calculations are more comparable to the normal approximation when the difference between the rates is less than 1.3 units. The proposed method is more intuitive, efficient and less subjective compared to the normal approximation method. An intuitive method to calculate the interim and final critical values for the exact method by using O'Brien Fleming critical values as the starting point is also proposed. Simulation results demonstrate that the type I error rate and power is maintained at the desired level. A simple code is developed in R-software to estimate the sample size and critical values.