522 – Contributed Oral Poster Presentations: Biopharmaceutical Section
Power and Sample Size Estimation in Clinical Trials with Multiple Co-Primary Endpoints
Zuoshun Zhang
Celgene Corporation
For some diseases, there needs to show the significance of a treatment in multiple co-primary endpoints. In the design stage, it will be helpful to have easily accessible tools for power and sample size estimation. Sozu et al. (2011) provided formulas and gave examples for 2 and 3 co-primary endpoints using integration. For any number of co-primary endpoints, we give an alternative way for accurate and quick estimation for power and sample size. We factor the positive definite correlation matrix into a product using Cholesky decomposition and generate independent standard normal random numbers. The random numbers are transformed into correlated vectors corresponding to the sufficient test statistics of simulated trials, from which the power and sample size are estimated. For 2 or 3 co-primary continuous endpoints, we estimated the sample sizes and compared with those in Sozu et al (2011); the differences are within 1% for each of the 100 cases. We further estimate sample sizes for cases with 4 and 5 co-primary endpoints. We provided a program using SAS IML and base language, which can be easily adapted to other designs including those with binary endpoints.
