In clinical research, investigators are often interested in the occurrence of certain events. Therefore, survival endpoints are commonly used in drug or therapy evaluations, for example, treatments for cancer, and organ transplant. It is often of interest to compare the hazard rate in the experimental group with that in the control group. Sometimes, the experimental therapy may be acceptable as long as there is evidence that it is not worse than the control therapy due to the consideration of toxicity, cost et al. Therefore, non-inferiority trials may be conducted.

One commonly raised problem is the calculation of the sample size in non-inferiority trials with survival endpoints. This roundtable will focus on the discussion of the methodology for sample size determination in clinical trials of this type. We will also discuss some other related issues such as margin determination and multi-region problems.

Key discussion questions: 1. What are the current methodologies commonly used in the sample size determination in the non-inferiority trial with survival endpoints? Assumptions? 2. Exponential model, Cox’s proportional hazard model and the Logrank test are three commonly used methods to determine the sample size needed in a non-inferiority trial. Both the Exponential model and the Cox model assume constant hazard ratios over time. The Logrank test is most powerful under the proportional hazard assumption. Question: what is the consequence of using these methods if the proportional hazard assumption is violated? Larger sample size? 3. When the proportional hazard assumption is violated, simulations may be used to determine the sample size. Weighted Logrank test can be used for testing equality of the hazard ratio functions. How can this approach be extended to a non-inferiority test? 4. Methods based on median survival. Advantages and drawbacks? 5. More recently, developed methods based on restricted means? 6. How to incorporate the drop-out rate? 7. How to choose the margin in the sample size determination? 8. Many clinical trials involve multiple regions. If the regions are heterogeneous, how to analyze the data? Random effect models? Region-specific margin?