Biostatistics Seminar Series: Keaven M. Anderson, PhD

Title: A proposal for design and analysis and clinical trials with potentially non-proportional hazards.Abstract: Methods efficient and effective for proportional hazards are standards for application in survival analysis in clinical trials, but are both inefficient and less effective for other situations. It is common in multiple therapeutic areas (e.g., vaccines, cardiology, oncology) to have a treatment effect that may take time to develop or a treatment effect that diminishes over time. In these cases, the first issue is that hazard ratios and median time-to-event may not be effective for describing treatment differences. There is also the issue that logrank tests can be less powerful than other methods for demonstrating a treatment effect. In this context, the FDA initiated a cross-PhRMA effort to investigate alternative methods that might be effective for regulatory use where the proportional hazards assumption may be questioned a priori. Following 2 years of investigation, this effort has brought forward proposals for more meaningful description of treatment effect and statistical testing that is robustly powerful across a variety of assumptions. We will present these methods and some of the challenges with moving them forward for application in a regulatory setting.