Hypothesis Testing Under Non-Proportion Hazards: Methodologic Challenges and Interpretations

• Activity Number: 271 - Recent Statistical Advances of Handling Nonproportional Hazards in Study Design and Data Interpretation
• Type: Topic Contributed
• Date/Time: Tuesday, August 9, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jianghua He* and Huan Cheng
• Companies: University of Kansas Medical Center and University of Kansas Medical Center
• Keywords: Survival Analysis; Cox Models; Clinical Trails; Censoring

Abstract

The commonly used proportional hazards model and log-rank test for survival outcomes are built upon and most powerful under the proportional hazards (PH) assumption that is often violated in oncology trials. Various research methods have been proposed to deal with Non PH (NPH) situations, such as modeling time-varying effects, estimating average effects, and various weighted log-rank/combo tests. There are some methodologic and clinical challenges in applying these methods. For example, independent censoring becomes an issue that affect testing and estimates and the estimates losses its original interpretations; weighted log-rank tests weight the effect differently at different time points without adequate justifications; different weighted log-rank tests may get opposite conclusions etc. In this presentation, simulation studies are done to show the impact of independent censoring on data analysis using Cox model or (weighted) log-rank/combo tests and explore the interpretations of the underlying hypotheses and testing results. Potential remedial methods are also presented, and their performances are examined.