Assessing the Impact of Nonproportional Hazards on Estimation of Treatment Effects of Progression-Free Survival

• Activity Number: 646
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Adam Boyd*+ and David Raunig
• Companies: Array BioPharma and ICON Medical Imaging
• Keywords: survival analysis ; interval censoring ; oncology

Abstract

The majority (18 / 29, 62%) of recent FDA regulatory submissions leading to approval of drugs to treat solid tumors were based on progression-free survival (PFS) or other tumor-based, interval-censored endpoints (Huang et al, DIJ, 2012). In randomized studies with time-to-event endpoints such as PFS, it is common to compare treatments using the log-rank test, and to summarize treatment effect estimates via the hazard ratio as estimated from the proportional hazards model. In most cases, the fact that the data are interval-censored is ignored and right-censoring methods are applied, usually by imputing the right-most point of the time interval as the progression date ("right-point imputation"). In this paper we investigate the impact of non-proportional hazards (NPH) on estimation and testing of the hazard ratio using the proportional hazards model when the data are interval censored and right-point imputation is applied. Robustness properties of other approaches, such as Finkelstein's method, are also explored. Recommendations for estimation in the presence of NPH are provided.