Should Progression-Free Survival Be the Primary End Point for Registration of Oncology Drugs?

John M. Burke, MD

Hematologist and Medical Oncologist

Rocky Mountain Cancer Centers

Associate Chair

US Oncology Hematology Research Program

Aurora, CO

The November 10 issue of the Journal of Clinical Oncology contains a Comments and Controversies article that caught my attention. Written by Christopher Booth, MD, and colleagues, the article is provocatively titled “Progression-Free Survival [PFS] Should Not Be Used as a Primary End Point for Registration of Anticancer Drugs.” If the title is correct, then many randomized trials in oncology are improperly designed and many drugs used in oncology have been approved based on the wrong end point.

In the paper, the authors outline several limitations of PFS as an end point for oncology drugs, including lack of consistent surrogacy for overall survival (OS), bias, and error inherent in measuring tumor sizes; arbitrariness in the definition of progression; and the dependency of PFS on the frequency of imaging. Perhaps the most important limitation in the PFS end point is something that I knew little about called informative censoring. This is the censoring of patients who drop out of the trial before undergoing a response assessment. For example, if a patient drops out of a trial prior to response assessment because of toxicities of treatment, that patient is censored from the PFS curve and a PFS event is not recorded. Thus, if 20% of patients in an experimental arm of a trial experience toxic effects forcing discontinuation of treatment prior to response assessment, then the PFS will be based on an analysis of 80% of patients who could tolerate the treatment. In a randomized trial in which the design is standard therapy vs standard therapy plus drug X, informative censoring will lead to a bias in favor of the investigational arm.

Potential methods for mitigating this include performing a sensitivity analysis in which it is assumed that patients who drop out of trials did so for progression; analyzing time-to-treatment-failure curves, which are not affected by informative censoring; and placing PFS results in the context of discontinuation rates, toxicities of regimens, quality-of-life assessments, and OS. The authors argue that a better way to design randomized trials in oncology would be to designate OS as the primary end point and to make PFS a secondary end point.

It's worth keeping the limitations of the PFS end point in mind when we observe data. The 2 ways in which we physicians can help our patients is to make them live longer or improve their quality of life. PFS does not measure either of these outcomes, is more subject to bias and error than OS, and does not incorporate any measure of harm. However, safety, discontinuation rate, quality of life, and OS end points incorporate measures of harm and are critical when analyzing the risk-benefit ratios of the treatments we study.