Overall Survival as Endpoint
A commentary published recently in the Journal of Clinical Oncology by Dr. Edward Korn of the NCI, discusses the use of overall survival (OS) as the measure of clinical benefit. This review is divided into three sections, beginning first with situations where identical subsequent therapy is used after both treatment arms and is either differentially effective or equally effective. Second, Dr. Korn and colleagues discuss special challenges in trials where patients cross over from standard to experimental treatments. The third section illustrates examples as well as conclusions on the use of OS for clinical trial endpoints.
Typically in randomized controlled trials (RCTs), standard and experimental treatments are followed by identical subsequent therapies. Therefore, measurements of OS that compare standard and experimental arms without the subsequent therapies are not observed. It is possible that a standard or experimental treatment alone may have improved survival, but the subsequent therapy influenced smaller or larger differences in OS. Intermediate endpoints measured before subsequent therapy may show improved OS, but the authors agree that conclusions cannot be made based on these intermediate endpoints. Using statistical methods, it is possible to calculate a hypothetical OS with no subsequent therapy. However, the study states that this does not occur in a clinical setting and is therefore irrelevant and infeasible to use as a statistical measure for OS. Even if both treatment arms experience increased OS after subsequent therapy, for statistical reasons, a larger sample size is needed to detect whether the experimental treatment is working.
In some trials, patients in the standard treatment arm whose condition worsens can be moved to the experimental treatment. Such cross-over studies pose different approaches to evaluating OS. Novel experimental therapies that have not been tested before usually have high interest from patients, leading to higher participation and enrollment. However, according to the review, this results in underestimating the clinical benefit of the experimental agent, as OS will appear similar in both arms regardless of actual clinical benefit. Methods to more accurately evaluate OS in cross-over situations, such as censoring OS data for all patients at the time of cross-over, can lead to unwanted skews in mortality data and confidence intervals, and ultimately produce inconclusive OS results. Cross-over also occurs when therapy that is effective in later-line treatment is tested as first-line treatment in an RCT. This can lead to an attenuated OS difference; however, the authors argue that this is an appropriate method for assessing the clinical benefit later-line treatments as first-line treatments.
The review concludes with recommendations that intermediate endpoints should not be used as surrogate outcomes for OS, and asserts the importance of following patients to the OS endpoint even if intermediate endpoint results are definitive and released. In addition, analyses of intermediate endpoints need to take into consideration cross-over patients and the consequential impact on OS differences. Predictions of OS should always be made in the context of the subsequent therapy used and not in hypothetical situations where subsequent therapies do not exist. Lastly, the authors recommend that when testing a new experimental therapy, trial designs with standard arm cross-over should be avoided when possible.
Korn EL, Freidlin B, Abrams JS. Overall Survival As the Outcome for Randomized Clinical Trials With Effective Subsequent Therapies. JCO. May 9, 2011: JCO.2011.34.6056; published online on May 9, 2011.
MYTHS VS. TRUTHS
Brush up on your breast cancer myths and truths. Find out if you know more about breast cancer than the American public.
Speak up and out about breast cancer. Say, “I’ll Talk” and commit to a conversation about breast cancer with people you know and care about.