Applying Sensitivity Analyses to Overall Survival in Cancer Clinical Trials

Abstract

TO THE EDITOR: “Role of Sensitivity Analyses in Assessing Progression-Free Survival in Late-Stage Oncology Trials” by Bhattacharya et al is an important statistical technique that assesses whether the results of phase III trials are robust and likely to be generalizable. The authors focus their comments and state that it is “important to conduct additional analyses (sensitivity analyses) of studies where progression-free survival (PFS) or time to tumor progression (TTP) is the primary end point.” Thus, implied in their emphasis is that overall survival (OS), the ultimate gold-standard outcome (the time from random assignment to death), may not be subject to confounding issues and does not require similar consideration. Bhattacharya et al state that “the documented date of first progression is often after the actual true (unknown) date of progression, which is not the case for OS.” However, they do not consider events that could similarly affect the timing of the OS primary event. Death, although an objective and measurable event, is affected by various clinical acts that can substantially influence this outcome. It is not the issue of when death occurred but various influences that affect the outcome. For instance, similar patients failing interventional therapy at the same time point after therapy may be kept alive (eg, avoiding the primary outcome) by heroic measures such as intensive care unit care or simple measures such as treating a pneumonia, or some may have only basic noninterventional hospice/end-of-life therapy offered that does not artificially prolong survival. Additionally, patients with bulky tumors compared with those with trivial tumor burdens, similarly failing therapy at the same time point without further therapy, may have different survival outcomes unrelated to the primary interventional therapy. It is therefore equally important for sensitivity testing to be pursued when phase III studies are positive for OS when differences are statistically different, but not robust biologically, because it is becoming more common for outcomes to be measured in a few weeks or months when large numbers of patients are entered on a trial. This is especially important when drugs or biologic are approved for generalizable intervention based on modest time increments for OS.