Hidden Biases in an Observational Study of Bevacizumab Beyond Progression

TO THE EDITOR: Grothey et al 1 recently reported a prospective observational study that monitored oncologists’ use of bevacizumab after patients experienced disease progression while receiving a prior bevacizumab-containing regimen. The conclusion formed on the basis of this study was that there was a suggested survival benefit from continuing bevacizumab after failure of a first-line bevacizumabcontaining regimen, with a hazard ratio (HR) of 0.46 on multivariate analysis. Generally, observational studies can provide important data on toxicities and practice patterns, but these types of studies have significant limitations when they are used to answer comparative questions of efficacy. Grothey et al 1 clearly acknowledge the potential bias with this study design and attempt to adjust for several factors. Unfortunately, it is impossible to quantify the extent to which this bias formed or altered the outcome of the trial. Several important sources of bias cannot be controlled for in the absence of randomization to treatment in this study, as will undoubtedly be delineated in this space by other readers of this journal. As a result, factors that are correlated with survival also affect the oncologist’s decision to continue bevacizumab. Several poor prognosis factors that may decrease an oncologist’s enthusiasm to continue bevacizumab include poor performance status at progression, higher vascular comorbidities, rapid tumor growth while receiving the prior bevacizumab-containing regimen, or peritoneal disease. Each of these factors would generally have predicted shorter survival in patients in whom bevacizumab was discontinued. However, in lieu of enumerating the number and reasoning for such biases in greater detail, we conducted similar analyses that may help to highlight the extent to which the conclusions in this observational trial may have been formed by these unseen biases. Since the US Food and Drug Administration’s approval of bevacizumab in 2004, most previously untreated metastatic colorectal patients at our institution have been treated with a bevacizumab-containing regimen. We chose to retrospectively examine these sequential patients to determine if patients treated again with bevacizumab as part of a later-line regimen had improved outcomes than those who did not receive a later-line bevacizumab-containing regimen. Indeed, we were able to demonstrate a similar HR of 0.62 (P .01) in favor of patients receiving bevacizumab after failing a prior bevacizumab-containing regimen.