|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 25, No 33 (November 20), 2007: pp. 5153-5154 © 2007 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.13.6796
Toward Progression-Free Survival As a Primary End Point in Advanced Colorectal CancerDepartment of Biostatistics, University of Pittsburgh; and the National Surgical Adjuvant Breast and Bowel Project Biostatistical Center, Pittsburgh, PA In this issue of the Journal of Clinical Oncology, Buyse et al1 evaluate progression-free survival (PFS) as a surrogate end point for overall survival (OS) in advanced colorectal cancer (CRC). They employ a general methodology for validating surrogate end points across multiple trials, as proposed by Buyse et al2 in 2000. They considered trials where leucovorin-modulated fluorouracil (FU) was compared with either FU without leucovorin or with raltitrexed. In these trials, they were able to show modest correlation between the proportion alive and progression free at 6 months and the proportion alive at 12 months, as well as good correlation when data across all time were considered. More importantly, they showed that the effect of treatment on each of these end points was highly correlated both when data across all time were considered and when the PFS and OS data were censored at 6 and 12 months, respectively. This latter step is an important advance over previous attempts to establish surrogacy.3 The model relating treatment effects between the PFS and OS end points was validated on a set of three contemporary trials in which leucovorin-modulated FU was compared with regimens that added irinotecan or oxaliplatin. The model was used to predict the treatment effect on OS given the observed treatment effect on PFS. The observed treatment effect on OS for the validation trials fell within the prediction limits for all three trials. The authors conclude that PFS is an acceptable surrogate for OS in advanced CRC and provide a surrogate threshold effect (STE), which is the treatment effect size for PFS which, if exceeded, would rule out equivalence with 95% confidence for the OS end point based on the model's prediction limits. PFS has now been validated as a surrogate end point for OS in advanced CRC, but OS is not a perfect end point for this disease. In advanced CRC, the goal of therapy in order of desirability of outcome is to eliminate cancer, reduce cancer burden, stabilize cancer, or slow cancer progression. Prolongation of life is secondary to the goal of therapy. As more and better options for second- and third-line therapies in advanced CRC become available, the relevance of OS as the primary end point is increasingly coming into question.1,3 Let's take a moment to compare and contrast the virtues of PFS and OS as potential primary end points for first-line treatment of metastatic CRC. OS is usually defined as time to death from any cause and may be considered a composite end point based on the cause of death where the event is the first of death related to the same cancer, death related to another cancer, death related to treatment, or death resulting from other causes. PFS is a composite end point where the event is the first of progression resulting from increasing size of tumor, progression resulting from formation of a new tumor, death related to the same cancer, death related to another cancer, death related to treatment, or death resulting from other causes. The component events are listed in Table 1 for the PFS and OS end points along with their relationship to the disease process under study or the therapies being studied.
Events related to the disease process under study or to treatment toxicity provide important information for comparing potential therapies. Events that are unrelated to the disease process under study or to treatment toxicity are essentially background noise and provide little useful information for comparing potential therapies. We don't want to ignore these likely irrelevant events, because sometimes treatments have unintended consequences. The underlying rates of these so-called background events should be equivalent on the treatment arms, because they are by definition unrelated to the disease process or treatments. Some desirable attributes for a primary end point in advanced CRC are listed in Table 2. Event status determination is extremely reliable in OS. The difficulties associated with determining progression status lead to a less reliable determination of event status with PFS. This deficiency for PFS can be mitigated by clear protocol criteria for progression and consistent access to high-quality imaging for all trial participants. Time to event can be ascertained with excellent reliability for OS, but the difficulties in ascertaining time to progression lead to less reliability for PFS. Reliability of time to progression can be improved by incorporating frequent imaging into the protocol and minimizing any potential treatment-related bias in assessment for progression.
The component events listed in Table 1 can be thought of as competing events. If we follow patients long enough, they will all experience an event. Hypothetically, a patient could experience progression at 6 months, die as a passenger in a car accident at 8 months—totally unrelated to CRC or treatment—and perhaps would have died at 12 months as a result of CRC had it not been for the car accident. For the OS end point, the potentially informative death from CRC is masked by an uninformative death from a car accident. But for PFS, we have an informative event at 8 months with progression. Some would argue for central review of events to determine relationship to disease or treatment and for censoring or ignoring of unrelated events. I believe the relationship to the disease process or treatment is too difficult to classify for some events. Consider our hypothetical patient again. What if this patient were driving the car when the accident occurred? The disease process or treatment toxicity could conceivably have influenced the patient's reaction time and contributed to causing the death. Eliminating reviews of relationship to the disease process or treatment also eliminates a potential source of bias. Although we don't generally know whether a particular event is informative or not, we want to design an end point so that a high proportion of events are likely informative. PFS counts progression events and OS does not. Progression events are likely related to the disease process, and some of these progression events will mask future deaths that are unrelated to the disease process or treatment toxicity; thus, PFS will generally have a higher proportion of informative events than will OS. Because only informative events are likely to be influenced by therapy, PFS should be a more sensitive indicator of treatment effect. Although this may presently be only a minor disparity between PFS and OS in advanced CRC, it promises to become an increasing disparity as survival times are extended in the future. The sensitivity of OS to second- and third-line therapies has become increasingly problematic with the recent proliferation of effective therapies for advanced CRC.1,3 PFS is less sensitive to subsequent therapy because most patients who go on to receive second- or third-line therapy will have already experienced a progression event. This is perhaps the most significant difference between PFS and OS. Buyse et al1 have shown that PFS is a valid surrogate for OS, but I believe PFS is a reasonable primary end point on its own merits. Although OS may be more reliable in classifying event status and quantifying time to event, PFS should be a more sensitive indicator of treatment effect because of its higher proportion of informative events, and should be less sensitive to subsequent therapy outside the protocol. PFS has the further advantage of being a more timely end point, with median PFS being reached in about 6 months whereas median OS is not reached until 12 months.1 Weighing all the pros and cons, I believe we will increasingly see PFS adopted as a primary end point in advanced CRC. AUTHOR'S DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST The author(s) indicated no potential conflicts of interest.
REFERENCES
1. Buyse MM, Bursykowski T, Carroll K, et al: Progression-free survival is a surrogate for survival in advanced colorectal cancer. J Clin Oncol 25:5218-5224, 2007 2. Buyse M, Molenberghs G, Burzykowski T, et al: The validation of surrogate endpoints in meta-analyses of randomized experiments. Biostatistics 1:49-67, 2000[Medline] 3. Louvet C, de Gramont A, Tournigand C, et al: Correlation between progression free survival and response rate in patients with metastatic colorectal carcinoma, Cancer 91:2033-2038, 2001[CrossRef][Medline]
Related Article
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2007 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
