|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2005.92.008 on June 13 2005 © 2005 American Society of Clinical Oncology.
Objective Response Rate As a Surrogate End Point: A CommentaryUniversity of Washington, Seattle, WA Bruzzi et al1 have evaluated objective response rate (ORR) as a potential surrogate end point for survival in metastatic breast cancer patients. Throughout oncology, many biomarkers have been evaluated. Examples include direct measures of the tumor burden process, such as ORR or progression-free survival (PFS), or inherently less reliable indirect measures, such as carcinoembryonic antigen or prostate-specific antigen. Biomarkers can serve several useful purposes.2 These purposes include the following: (1) detecting disease or assessing prognosis, and to serve this purpose, biomarkers, such as carcinoembryonic antigen or prostate-specific antigen, only need to be correlated with disease presence or prognosis, rather than being the causal mechanism through which the disease process induces effects on true clinical end points like survival or disease-related symptoms (in oncology, this causal mechanism usually is the tumor burden process); (2) targeting those patients who will benefit from treatment, and to be truly useful in this context, the biomarker needs to be known before initiation of treatment, such as assessing the level of overexpression of human epidermal growth factor receptor 2 in deciding whether to use trastuzumab; and (3) as a surrogate (or replacement) end point in trials designed to provide conclusive evidence about the benefit-to-risk profile of an intervention. How useful is ORR regarding purposes 2 and 3? Although it is a direct measure of the tumor burden process, ORR could underestimate treatment effects on clinical end points, such as survival, by failing to adequately capture the magnitude, breadth, and, in particular, the duration of effects on tumor burden. Conversely, ORR could overestimate impact on survival or other clinical end points if response is brief or if this measure fails to capture unintended harmful mechanisms of action of treatment. How does one validate a biomarker in general? In particular, in the context of metastatic breast cancer in which survival is the clinical end point, is ORR reliable in addressing the useful purposes identified earlier? Three levels of statistical evidence should be considered. First, is the biomarker correlated with the clinical end point? Such correlation is a necessary but not sufficient condition to establish the validity of a surrogate end point. Simply showing that responders live longer than nonresponders does not rule out the possibility that achievement of response is merely putting a label on those who already would have lived longer, independent of treatment. Second, does evidence suggest that ORR fully captures the effect of treatment on survival, when conducting a Cox regression analysis using ORR as a time varying covariate3-5 or when generating survival curves by tumor response using the landmark method?6 Although these statistical analyses provide evidence concerning the plausibility that ORR is a useful surrogate end point, one must use considerable caution in interpreting these results. The conclusion by Bruzzi et al1 from such analyses that the "survival difference was entirely a result of the effect of experimental FU regimens on response rate" seems to be an overstatement because of several limitations. The first limitation is that these analyses only address whether ORR is capturing the net effect of treatment on survival.4 For example, suppose that the treatment provides a 6% reduction in death rate and that that is exactly what would be expected from the identified effect on ORR. In such settings, treatment might also be providing additional beneficial effects on survival through mechanisms other than induction of response, but these additional benefits may be offset by the treatment's unintended adverse effects on survival that are not captured by ORR. (It also follows that it is unwise to generalize the relationship between effects on ORR and effects on survival found with one class of agents to other classes.) The second limitation is that the effect of treatment might not be a result of ORR but, instead, a result of a causal mechanism that is correlated with the ORR, such as prevention of long-term worsening of tumor burden if that is induced in predominantly the same patients who experience tumor response. The third limitation is the substantial variability in parameter estimates that is inherent with these methods. This has an important impact on the reliability of these methods, particularly in settings such as those chosen by Bruzzi et al,1 where treatment has a small effect on survival. The landmark analyses also have a fourth limitation as result of the risk of bias arising from missing data, exclusion of early deaths, and dependence on strong assumptions regarding independent censoring that are unlikely to hold, and as a result of the compromised interpretability when there is large variability in response time and when the landmark time is substantially different from time of randomization, as is true in the setting considered by Bruzzi et al.1 Third, and most importantly, in a meta-analysis of trials evaluating a class of agents in a clinical setting, does the treatment effect on ORR reliably predict the effect on survival? Such analyses provide the principal information regarding whether biomarkers are valid surrogate end points. On the basis of these considerations and the results of Bruzzi et al,1 several conclusions seem to be justified. First, assuming they avoided publication bias in the selection of their 10 randomized trials comparing standard with intensified epirubicin-containing chemotherapy, Bruzzi et al1 provide important insights into the usefulness of ORR for evaluating such regimens in the metastatic breast cancer setting through their recognition that such assessments should be based on this type of meta-analysis. Second, the meta-analysis suggests that the experimental regimen provides a statistically nonsignificant modest benefit of only a few weeks in duration of survival and statistically rules out a 14% reduction in death rate. In the context of the toxicity, inconvenience, and cost of the regimen, why are such effects clinically meaningful? These modest effects on the clinical end point of interest also reduce the relevance and reliability of analyses regarding the validity of surrogate end points in such settings. Third, regarding the useful purpose of targeting those patients who will benefit from treatment, results by Bruzzi et al1 from their landmark analysis (as shown in Fig 3) do not allow one to determine who should receive the experimental intensified epirubicin-containing chemotherapy, even though they suggest that nonresponders are unlikely to receive survival benefit. Unfortunately, unlike the trastuzumab example, in which levels of overexpression of human epidermal growth factor receptor 2 can be identified before initiation of treatment, one must receive several months of the intensified epirubicin regimen before nonresponse can be established. Fourth, the most important insights in the Bruzzi et al1 article, provided in Figure 4, indicate a low correlation between the intensified epirubicin regimen's effects on ORR and on survival, with an R2 = 0.10. In contrast, a meta-analysis regarding the validity of disease-free survival as a surrogate end point for survival for fluorouracil-based regimens in the colon adjuvant setting based on 18 randomized comparisons yielded an R2 of 0.87.2,7 Bruzzi et al1 appropriately conclude that their results provide evidence "against the use of objective response as the primary end point in clinical trials," and that "response could not be considered a valid surrogate for survival for the purpose of testing new treatments." Finally, Bruzzi et al1 provide the important recognition that analyses regarding the validation of surrogate end points are "specific to well-defined disease, clinical outcome, and treatment." Hence, one should be cautious about the generalization of such analyses to other settings. Author's Disclosures of Potential Conflicts of Interest The author indicated no potential conflicts of interest. REFERENCES 1. Bruzzi P, Del Mastro L, Sormani MP, et al: Objective response to chemotherapy as a potential surrogate end point of survival in metastatic breast cancer patients. J Clin Oncol 23: 5117-5126, 2005
2. Fleming TR: Surrogate end points and FDA's accelerated approval process. Health Aff 24: 67-78, 2005 3. Freedman LS, Graubard BL, Schatzkin A: Statistical validation of intermediate end points for chronic diseases. Stat Med 11: 167-178, 1992[Medline] 4. DeGruttola V, Fleming TR, Coombs R, et al: Validating surrogate markers: Are we being naïve? J Infect Dis 175: 237-246, 1997[Medline] 5. Lin DY, Fleming TR, DeGruttola V: Estimating the proportion of treatment effect explained by a surrogate marker. Stat Med 16: 1515-1527, 1997[CrossRef][Medline] 6. Simon R, Makuch RW: A nonparametric graphical presentation of the relationship between survival and the occurrence of an event: Application to responders versus nonresponders bias. Stat Med 3: 35-44, 1984[Medline] 7. Sargent D: Disease-free survival versus overall survival as an end point for adjuvant colon cancer studies: Data from randomized trials. Meeting Transcript of the United States Food and Drug Administration Oncology Drugs Advisory Committee, Washington DC, March 12-13, 2003 Related Article
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2005 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
