|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 25, No 22 (August 1), 2007: pp. 3191-3193 © 2007 American Society of Clinical Oncology. DOI: 10.1200/JCO.2007.11.5089
Sound Footing or Slippery Slope? The Value of Secondary Analyses of Randomized TrialsMayo Clinic, College of Medicine, Rochester, MN The accepted standard for the adoption of a new therapy, whether a drug, device, or treatment strategy, is the randomized phase III clinical trial. The justification for basing patient care decisions on such rigorous investigations is clear. These trials have prospectively specified hypotheses; limit bias through well-defined patient selection, random assignment, and blinding (when possible); have clearly defined primary and secondary end points; and have analysis plans that are prespecified. However, the cost, complexity, and time to perform such trials, as well as practical and ethical considerations, are such that not every question can be addressed in a prospective phase III randomized trial. Because datasets from randomized clinical trials contain vast amounts of information, additional hypotheses can be generated, optimally to be explored in subsequent trials. Thus, techniques such as meta-analyses, retrospective analyses, and statistical modeling are increasingly being used to enhance our base of knowledge. On the basis of the results of several phase III trials, a combination regimen of infusional fluorouracil (FU) with leucovorin plus oxaliplatin has emerged as a standard of care for first-line chemotherapy in advanced colorectal cancer.1,2 The duration of oxaliplatin therapy is commonly limited by its well-known adverse effect, cumulative sensory neurotoxicity. Although the median progression-free survival time of FU with leucovorin plus oxaliplatin in first-line advanced colorectal cancer is approximately 9 months, median time to treatment failure, the time when patients discontinue oxaliplatin-based chemotherapy, has consistently been 5 to 6 months. In fact, more than 60% of patients discontinue oxaliplatin as a component of first-line therapy for reasons other than progressive disease,3 implying that many patients stop oxaliplatin for toxicity reasons when the drug may still have activity against their cancer. Although oxaliplatin-induced neurotoxicity is reversible in principle, it might take months to years until resolution is achieved. Thus, patients and physicians might be reluctant to reuse oxaliplatin on progression of a potentially still oxaliplatin-sensitive tumor. To address the clinical value of reintroduction of oxaliplatin as a component of therapy, in this issue, de Gramont et al4 present analyses using data from the OPTIMOX1 trial, which randomly assigned patients to continuous versus intermittent (stop-and-go) oxaliplatin in first-line treatment.5 The primary results from this previously published important trial demonstrated that stopping oxaliplatin after a fixed number of treatment cycles, with reintroduction at disease progression after an FU-based maintenance treatment phase, resulted in no decrement in efficacy and improved tolerability (decreased grade 3 neuropathy). However, protocol adherence was variable among patients and treating locations, such that in the stop-and-go arm, only 40% of patients had oxaliplatin reintroduction per protocol. Is there a method to validly use the data from the OPTIMOX1 study to estimate what patient outcome might have been if protocol compliance was improved? Analyses seeking to address questions of how a therapy would work if the protocol were followed are simultaneously extremely important and extremely challenging. There is an entire field of statistical theory, known as causal inference, devoted to this subject.6 The challenges with any such analysis are the known and unknown biases that may have limited physician or patient protocol compliance. Among other reasons, these biases are why the intent-to-treat approach is the preferred primary analysis approach to phase III trials. de Gramont et al4 did not use the causal inference methodology, which is controversial, but instead used innovative approaches based on extensions to the Cox proportional hazard regression model.7 In one approach, the authors used time-varying covariates (covariates whose values are allowed to change in the model over time) as well as extensive baseline adjustments to attempt to remove, as much as possible, patient selection issues and length-biased sampling. In this analysis, they found that, as expected, irinotecan introduction was associated with improved outcome (survival) and that oxaliplatin reintroduction was associated with improved survival as well. These findings held true overall but were also true for patients on the continuous treatment arm, where more patients would have stopped treatment as a result of progression. There are several limitations to definitive conclusions based on these findings. First, only baseline patient characteristics, not those at the time of therapy reinitiation, were included in the model. Second, analyses are not presented for the subset of patients on the control arm who did truly experience progression while on oxaliplatin (as opposed to patients who discontinued treatment as a result of toxicity), limiting our ability to evaluate the value of oxaliplatin reintroduction in patients who experienced progression through active oxaliplatin treatment. Third, because both reintroduction of oxaliplatin and introduction of irinotecan were prognostic factors for patient benefit, despite the covariate adjustments, it is impossible to definitively conclude from this analysis whether oxaliplatin reintroduction itself is helpful or whether it is a marker of a good prognosis patient who would have a better outcome regardless of what therapy was administered. Insight into this question could have been obtained by including other (nonirinotecan, nonoxaliplatin) postprotocol therapy, which was used in 30% to 40% of patients, with regimens and agents known to have little second- or third-line activity (eg, FU, capecitabine, uracil-tegafur). If use of any postprotocol therapy, including these less effective therapies, was a beneficial prognostic factor, it would suggest a patient selection effect, rather than a therapeutic effect. However, the lack of a benefit of less effective therapies would have strengthened the conclusion that it is specific therapy, not patient selection, that is of greatest importance. A second approach described in the article by de Gramont et al4 used a treating center–specific effect, as opposed to a patient-specific effect, based on the likelihood of a patient to be rechallenged with oxaliplatin. This analysis is founded on the idea that reintroduction of oxaliplatin per protocol may have been dictated more by practice at different treatment centers than by patient-level decision. The authors compared patient outcomes based on the propensity for a treating institution to reintroduce oxaliplatin and found a significant and consistent trend toward improved survival in patients treated at centers where a higher proportion of patients had oxaliplatin reintroduced. This analysis clearly suggests that a strategy of oxaliplatin reintroduction is in the patient's best interest. However, the potential for bias is present in this approach as well. Was the propensity to reintroduce oxaliplatin the only factor that differentiated sites, or did sites also have differences in patient mixes, frequency of follow-up, supportive care, and access to experimental third- and fourth-line treatment? The Cox model approach used allowed for the adjustment of some confounding factors, but site-level covariates, such as those mentioned, were not included in the model. On the basis of these findings and caveats, what can we conclude? Is access to all active agents important? Almost certainly. Note that, in the study by de Gramont et al,4 the impact on survival of second-line irinotecan (hazard ratio = 0.40) was greater than the impact of oxaliplatin reintroduction (hazard ratio = 0.56). Is stop-and-go oxaliplatin a sound clinical strategy? Again, almost certainly. These results strongly support the primary OPTIMOX1 findings that showed no efficacy decrement to pausing and reinitiating oxaliplatin and suggest that there may even be some benefit to such a strategy. Is oxaliplatin reintroduction beneficial to a nonprotocol patient previously exposed to oxaliplatin? Likely. The magnitude of effect observed and the consistency seen in the various analyses, each with their own potential biases, are strongly suggestive. Is oxaliplatin reintroduction beneficial in a patient whose tumor has progressed on oxaliplatin at an earlier time point, and does the time point of oxaliplatin reintroduction (before or after irinotecan) matter? The answers to these questions are unclear because these analyses were not reported. These questions also highlight current dilemmas in the design and interpretation of clinical trials that test the sequential administration of drugs and the implementation of (however defined) treatment holidays. Such trials defy the conventional definition of lines of therapy and challenge commonly used end points such as progression-free survival to describe patient benefit from sequencing strategies. Novel, innovative end points are warranted to capture the effects of complex therapeutic interventions in an era where the idea of lines of therapy is becoming increasingly obsolete and treatment holidays are increasing, even if we are not all in agreement as to what constitutes an optimal treatment break. Well-conducted secondary analyses of randomized clinical trials are critical to aiding our understanding of important clinical questions that may require years before (if ever) they are prospectively addressed. de Gramont et al4 are to be congratulated on a fine example of such work, which was the joint effort of multiple clinicians and statisticians. Findings based on such analyses must be interpreted with the appropriate caveats, and as the authors correctly acknowledge, the level of evidence is clearly below that from the prospective trial. However, pressing clinically relevant questions demand that we make optimal use of available data to provide whatever insights may be possible. For example, recent analyses of data collected in a nonrandomized fashion addressed whether continuation of treatment with bevacizumab beyond first-line progression provides clinical benefit to patients with advanced colorectal cancer. These analyses, reported only in abstract form but using many methods similar to those of de Gramont's work, suggested that this may indeed be the case.4,8 The concept of continuation of bevacizumab beyond first-line progression is currently being tested prospectively in the ongoing randomized phase III Intergroup trial S0600. The following strategy of trial development is, in our opinion, optimal practice: retrospective analyses presenting hypotheses to be tested in a randomized trial. In many cases, however, such a randomized trial is years from providing data or may never be performed. It is our opinion that, in the absence of randomized data on a certain issue, retrospective, exploratory analyses, with all their limitations, are preferable to obtain information for a clinical decision, rather than refusing to offer any opinion, provided that these analyses are accompanied by a full discussion of the limitations and are appropriate in the strength of conclusions that can be drawn. This is particularly important at a time when the number of potential treatment options vastly exceeds the capability of the current clinical trials system to provide phase III level of evidence for all pertinent, clinically relevant questions. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a "U" are those for which no compensation was received; those relationships marked with a "C" were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors. Employment or Leadership Position: None Consultant or Advisory Role: Daniel Sargent, Sanofi-Aventis (C), Genentech (C), Pfizer Inc (C), Roche (C), Amgen (C); Axel Grothey, Sanofi-Aventis (C), Genentech (C), Bristol-Myers Squibb Co (C), Roche (C), Amgen Inc (C) Stock Ownership: None Honoraria: Daniel Sargent, Sanofi-Aventis, Genentech, Roche, Pfizer Inc, Amgen; Axel Grothey, Sanofi-Aventis, Roche, Genentech, Bristol-Myers Squibb Co, Amgen Inc Research Funding: None Expert Testimony: None Other Remuneration: None AUTHOR CONTRIBUTIONS Conception and design: Daniel Sargent, Axel Grothey Administrative support: Daniel Sargent Data analysis and interpretation: Daniel Sargent, Axel Grothey Manuscript writing: Daniel Sargent, Axel Grothey Final approval of manuscript: Daniel Sargent, Axel Grothey REFERENCES
1. de Gramont A, Figer A, Seymour M, et al: Leucovorin and fluorouracil with or without oxaliplatin as first-line treatment in advanced colorectal cancer. J Clin Oncol 18:2938-2947, 2000 2. Goldberg RM, Sargent D, Morton RF, et al: N9741: FOLFOX (oxaliplatin (Oxal)/5-fluorouracil (5-FU)/leucovorin (LV)) or reduced dose R-IFL (CPT-11 + 5-FU/LV) in advanced colorectal cancer (CRC): Final efficacy data from an intergroup study. J Clin Oncol 22:275, 2004 (suppl, abstr 3621) 3. Green E, Sargent D, Goldberg RM, et al: Detailed analysis of oxaliplatin-associated neurotoxicity in Intergroup trial N9741. Ann Oncol 15:70, 2004 (suppl 3, abstr 2540) 4. de Gramont A, Buyse M, Cortinas Abrahantes J, et al: Reintroduction of oxaliplatin is associated with improved survival in advanced colorectal cancer. J Clin Oncol 25:3224-3229, 2007 5. Tournigand C, Cervantes A, Figer A, et al: OPTIMOX1: A randomized study of FOLFOX4 or FOLFOX7 with oxaliplatin in a stop-and-go fashion in advanced colorectal cancer—A GERCOR study. J Clin Oncol 24:394-400, 2006 6. Bellamy SL, Lin JY, Ten Have TR: An introduction to causal modeling in clinical trials. Clin Trials 4:58-73, 2007 7. Cox DR: Regression models and life tables. J R Stat Soc B 34:187-220, 1972 8. Grothey A, Sugrue M, Hedrick E, et al: Association between exposure to bevacizumab (BV) beyond first progression and overall survival (OS) in patients (pts) with metastatic colorectal cancer (mCRC): Results from a large observational study (BRiTE). J Clin Oncol 25:18s, 2007 (suppl; abstr 4036)
Related Article
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2007 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
