Disease-Free Survival Versus Overall Survival As a Primary End Point for Adjuvant Colon Cancer Studies: Individual Patient Data From 20,898
http://www.100md.com
《临床肿瘤学》
the North Central Cancer Treatment Group, Mayo Clinic, Rochester, MN
National Surgical Adjuvant Breast and Bowel Project Statistical Center and Operations Office, Pittsburgh
Abramson Cancer Center, University of Pennsylvania, Philadelphia, PA
QUASAR Collaborative Group, Birmingham and Oxford, United Kingdom
Southwest Oncology Group Statistical Center, Seattle, WA
International Drug Development Institute, Brussels, Belgium
Ospedali Riuniti, Bergamo
University of Siena, Siena, Italy
University of the Mediterranean, Marseilles
Groupe d'Etude et de Recherche Clinique en Oncologie et Radiothérapie, Paris
Hospital Saint Antoine, Paris, France
National Cancer Institute of Canada Clinical Trials Group, Queens University, Kingston, Ontario, Canada
Eastern Cooperative Oncology Group Statistical Center, Boston, MA
Oregon Health and Science University Cancer Institute, Portland, OR
University of North Carolina, Chapel Hill, NC
ABSTRACT
PURPOSE: A traditional end point for colon adjuvant clinical trials is overall survival (OS), with 5 years demonstrating adequate follow-up. A shorter-term end point providing convincing evidence to allow treatment comparisons could significantly speed the translation of advances into practice.
METHODS: Individual patient data were pooled from 18 randomized phase III colon cancer adjuvant clinical trials. Trials included 43 arms, with a pooled sample size of 20,898 patients. The primary hypothesis was that disease-free survival (DFS), with 3 years of follow-up, is an appropriate primary end point to replace OS with 5 years of follow-up.
RESULTS: The recurrence rates for years 1 through 5 were 12%, 14%, 8%, 5%, and 3%, respectively. Median time from recurrence to death was 12 months. Eighty percent of recurrences were in the first 3 years; 91% of patients with recurrence by 3 years died before 5 years. Correlation between 3-year DFS and 5-year OS was 0.89. Comparing control versus experimental arms within each trial, the correlation between hazard ratios for DFS and OS was 0.92. Within-trial log-rank testing using both DFS and OS provided the same conclusion in 23 (92%) of 25 cases. Formal measures of surrogacy were satisfied.
CONCLUSION: In patients treated on phase III adjuvant colon clinical trials, DFS and OS are highly correlated, both within patients and across trials. These results suggest that DFS after 3 years of median follow-up is an appropriate end point for adjuvant colon cancer clinical trials of fluorouracil-based regimens, although marginally significant DFS improvements may not translate into significant OS benefits.
INTRODUCTION
The benefit of adjuvant chemotherapy on overall survival (OS) for patients with node-positive, locally advanced colon cancer is well established.1 In a pooled-analysis, fluorouracil (FU)-based chemotherapy was shown to increase the 5-year OS from 64% to 71% in a mixed group of node-negative (American Joint Committee on Cancer stage II) and node-positive (American Joint Committee on Cancer stage III) patients.2 The benefit of treatment in patients with stage II disease has been less clear, and current American Society of Clinical Oncology guidelines do not recommend the routine use of adjuvant chemotherapy for all patients with stage II disease.3 However, two large studies (one randomized clinical trial4 and one pooled analysis of seven randomized clinical trials5) have both suggested a modest but consistent treatment benefit from FU-based chemotherapy in these patients.
Clearly, long-term cure is the most unambiguous, and ultimately important, outcome for colon cancer patients. The goal of improving long-term outcome has translated into the use of OS as the primary end point for most stage II and stage III colon cancer adjuvant therapy trials. Historically, the OS rate at 5 years has been the most typically quoted metric for judging the success of a particular treatment regimen. This end point has the advantage of being simple to measure, easy to interpret, clinically meaningful, and straightforward to explain. The disadvantage of this end point is that it requires extended follow-up. An end point that is reached more rapidly has obvious advantages, including but not limited to answering the question posed by the trial more rapidly and, potentially, quicker drug approval and availability. The goal of this study is to explore disease-free survival (DFS) as a possible alternative end point to OS, in clinical trials testing adjuvant treatment for patients with stages II and III colon cancer.
Considerable attention has been paid to the field of surrogate markers, with some conflict as to the most appropriate definition of an adequate surrogate.6-8 The most widely cited criteria are proposed by Prentice,6 who stated that for a surrogate marker to be valid, it must be correlated with the true end point, and it must fully capture the net effect of the treatment on the true end point. These criteria have been criticized for several reasons, one of which is that they are too stringent to be valid in practice. More recently, approaches have been proposed suggesting that for a surrogate to be appropriate, the effect of treatment on the surrogate end point must predict the effect of treatment on the true end point or that a test of a statistical hypothesis based on the surrogate end point should come to the same conclusion as a statistical hypothesis test based on the primary end point.8,9 This article will not attempt to resolve any definitional issues regarding surrogate markers, rather it will present data designed to evaluate the appropriateness of alternative end points for future adjuvant clinical trials in colon cancer through a data-based approach.
METHODS
Through contact with principal investigators, we combined individual patient data from large randomized phase III trials in adjuvant colon cancer. This is not a formal meta-analysis in that an exhaustive attempt to include every completed colon adjuvant trial was not undertaken. Trials were selected based on relevance, maturity, and data availability. For each trial, following the original protocol, one arm was defined as the control arm, even if chemotherapy was used in all patients. Two trials included a small number of patients with high rectal tumors; these patients are included in this analysis. OS was defined as the time from randomization to death from any cause. DFS was defined as the time from randomization to the first event of either recurrent disease or death. Second primary tumors were not counted as events in the DFS analysis. All randomly assigned patients were included in the analyses according to the intention to treat principle. This retrospective pooled analysis was institutional review board–approved; each trial included was originally approved through appropriate local mechanisms.
The long-term follow-up practices of the various studies differed. To reduce the possibility of bias owing to this differential practice, the follow-up data for all patients were censored at 8 years from randomization. For all analyses, efforts were made to replicate the actual conduct of a clinical trial, where patients enter over a period of several years, and thus at any given calendar date have differing durations of follow-up. For each analyses conducted at a specific time point (eg, 3 years), the outcome data for each patient were censored at the point in time at which the median follow-up in the trial was 3 years. For example, if a trial accrued evenly over a period of 3 years, then the median follow-up would be 3 years after 1.5 years had elapsed from the close of randomization. At this point, the first patient enrolled would have 4.5 years of follow-up, whereas the last patient enrolled would have 1.5 years of follow-up. Analyses were repeated after applying a censoring pattern 3 years after the date of the last patient randomization to that trial. These two analyses gave similar conclusions; we therefore present the results with 3 years of median follow-up.
Multiple analysis techniques were used. For several end points, simple percentage agreement was calculated based on the individual patient data. Hazard ratios comparing each experimental arm with the control arm in that trial were calculated using the Cox proportional hazards regression model10 separately for each individual trial. Statistical significance of the comparison between the experimental and control regimen of each trial for each end point was evaluated using the log-rank test, with a two-sided P < .05 used to denote statistical significance. The relationship between various end points was estimated using weighted linear regression, with weights equal to the sample size of the trial from which the data were derived.
Because of the evolving literature on surrogate end points, formal testing for surrogacy was also explored with multiple criteria. The first approach followed Freedman et al,11 fitting two Cox proportional hazards models and calculating the proportion of net treatment effect explained by the surrogate end point. The second approach followed Burzykowski et al,12 where both the trial level and individual level association between the proposed surrogate and the true end point is estimated using a bivariate survival model. Both of these measures are based on a 0 to 1 scale, with values close to 1 indicating effective surrogacy. In addition, we explored the proportion of agreement of trial results, as suggested by Begg and Leung,9 where the proportion of times the trials reached the same conclusion based on statistical significance testing for the two end points (DFS and OS) was calculated. The use of this variety of methods was designed to make our findings robust to the particulars of any specific approach, which each have advantages and drawbacks.
To test the validity of the model associating within-trial hazard ratios for DFS and OS, leave-one-out cross-validation was performed.13 In short, the regression model was repeatedly fit using data from all trials except the one left out, then used to predict the results for the trial for which data were not included in the model fitting. Ninety-five percent prediction intervals were computed for each excluded trial, and the actual data were compared with the prediction to explore the accuracy of the prediction.
RESULTS
Patients and Trials
A total of 18 randomized phase III clinical trials including 43 different treatment arms were included in the analysis (Table 1). Two trials included a levamisole-alone control arm15,18; these two arms were excluded from the analysis. The total sample size is 20,898 patients. The median follow-up on living patients was 8 years; 5-year data were available for 89% of patients. The period of trial enrollment spanned from 1977 to 1999. All trials included at least one arm with an FU-containing regimen. Nine trials included a no-treatment control arm.
Patient characteristics are outlined in Table 2. Sixty-six percent of patients were stage III, 55% were male, and 19% were older than 70 years of age. Eighty-eight percent of patients were randomly assigned to a chemotherapy-containing treatment arm.
Per-Patient Measures
The recurrence rates for years 1 through 5 were 12%, 14%, 8%, 5%, and 3% respectively. Eighty percent of patients who experienced disease recurrence during the 8-year follow-up period experienced recurrence within the first 3 years after study enrollment. The median time from the diagnosis of recurrent disease to death was 12 months. Ninety-one percent of patients who experienced a recurrence within 3 years of randomization died before 5 years from randomization.
Per–Study Arm Measures
The relationship between the DFS rate after 3 years of median follow-up and the OS rate after 5 years of median follow-up by study arm is shown in Figure 1. The R2 value from a weighted linear regression is 0.85; the Spearman rank correlation coefficient between the two end points is 0.88. For 33 of the 43 study arms, the difference between the 3-year DFS and the 5-year OS was 3%. The largest single difference in the two measures for any single study arm was 8%. The weighted linear regression for the two end points is shown in Figure 1; the SE for the intercept is 0.047 and for the slope is 0.068. These two coefficients do not differ significantly from 0 and 1 respectively (P > .20).
Per-Study Measures
The relationship between the within-study hazard ratio comparing the experimental and the control arms for DFS and OS is shown in Figure 2. The R2 value from a weighted linear regression is 0.90; the Spearman rank correlation coefficient between the 2 end points is 0.94. The weighted linear regression for the two end points is as follows: the hazard ratio for OS = 0.12 + 0.89 x (hazard ratio for DFS), the SE for the intercept is 0.056, and the SE for the slope is 0.061. These two coefficients do differ significantly from 0 and 1 respectively (P = .03 and < .0001, respectively) and indicate a slight attenuation of effect on the hazard ratio between the two end points (Table 3).
Surrogate End Point Measures
The proportion of treatment effect of chemotherapy on the end point of OS explained by DFS was greater than 100%. This curious result is due to the role of recurrence as such a dominant predictor of death that the coefficient for treatment in the Cox model changed from negative to positive when recurrence was included in the model. The individual level correlation between the two end points, estimated using a bivariate survival model, is 0.873 (95% CI, 0.869 to 0.877), and the trial level correlation, estimated from the same model, is 0.78 (95% CI, 0.60 to 0.96). Of 25 within-trial comparisons (control v treatment within a trial), using P < .05 as a measure of concordance, 23 of 25 trials obtained the same conclusion regardless of end point (no difference between arms for 18 within-trial comparisons for either end point; five within-trial comparisons were significant for both end points). In the two trials with discordance, the P values for DFS were .005 and .038 and for OS were .26 and .12, respectively.
Model Validation
The results of the leave-one-out cross-validation for the fitted weighted regression model for within-trial hazard ratios are shown in Figure 3. Twenty-four of the 25 observed OS hazard ratios fell within the 95% prediction intervals, with approximately an equal number of overprediction and underpredictions, indicating excellent model calibration.
DISCUSSION
This analysis was initiated to address a simple question: Can we make a decision regarding the efficacy of a new adjuvant colon cancer treatment regimen more quickly? The implications of such a finding, if confirmed, would be several. Clinical trials could be completed more quickly with reduced cost, drug development time could be shortened, and, most importantly, promising new agents or combinations of agents could be made available to patients more rapidly. In the setting of FU-based treatments for adjuvant colon cancer, the answer to the question seems to be yes.
During the first 8 years after surgical resection, 80% of patient deaths are preceded by a tumor recurrence.2 On an individual patient basis, in this analysis, 80% of patients who experienced disease recurrence experienced their recurrence within 3 years of study entry. During the time in which these trials were conducted (1977 through 1999), the efficacy of postrecurrence chemotherapy was limited, with a median time from recurrence to death of 12 months, which is consistent with that reported in most phase III advanced colon cancer trials conducted before the late 1990s. Thus by 3 years from study entry, most of the recurrences that were to occur had occurred, and by 5 years, the vast majority of these patients with recurrent disease had died. In addition, based on a bivariate survival model,12 the patient level correlation between the two end points of DFS and OS is 0.87, which is an extremely high correlation.
It is possible for the within-patient correlation between end points to be high, but the within-trial concordance to be low.30 For this reason, individual patient correlation is generally not accepted as sufficient evidence of surrogacy. Regarding trial level surrogacy, we feel the most meaningful metric to examine is the concordance of hazard ratios (comparing treatment with control within each study). This summary takes into account the entire pattern of events for each treatment arm, as opposed to single time-point summaries. For this measure, the hazard ratio for DFS, with 3 years of median follow-up, and OS, with 5 years of median follow-up, showed tight agreement. In addition, the formal measures examined also showed strong evidence for surrogacy, and predictions from the model for the association between the two end points showed excellent concordance with the actual data.
The concordance of results measure examining whether the trial gives the same results based on the presumed surrogate end point as based on the true end point has intuitive appeal. In this pooled analysis, the conclusions of the trial were concordant for the two end points in 23 of 25 within-trial comparisons. In the case of one discrepancy, the P value for DFS was .038 (for one of the three comparisons in Intergroup 008924). In the other discrepancy, the P value for DFS was .005, which would standardly be considered highly significant. The P values for OS for these two comparisons were .12 and .26, respectively. The trial responsible for the major discrepancy, North Central Cancer Treatment Group Trial 78-48-52,15 was the second smallest trial included in this pooled analysis, with a sample size of only 247 patients. Modern trials generally include sample sizes of an order of magnitude greater than this, which by the results of this analysis, suggest that such a discrepancy in a present-day trial is unlikely.
The concordance of outcomes measure has been criticized because the power for the true outcome may be less than that for the surrogate, thus P values may not be significant for the true outcome because of inadequate power as opposed to a true lack of effect. Many but not all of the trials included in this analysis were well powered (at least 80% power) to detect a difference in OS, thus overcoming that criticism. In addition, in these trials, the event rates for DFS at 3 years and OS at 5 years are statistically indistinguishable (intercept and slope from the weighted linear regression not significantly different from 0 and 1, respectively). This implies that for trials included in this analysis, and for future trials, power considerations and sample size requirements should not differ substantially for the two end points.
The data from these trials suggest a slight attenuation of the hazard ratio comparing the experimental with the control regimen between the two end points (Table 3). This could be explained by several reasons. Because 80% of deaths in this population are preceded by recurrence, the DFS end point is dominated by recurrence. Deaths from competing causes, which should be balanced between the two arms, play a larger role in the OS comparison, thus affecting the difference in outcomes between the arms. A second possible explanation for the attenuation is a form of cross-over: patients on the control arm who experience recurrence may receive the experimental therapy at the time of recurrence, whereas for those on the experimental arm, re-treatment may not be as effective. However, nine of the 18 trials included had chemotherapy in all arms, and in the majority of the time periods spanned by these trials, the only therapy available for recurrence in patients who had experienced recurrence was FU-based, with its known limited impact as sole treatment in recurrent disease. In either case, for these trials, the attenuation is modest. However, it does suggest caution when interpreting marginally significant DFS results.
In conducting this analysis, we attempted to emulate the conduct of an actual trial as closely as possible, by inducing a censoring pattern that would occur for an analysis at a particular time point (the time at which the median follow-up for patients still alive was 3 years). Of course perfect emulation is not possible, because in any given trial, follow-up will not be complete and up-to-date on all patients at this exact time point (because of data delinquency, schedules of clinic visits, or other reasons).
In the 8 years of follow-up considered in this analysis, 80% of the recurrences occurred within the first 3 years. Although this may imply that 3 years is sufficient to dictate the results of a clinical trial, this may not translate into a practice guideline for any particular patient, which is an area of active clinical investigation.31,32
The trials included in this analysis spanned the accrual period of 1977 through 1999. In the intervening years, advances have been made in areas of treatment, monitoring for recurrence, and surgical resection for patients with limited recurrence that may impact on the applicability of these results to modern trials. In the area of diagnosis, the ability of imaging to detect recurrent disease earlier may introduce a lead-time bias, appearing to increase the duration from recurrence to death. Unless this earlier detection allows for more effective, potentially curable treatment of small metastases, the impact of this bias on these findings may impact the time points chosen (3 years and 5 years), but not the fundamental relationship.
In the treatment of advanced disease, new agents, such as irinotecan, oxaliplatin, and bevacizumab33-35 have extended the median survival with advanced disease to approximately 20 months.36 However, the long-term survival prospects with recurrent disease, even with these new therapies, remain limited. Assuming the majority of recurrences continue to occur within the first 3 years, even with a 20-month median survival, most patients with recurrent disease will have died before 5 years. As median survival after recurrence continues to improve to a hopeful 24 months or greater, 5 years may no longer be an appropriate time at which to judge long-term cure. However, until regimens are developed that cure more than a small proportion of patients with recurrent disease, these newer treatments for recurrent disease may again only impact the time points chosen for analysis, but not the fundamental relationship between DFS and OS.
Recently, two combination regimens of FU/leucovorin and oxaliplatin have shown promise in the setting of adjuvant disease in stage II and III patients, based on the end point of DFS assessed after a median 3 years of follow-up.37,38 If oxaliplatin or other future agents that reduce the risk of recurrence truly prevent recurrence, then the findings based on the FU-based regimens included in this analysis should translate to these newer regimens. If, however, these new regimens simply delay, as opposed to prevent, recurrence, the time points for analysis may need to be reconsidered. If the adjuvant regimen impacts the nature of the recurrence, such as rendering the recurrent tumor more or less aggressive, then the fundamental nature of the relationship between the end points could be altered.
Recent discussions39 have expanded the scope of DFS as an end point in adjuvant colon cancer trials. These discussions have included the role of DFS as an end point of its own merit, regardless of its relationship with OS. The high subsequent costs, quality-of-life impact, and debilitating consequences of recurrent disease suggest that prevention of recurrence is of considerable importance, regardless of any eventual impact on OS. In colon cancer, FU-based therapy has been shown to improve both DFS and OS, thus the debate has been academic to this point. However, this issue will increase in importance as strategies for the treatment of recurrent disease improve.
In summary, in this pooled analysis of more than 20,800 patients and 18 trials of FU-based regimens in the setting of adjuvant colon cancer, we have demonstrated a consistent and strong association between the end points of DFS and OS. Highly significant observed differences in DFS, assessed after 3 years, were highly likely to be present in OS after 5 years of follow-up. Marginally significant DFS advantages may not translate into OS benefit. These findings suggest that DFS can be considered an appropriate primary end point in the setting of clinical trials in adjuvant colon cancer.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Supported in part by National Cancer Institute Grant No. CA25224.
Presented in part at the 40th Annual Meeting of the American Society of Clinical Oncology, June 5-8, 2004, New Orleans, LA.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
National Institutes of Health: Adjuvant therapy for patients with colon and rectum cancer: NIH Consensus Statement (vol 8). Bethesda, MD, National Institutes of Health, 1990, pp 1-25
Sargent DJ, Goldberg RM, Jacobson SD, et al: A pooled analysis of adjuvant chemotherapy for resected colon cancer in elderly patients. N Engl J Med 345:1091-1097, 2001
Benson AB 3rd, Schrag D, Somerfield MR, et al: American Society of Clinical Oncology recommendations on adjuvant chemotherapy for stage II colon cancer. J Clin Oncol 22:3408-3419, 2004
Gray RG, Barnwell J, Hills R, et al: QUASAR: A randomized study of adjuvant chemotherapy (CT) vs. observation including 3238 colorectal cancer patients. J Clin Oncol 23:245s, 2004 (abstr 3501)
Gill S, Loprinzi CL, Sargent DJ, et al: Pooled analysis of fluorouracil-based adjuvant therapy for stage II and III colon cancer: Who benefits and by how much? J Clin Oncol 22:1797-1806, 2004
Prentice RL: Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med 8:431-440, 1989
Fleming TR, DeMets DL: Surrogate end points in clinical trials: Are we being misled? Ann Intern Med 125:605-613, 1996
Buyse M, Molenberghs G, Burzykowski T, et al: The validation of surrogate endpoints in meta-analyses of randomized experiments. Biostatistics 1:49-67, 2000
Begg CB, Leung DHY: On the use of surrogate end points in randomized trials. J R Stat Soc (Ser A) 163:15-24, 2000
Cox DR: Regression models and life tables (with discussion). J R Stat Soc B 34:187-220, 1972
Freedman LS, Graubard BI, Schatzkin A: Statistical validation of intermediate endpoints for chronic diseases. Stat Med 11:167-178, 1992
Burzykowski T, Molenberghs G, Buyse M, et al: Validation of surrogate end points in multiple randomized clinical trials with failure time end points. Appl Stat 50:405-422, 2001
Efron B, Gong G: A leisurely look at the bootstrap, the jackknife, and cross-validation. Am Stat 37:36-48, 1983
Wolmark N, Fisher B, Rockette H: Postoperative adjuvant chemotherapy or BCG for colon cancer: Results from NSABP protocol C-01. J Natl Cancer Inst 80:30-36, 1988
Laurie JA, Moertel CG, Fleming TR, et al: Surgical adjuvant therapy of large-bowel carcinoma: An evaluation of levamisole and the combination of levamisole and fluorouracil. J Clin Oncol 7:1447-1456, 1989
International Multicentre Pooled Analysis of Colon Cancer Trials (IMPACT) investigators: Efficacy of adjuvant fluorouracil and folinic acid in colon cancer. Lancet 345:939-944,1995
Wolmark N, Rockette H, Wickerham DL, et al: Adjuvant therapy of Dukes' A, B, and C adenocarcinoma of the colon with portal-vein fluorouracil hepatic infusion: Preliminary results of National Surgical Adjuvant Breast and Bowel Project Protocol C-02. J Clin Oncol 8:1466-1475, 1990
Moertel CG, Fleming TR, Macdonald JS, et al: Levamisole and fluorouracil for adjuvant therapy of resected colon carcinoma. N Engl J Med 322:352-358, 1990
Francini G, Petrioli R, Lorenzini L, et al: Folinic acid and 5-fluorouracil as adjuvant chemotherapy in colon cancer. Gastroenterology 106:899-906, 1994
Wolmark N, Rockette H, Fisher B, et al: The benefit of leucovorin-modulated fluorouracil as postoperative adjuvant therapy for primary colon cancer: Results from National Surgical Adjuvant Breast and Bowel Project protocol C-03. J Clin Oncol 11:1879-1887, 1993
O'Connell MJ, Mailliard JA, Kahn MJ, et al: Controlled clinical trial of fluorouracil and low-dose leucovorin given for 6 months as postoperative adjuvant therapy for colon cancer. J Clin Oncol 15:246-250, 1997
O'Connell MJ, Laurie JA, Kahn M, et al: Prospectively randomized trial of postoperative adjuvant chemotherapy in patients with high-risk colon cancer. J Clin Oncol 16:295-300, 1998
Wolmark N, Rockette H, Mamounas E, et al: Clinical trial to assess the relative efficacy of fluorouracil and leucovorin, fluorouracil and levamisole, and fluorouracil, leucovorin, and levamisole in patients with Dukes' B and C carcinoma of the colon: Results from National Surgical Adjuvant Breast and Bowel Project C-04. J Clin Oncol 17:3553-3559, 1999
Haller DG, Catalano PJ, Macdonald JS, et al: Fluorouracil (FU), leucovorin (LV), and levamisole (LEV) adjuvant therapy for colon cancer: Five-year final report of INT-0089. Proc Am Soc Clin Oncol 17:256a, 1998 (abstr 992)
Wolmark N, Bryant J, Smith R, et al: Adjuvant 5-fluorouracil and leucovorin with or without interferon alfa-2a in colon carcinoma: National Surgical Adjuvant Breast and Bowel Project protocol C-05. J Natl Cancer Inst 90:1810-1816, 1998
O'Connell M, Skillings J, Windschitl H, et al: Phase III trial of high dose levamisole (LEV) + 5-fluorouracil (5FU) + leucovorin (CF) as adjuvant treatment (Adj) for high risk colon cancer (CC): An NCCTG/NCIC study. Proc Am Soc Clin Oncol 20:132a, 2001 (abstr 525)
Poplin E, Benedetti J, Estes N, et al: Phase III randomized trial of bolus 5-FU/leucovorin/levamisole versus 5-FU continuous infusion/levamisole as adjuvant therapy for high risk colon cancer (SWOG 8415/INT-0153). Proc Am Soc Clin Oncol 19:240a, 2000 (abstr)
Andre T, Colin P, Louvet C, et al: Semimonthly versus monthly regimen of fluorouracil and leucovorin administered for 24 or 36 weeks as adjuvant therapy in stage II and III colon cancer: Results of a randomized trial. J Clin Oncol 21:2896-2903, 2003
QUASAR Collaborative Group: Comparison of fluorouracil with additional levamisole, higher-dose folinic acid, or both, as adjuvant chemotherapy for colorectal cancer: A randomized trial. Lancet 355:1588-1596, 2000
Korn EL, Albert PS, McShane LM: Assessing surrogates as trial endpoints using mixed models. Stat Med 24:163-182, 2005
Benson III AB, Desch CE, Flynn PJ, et al: 2000 update of American Society of Clinical Oncology colorectal cancer surveillance guidelines. J Clin Oncol 18:3586-3588, 2000
Chau I, Allen MJ, Cunningham D, et al: The value of routine serum carcino-embryonic antigen measurement and computed tomography in surveillance of patients after adjuvant chemotherapy for colorectal cancer. J Clin Oncol 22:1420-1429, 2004
Saltz LB, Cox JV, Blanke C, et al: Irinotecan plus fluorouracil and leucovorin for metastatic colorectal cancer. N Engl J Med 343:905-914, 2000
Goldberg RM, Sargent DJ, Morton RF, et al: A randomized controlled trial of fluorouracil plus leucovorin, irinotecan, and oxaliplatin combinations in patients with previously untreated metastatic colorectal cancer. J Clin Oncol 22:23-30, 2004
Hurwitz H, Fehrenbacher L, Novotny W, et al: Bevacizumab plus irinotecan, fluorouracil, and leucovorin for metastatic colorectal cancer. N Engl J Med 350:2335-2342, 2004
Grothey A, Sargent D, Goldberg RM, et al: Survival of patients with advanced colorectal cancer improves with the availability of fluorouracil-leucovorin, irinotecan, and oxaliplatin in the course of treatment. J Clin Oncol 22:1209-1214, 2004
Andre T, Boni C, Mounedji-Boudiaf L, et al: Oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment for colon cancer. N Engl J Med 350:2343-2351, 2004
United States Food and Drug Administration: 3 year DFS vs. 5 year OS as an endpoint for adjuvant colon cancer studies: Data from randomized trials. http://www.fda.gov/cder/drug/cancer_endpoints/Sargent/index.htm(Daniel J. Sargent, Harry )
National Surgical Adjuvant Breast and Bowel Project Statistical Center and Operations Office, Pittsburgh
Abramson Cancer Center, University of Pennsylvania, Philadelphia, PA
QUASAR Collaborative Group, Birmingham and Oxford, United Kingdom
Southwest Oncology Group Statistical Center, Seattle, WA
International Drug Development Institute, Brussels, Belgium
Ospedali Riuniti, Bergamo
University of Siena, Siena, Italy
University of the Mediterranean, Marseilles
Groupe d'Etude et de Recherche Clinique en Oncologie et Radiothérapie, Paris
Hospital Saint Antoine, Paris, France
National Cancer Institute of Canada Clinical Trials Group, Queens University, Kingston, Ontario, Canada
Eastern Cooperative Oncology Group Statistical Center, Boston, MA
Oregon Health and Science University Cancer Institute, Portland, OR
University of North Carolina, Chapel Hill, NC
ABSTRACT
PURPOSE: A traditional end point for colon adjuvant clinical trials is overall survival (OS), with 5 years demonstrating adequate follow-up. A shorter-term end point providing convincing evidence to allow treatment comparisons could significantly speed the translation of advances into practice.
METHODS: Individual patient data were pooled from 18 randomized phase III colon cancer adjuvant clinical trials. Trials included 43 arms, with a pooled sample size of 20,898 patients. The primary hypothesis was that disease-free survival (DFS), with 3 years of follow-up, is an appropriate primary end point to replace OS with 5 years of follow-up.
RESULTS: The recurrence rates for years 1 through 5 were 12%, 14%, 8%, 5%, and 3%, respectively. Median time from recurrence to death was 12 months. Eighty percent of recurrences were in the first 3 years; 91% of patients with recurrence by 3 years died before 5 years. Correlation between 3-year DFS and 5-year OS was 0.89. Comparing control versus experimental arms within each trial, the correlation between hazard ratios for DFS and OS was 0.92. Within-trial log-rank testing using both DFS and OS provided the same conclusion in 23 (92%) of 25 cases. Formal measures of surrogacy were satisfied.
CONCLUSION: In patients treated on phase III adjuvant colon clinical trials, DFS and OS are highly correlated, both within patients and across trials. These results suggest that DFS after 3 years of median follow-up is an appropriate end point for adjuvant colon cancer clinical trials of fluorouracil-based regimens, although marginally significant DFS improvements may not translate into significant OS benefits.
INTRODUCTION
The benefit of adjuvant chemotherapy on overall survival (OS) for patients with node-positive, locally advanced colon cancer is well established.1 In a pooled-analysis, fluorouracil (FU)-based chemotherapy was shown to increase the 5-year OS from 64% to 71% in a mixed group of node-negative (American Joint Committee on Cancer stage II) and node-positive (American Joint Committee on Cancer stage III) patients.2 The benefit of treatment in patients with stage II disease has been less clear, and current American Society of Clinical Oncology guidelines do not recommend the routine use of adjuvant chemotherapy for all patients with stage II disease.3 However, two large studies (one randomized clinical trial4 and one pooled analysis of seven randomized clinical trials5) have both suggested a modest but consistent treatment benefit from FU-based chemotherapy in these patients.
Clearly, long-term cure is the most unambiguous, and ultimately important, outcome for colon cancer patients. The goal of improving long-term outcome has translated into the use of OS as the primary end point for most stage II and stage III colon cancer adjuvant therapy trials. Historically, the OS rate at 5 years has been the most typically quoted metric for judging the success of a particular treatment regimen. This end point has the advantage of being simple to measure, easy to interpret, clinically meaningful, and straightforward to explain. The disadvantage of this end point is that it requires extended follow-up. An end point that is reached more rapidly has obvious advantages, including but not limited to answering the question posed by the trial more rapidly and, potentially, quicker drug approval and availability. The goal of this study is to explore disease-free survival (DFS) as a possible alternative end point to OS, in clinical trials testing adjuvant treatment for patients with stages II and III colon cancer.
Considerable attention has been paid to the field of surrogate markers, with some conflict as to the most appropriate definition of an adequate surrogate.6-8 The most widely cited criteria are proposed by Prentice,6 who stated that for a surrogate marker to be valid, it must be correlated with the true end point, and it must fully capture the net effect of the treatment on the true end point. These criteria have been criticized for several reasons, one of which is that they are too stringent to be valid in practice. More recently, approaches have been proposed suggesting that for a surrogate to be appropriate, the effect of treatment on the surrogate end point must predict the effect of treatment on the true end point or that a test of a statistical hypothesis based on the surrogate end point should come to the same conclusion as a statistical hypothesis test based on the primary end point.8,9 This article will not attempt to resolve any definitional issues regarding surrogate markers, rather it will present data designed to evaluate the appropriateness of alternative end points for future adjuvant clinical trials in colon cancer through a data-based approach.
METHODS
Through contact with principal investigators, we combined individual patient data from large randomized phase III trials in adjuvant colon cancer. This is not a formal meta-analysis in that an exhaustive attempt to include every completed colon adjuvant trial was not undertaken. Trials were selected based on relevance, maturity, and data availability. For each trial, following the original protocol, one arm was defined as the control arm, even if chemotherapy was used in all patients. Two trials included a small number of patients with high rectal tumors; these patients are included in this analysis. OS was defined as the time from randomization to death from any cause. DFS was defined as the time from randomization to the first event of either recurrent disease or death. Second primary tumors were not counted as events in the DFS analysis. All randomly assigned patients were included in the analyses according to the intention to treat principle. This retrospective pooled analysis was institutional review board–approved; each trial included was originally approved through appropriate local mechanisms.
The long-term follow-up practices of the various studies differed. To reduce the possibility of bias owing to this differential practice, the follow-up data for all patients were censored at 8 years from randomization. For all analyses, efforts were made to replicate the actual conduct of a clinical trial, where patients enter over a period of several years, and thus at any given calendar date have differing durations of follow-up. For each analyses conducted at a specific time point (eg, 3 years), the outcome data for each patient were censored at the point in time at which the median follow-up in the trial was 3 years. For example, if a trial accrued evenly over a period of 3 years, then the median follow-up would be 3 years after 1.5 years had elapsed from the close of randomization. At this point, the first patient enrolled would have 4.5 years of follow-up, whereas the last patient enrolled would have 1.5 years of follow-up. Analyses were repeated after applying a censoring pattern 3 years after the date of the last patient randomization to that trial. These two analyses gave similar conclusions; we therefore present the results with 3 years of median follow-up.
Multiple analysis techniques were used. For several end points, simple percentage agreement was calculated based on the individual patient data. Hazard ratios comparing each experimental arm with the control arm in that trial were calculated using the Cox proportional hazards regression model10 separately for each individual trial. Statistical significance of the comparison between the experimental and control regimen of each trial for each end point was evaluated using the log-rank test, with a two-sided P < .05 used to denote statistical significance. The relationship between various end points was estimated using weighted linear regression, with weights equal to the sample size of the trial from which the data were derived.
Because of the evolving literature on surrogate end points, formal testing for surrogacy was also explored with multiple criteria. The first approach followed Freedman et al,11 fitting two Cox proportional hazards models and calculating the proportion of net treatment effect explained by the surrogate end point. The second approach followed Burzykowski et al,12 where both the trial level and individual level association between the proposed surrogate and the true end point is estimated using a bivariate survival model. Both of these measures are based on a 0 to 1 scale, with values close to 1 indicating effective surrogacy. In addition, we explored the proportion of agreement of trial results, as suggested by Begg and Leung,9 where the proportion of times the trials reached the same conclusion based on statistical significance testing for the two end points (DFS and OS) was calculated. The use of this variety of methods was designed to make our findings robust to the particulars of any specific approach, which each have advantages and drawbacks.
To test the validity of the model associating within-trial hazard ratios for DFS and OS, leave-one-out cross-validation was performed.13 In short, the regression model was repeatedly fit using data from all trials except the one left out, then used to predict the results for the trial for which data were not included in the model fitting. Ninety-five percent prediction intervals were computed for each excluded trial, and the actual data were compared with the prediction to explore the accuracy of the prediction.
RESULTS
Patients and Trials
A total of 18 randomized phase III clinical trials including 43 different treatment arms were included in the analysis (Table 1). Two trials included a levamisole-alone control arm15,18; these two arms were excluded from the analysis. The total sample size is 20,898 patients. The median follow-up on living patients was 8 years; 5-year data were available for 89% of patients. The period of trial enrollment spanned from 1977 to 1999. All trials included at least one arm with an FU-containing regimen. Nine trials included a no-treatment control arm.
Patient characteristics are outlined in Table 2. Sixty-six percent of patients were stage III, 55% were male, and 19% were older than 70 years of age. Eighty-eight percent of patients were randomly assigned to a chemotherapy-containing treatment arm.
Per-Patient Measures
The recurrence rates for years 1 through 5 were 12%, 14%, 8%, 5%, and 3% respectively. Eighty percent of patients who experienced disease recurrence during the 8-year follow-up period experienced recurrence within the first 3 years after study enrollment. The median time from the diagnosis of recurrent disease to death was 12 months. Ninety-one percent of patients who experienced a recurrence within 3 years of randomization died before 5 years from randomization.
Per–Study Arm Measures
The relationship between the DFS rate after 3 years of median follow-up and the OS rate after 5 years of median follow-up by study arm is shown in Figure 1. The R2 value from a weighted linear regression is 0.85; the Spearman rank correlation coefficient between the two end points is 0.88. For 33 of the 43 study arms, the difference between the 3-year DFS and the 5-year OS was 3%. The largest single difference in the two measures for any single study arm was 8%. The weighted linear regression for the two end points is shown in Figure 1; the SE for the intercept is 0.047 and for the slope is 0.068. These two coefficients do not differ significantly from 0 and 1 respectively (P > .20).
Per-Study Measures
The relationship between the within-study hazard ratio comparing the experimental and the control arms for DFS and OS is shown in Figure 2. The R2 value from a weighted linear regression is 0.90; the Spearman rank correlation coefficient between the 2 end points is 0.94. The weighted linear regression for the two end points is as follows: the hazard ratio for OS = 0.12 + 0.89 x (hazard ratio for DFS), the SE for the intercept is 0.056, and the SE for the slope is 0.061. These two coefficients do differ significantly from 0 and 1 respectively (P = .03 and < .0001, respectively) and indicate a slight attenuation of effect on the hazard ratio between the two end points (Table 3).
Surrogate End Point Measures
The proportion of treatment effect of chemotherapy on the end point of OS explained by DFS was greater than 100%. This curious result is due to the role of recurrence as such a dominant predictor of death that the coefficient for treatment in the Cox model changed from negative to positive when recurrence was included in the model. The individual level correlation between the two end points, estimated using a bivariate survival model, is 0.873 (95% CI, 0.869 to 0.877), and the trial level correlation, estimated from the same model, is 0.78 (95% CI, 0.60 to 0.96). Of 25 within-trial comparisons (control v treatment within a trial), using P < .05 as a measure of concordance, 23 of 25 trials obtained the same conclusion regardless of end point (no difference between arms for 18 within-trial comparisons for either end point; five within-trial comparisons were significant for both end points). In the two trials with discordance, the P values for DFS were .005 and .038 and for OS were .26 and .12, respectively.
Model Validation
The results of the leave-one-out cross-validation for the fitted weighted regression model for within-trial hazard ratios are shown in Figure 3. Twenty-four of the 25 observed OS hazard ratios fell within the 95% prediction intervals, with approximately an equal number of overprediction and underpredictions, indicating excellent model calibration.
DISCUSSION
This analysis was initiated to address a simple question: Can we make a decision regarding the efficacy of a new adjuvant colon cancer treatment regimen more quickly? The implications of such a finding, if confirmed, would be several. Clinical trials could be completed more quickly with reduced cost, drug development time could be shortened, and, most importantly, promising new agents or combinations of agents could be made available to patients more rapidly. In the setting of FU-based treatments for adjuvant colon cancer, the answer to the question seems to be yes.
During the first 8 years after surgical resection, 80% of patient deaths are preceded by a tumor recurrence.2 On an individual patient basis, in this analysis, 80% of patients who experienced disease recurrence experienced their recurrence within 3 years of study entry. During the time in which these trials were conducted (1977 through 1999), the efficacy of postrecurrence chemotherapy was limited, with a median time from recurrence to death of 12 months, which is consistent with that reported in most phase III advanced colon cancer trials conducted before the late 1990s. Thus by 3 years from study entry, most of the recurrences that were to occur had occurred, and by 5 years, the vast majority of these patients with recurrent disease had died. In addition, based on a bivariate survival model,12 the patient level correlation between the two end points of DFS and OS is 0.87, which is an extremely high correlation.
It is possible for the within-patient correlation between end points to be high, but the within-trial concordance to be low.30 For this reason, individual patient correlation is generally not accepted as sufficient evidence of surrogacy. Regarding trial level surrogacy, we feel the most meaningful metric to examine is the concordance of hazard ratios (comparing treatment with control within each study). This summary takes into account the entire pattern of events for each treatment arm, as opposed to single time-point summaries. For this measure, the hazard ratio for DFS, with 3 years of median follow-up, and OS, with 5 years of median follow-up, showed tight agreement. In addition, the formal measures examined also showed strong evidence for surrogacy, and predictions from the model for the association between the two end points showed excellent concordance with the actual data.
The concordance of results measure examining whether the trial gives the same results based on the presumed surrogate end point as based on the true end point has intuitive appeal. In this pooled analysis, the conclusions of the trial were concordant for the two end points in 23 of 25 within-trial comparisons. In the case of one discrepancy, the P value for DFS was .038 (for one of the three comparisons in Intergroup 008924). In the other discrepancy, the P value for DFS was .005, which would standardly be considered highly significant. The P values for OS for these two comparisons were .12 and .26, respectively. The trial responsible for the major discrepancy, North Central Cancer Treatment Group Trial 78-48-52,15 was the second smallest trial included in this pooled analysis, with a sample size of only 247 patients. Modern trials generally include sample sizes of an order of magnitude greater than this, which by the results of this analysis, suggest that such a discrepancy in a present-day trial is unlikely.
The concordance of outcomes measure has been criticized because the power for the true outcome may be less than that for the surrogate, thus P values may not be significant for the true outcome because of inadequate power as opposed to a true lack of effect. Many but not all of the trials included in this analysis were well powered (at least 80% power) to detect a difference in OS, thus overcoming that criticism. In addition, in these trials, the event rates for DFS at 3 years and OS at 5 years are statistically indistinguishable (intercept and slope from the weighted linear regression not significantly different from 0 and 1, respectively). This implies that for trials included in this analysis, and for future trials, power considerations and sample size requirements should not differ substantially for the two end points.
The data from these trials suggest a slight attenuation of the hazard ratio comparing the experimental with the control regimen between the two end points (Table 3). This could be explained by several reasons. Because 80% of deaths in this population are preceded by recurrence, the DFS end point is dominated by recurrence. Deaths from competing causes, which should be balanced between the two arms, play a larger role in the OS comparison, thus affecting the difference in outcomes between the arms. A second possible explanation for the attenuation is a form of cross-over: patients on the control arm who experience recurrence may receive the experimental therapy at the time of recurrence, whereas for those on the experimental arm, re-treatment may not be as effective. However, nine of the 18 trials included had chemotherapy in all arms, and in the majority of the time periods spanned by these trials, the only therapy available for recurrence in patients who had experienced recurrence was FU-based, with its known limited impact as sole treatment in recurrent disease. In either case, for these trials, the attenuation is modest. However, it does suggest caution when interpreting marginally significant DFS results.
In conducting this analysis, we attempted to emulate the conduct of an actual trial as closely as possible, by inducing a censoring pattern that would occur for an analysis at a particular time point (the time at which the median follow-up for patients still alive was 3 years). Of course perfect emulation is not possible, because in any given trial, follow-up will not be complete and up-to-date on all patients at this exact time point (because of data delinquency, schedules of clinic visits, or other reasons).
In the 8 years of follow-up considered in this analysis, 80% of the recurrences occurred within the first 3 years. Although this may imply that 3 years is sufficient to dictate the results of a clinical trial, this may not translate into a practice guideline for any particular patient, which is an area of active clinical investigation.31,32
The trials included in this analysis spanned the accrual period of 1977 through 1999. In the intervening years, advances have been made in areas of treatment, monitoring for recurrence, and surgical resection for patients with limited recurrence that may impact on the applicability of these results to modern trials. In the area of diagnosis, the ability of imaging to detect recurrent disease earlier may introduce a lead-time bias, appearing to increase the duration from recurrence to death. Unless this earlier detection allows for more effective, potentially curable treatment of small metastases, the impact of this bias on these findings may impact the time points chosen (3 years and 5 years), but not the fundamental relationship.
In the treatment of advanced disease, new agents, such as irinotecan, oxaliplatin, and bevacizumab33-35 have extended the median survival with advanced disease to approximately 20 months.36 However, the long-term survival prospects with recurrent disease, even with these new therapies, remain limited. Assuming the majority of recurrences continue to occur within the first 3 years, even with a 20-month median survival, most patients with recurrent disease will have died before 5 years. As median survival after recurrence continues to improve to a hopeful 24 months or greater, 5 years may no longer be an appropriate time at which to judge long-term cure. However, until regimens are developed that cure more than a small proportion of patients with recurrent disease, these newer treatments for recurrent disease may again only impact the time points chosen for analysis, but not the fundamental relationship between DFS and OS.
Recently, two combination regimens of FU/leucovorin and oxaliplatin have shown promise in the setting of adjuvant disease in stage II and III patients, based on the end point of DFS assessed after a median 3 years of follow-up.37,38 If oxaliplatin or other future agents that reduce the risk of recurrence truly prevent recurrence, then the findings based on the FU-based regimens included in this analysis should translate to these newer regimens. If, however, these new regimens simply delay, as opposed to prevent, recurrence, the time points for analysis may need to be reconsidered. If the adjuvant regimen impacts the nature of the recurrence, such as rendering the recurrent tumor more or less aggressive, then the fundamental nature of the relationship between the end points could be altered.
Recent discussions39 have expanded the scope of DFS as an end point in adjuvant colon cancer trials. These discussions have included the role of DFS as an end point of its own merit, regardless of its relationship with OS. The high subsequent costs, quality-of-life impact, and debilitating consequences of recurrent disease suggest that prevention of recurrence is of considerable importance, regardless of any eventual impact on OS. In colon cancer, FU-based therapy has been shown to improve both DFS and OS, thus the debate has been academic to this point. However, this issue will increase in importance as strategies for the treatment of recurrent disease improve.
In summary, in this pooled analysis of more than 20,800 patients and 18 trials of FU-based regimens in the setting of adjuvant colon cancer, we have demonstrated a consistent and strong association between the end points of DFS and OS. Highly significant observed differences in DFS, assessed after 3 years, were highly likely to be present in OS after 5 years of follow-up. Marginally significant DFS advantages may not translate into OS benefit. These findings suggest that DFS can be considered an appropriate primary end point in the setting of clinical trials in adjuvant colon cancer.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Supported in part by National Cancer Institute Grant No. CA25224.
Presented in part at the 40th Annual Meeting of the American Society of Clinical Oncology, June 5-8, 2004, New Orleans, LA.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
National Institutes of Health: Adjuvant therapy for patients with colon and rectum cancer: NIH Consensus Statement (vol 8). Bethesda, MD, National Institutes of Health, 1990, pp 1-25
Sargent DJ, Goldberg RM, Jacobson SD, et al: A pooled analysis of adjuvant chemotherapy for resected colon cancer in elderly patients. N Engl J Med 345:1091-1097, 2001
Benson AB 3rd, Schrag D, Somerfield MR, et al: American Society of Clinical Oncology recommendations on adjuvant chemotherapy for stage II colon cancer. J Clin Oncol 22:3408-3419, 2004
Gray RG, Barnwell J, Hills R, et al: QUASAR: A randomized study of adjuvant chemotherapy (CT) vs. observation including 3238 colorectal cancer patients. J Clin Oncol 23:245s, 2004 (abstr 3501)
Gill S, Loprinzi CL, Sargent DJ, et al: Pooled analysis of fluorouracil-based adjuvant therapy for stage II and III colon cancer: Who benefits and by how much? J Clin Oncol 22:1797-1806, 2004
Prentice RL: Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med 8:431-440, 1989
Fleming TR, DeMets DL: Surrogate end points in clinical trials: Are we being misled? Ann Intern Med 125:605-613, 1996
Buyse M, Molenberghs G, Burzykowski T, et al: The validation of surrogate endpoints in meta-analyses of randomized experiments. Biostatistics 1:49-67, 2000
Begg CB, Leung DHY: On the use of surrogate end points in randomized trials. J R Stat Soc (Ser A) 163:15-24, 2000
Cox DR: Regression models and life tables (with discussion). J R Stat Soc B 34:187-220, 1972
Freedman LS, Graubard BI, Schatzkin A: Statistical validation of intermediate endpoints for chronic diseases. Stat Med 11:167-178, 1992
Burzykowski T, Molenberghs G, Buyse M, et al: Validation of surrogate end points in multiple randomized clinical trials with failure time end points. Appl Stat 50:405-422, 2001
Efron B, Gong G: A leisurely look at the bootstrap, the jackknife, and cross-validation. Am Stat 37:36-48, 1983
Wolmark N, Fisher B, Rockette H: Postoperative adjuvant chemotherapy or BCG for colon cancer: Results from NSABP protocol C-01. J Natl Cancer Inst 80:30-36, 1988
Laurie JA, Moertel CG, Fleming TR, et al: Surgical adjuvant therapy of large-bowel carcinoma: An evaluation of levamisole and the combination of levamisole and fluorouracil. J Clin Oncol 7:1447-1456, 1989
International Multicentre Pooled Analysis of Colon Cancer Trials (IMPACT) investigators: Efficacy of adjuvant fluorouracil and folinic acid in colon cancer. Lancet 345:939-944,1995
Wolmark N, Rockette H, Wickerham DL, et al: Adjuvant therapy of Dukes' A, B, and C adenocarcinoma of the colon with portal-vein fluorouracil hepatic infusion: Preliminary results of National Surgical Adjuvant Breast and Bowel Project Protocol C-02. J Clin Oncol 8:1466-1475, 1990
Moertel CG, Fleming TR, Macdonald JS, et al: Levamisole and fluorouracil for adjuvant therapy of resected colon carcinoma. N Engl J Med 322:352-358, 1990
Francini G, Petrioli R, Lorenzini L, et al: Folinic acid and 5-fluorouracil as adjuvant chemotherapy in colon cancer. Gastroenterology 106:899-906, 1994
Wolmark N, Rockette H, Fisher B, et al: The benefit of leucovorin-modulated fluorouracil as postoperative adjuvant therapy for primary colon cancer: Results from National Surgical Adjuvant Breast and Bowel Project protocol C-03. J Clin Oncol 11:1879-1887, 1993
O'Connell MJ, Mailliard JA, Kahn MJ, et al: Controlled clinical trial of fluorouracil and low-dose leucovorin given for 6 months as postoperative adjuvant therapy for colon cancer. J Clin Oncol 15:246-250, 1997
O'Connell MJ, Laurie JA, Kahn M, et al: Prospectively randomized trial of postoperative adjuvant chemotherapy in patients with high-risk colon cancer. J Clin Oncol 16:295-300, 1998
Wolmark N, Rockette H, Mamounas E, et al: Clinical trial to assess the relative efficacy of fluorouracil and leucovorin, fluorouracil and levamisole, and fluorouracil, leucovorin, and levamisole in patients with Dukes' B and C carcinoma of the colon: Results from National Surgical Adjuvant Breast and Bowel Project C-04. J Clin Oncol 17:3553-3559, 1999
Haller DG, Catalano PJ, Macdonald JS, et al: Fluorouracil (FU), leucovorin (LV), and levamisole (LEV) adjuvant therapy for colon cancer: Five-year final report of INT-0089. Proc Am Soc Clin Oncol 17:256a, 1998 (abstr 992)
Wolmark N, Bryant J, Smith R, et al: Adjuvant 5-fluorouracil and leucovorin with or without interferon alfa-2a in colon carcinoma: National Surgical Adjuvant Breast and Bowel Project protocol C-05. J Natl Cancer Inst 90:1810-1816, 1998
O'Connell M, Skillings J, Windschitl H, et al: Phase III trial of high dose levamisole (LEV) + 5-fluorouracil (5FU) + leucovorin (CF) as adjuvant treatment (Adj) for high risk colon cancer (CC): An NCCTG/NCIC study. Proc Am Soc Clin Oncol 20:132a, 2001 (abstr 525)
Poplin E, Benedetti J, Estes N, et al: Phase III randomized trial of bolus 5-FU/leucovorin/levamisole versus 5-FU continuous infusion/levamisole as adjuvant therapy for high risk colon cancer (SWOG 8415/INT-0153). Proc Am Soc Clin Oncol 19:240a, 2000 (abstr)
Andre T, Colin P, Louvet C, et al: Semimonthly versus monthly regimen of fluorouracil and leucovorin administered for 24 or 36 weeks as adjuvant therapy in stage II and III colon cancer: Results of a randomized trial. J Clin Oncol 21:2896-2903, 2003
QUASAR Collaborative Group: Comparison of fluorouracil with additional levamisole, higher-dose folinic acid, or both, as adjuvant chemotherapy for colorectal cancer: A randomized trial. Lancet 355:1588-1596, 2000
Korn EL, Albert PS, McShane LM: Assessing surrogates as trial endpoints using mixed models. Stat Med 24:163-182, 2005
Benson III AB, Desch CE, Flynn PJ, et al: 2000 update of American Society of Clinical Oncology colorectal cancer surveillance guidelines. J Clin Oncol 18:3586-3588, 2000
Chau I, Allen MJ, Cunningham D, et al: The value of routine serum carcino-embryonic antigen measurement and computed tomography in surveillance of patients after adjuvant chemotherapy for colorectal cancer. J Clin Oncol 22:1420-1429, 2004
Saltz LB, Cox JV, Blanke C, et al: Irinotecan plus fluorouracil and leucovorin for metastatic colorectal cancer. N Engl J Med 343:905-914, 2000
Goldberg RM, Sargent DJ, Morton RF, et al: A randomized controlled trial of fluorouracil plus leucovorin, irinotecan, and oxaliplatin combinations in patients with previously untreated metastatic colorectal cancer. J Clin Oncol 22:23-30, 2004
Hurwitz H, Fehrenbacher L, Novotny W, et al: Bevacizumab plus irinotecan, fluorouracil, and leucovorin for metastatic colorectal cancer. N Engl J Med 350:2335-2342, 2004
Grothey A, Sargent D, Goldberg RM, et al: Survival of patients with advanced colorectal cancer improves with the availability of fluorouracil-leucovorin, irinotecan, and oxaliplatin in the course of treatment. J Clin Oncol 22:1209-1214, 2004
Andre T, Boni C, Mounedji-Boudiaf L, et al: Oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment for colon cancer. N Engl J Med 350:2343-2351, 2004
United States Food and Drug Administration: 3 year DFS vs. 5 year OS as an endpoint for adjuvant colon cancer studies: Data from randomized trials. http://www.fda.gov/cder/drug/cancer_endpoints/Sargent/index.htm(Daniel J. Sargent, Harry )