- © 2005 by American Society of Clinical Oncology
Objective Response to Chemotherapy As a Potential Surrogate End Point of Survival in Metastatic Breast Cancer Patients
- Paolo Bruzzi,
- Lucia Del Mastro,
- Maria P. Sormani,
- Lars Bastholt,
- Marco Danova,
- Christian Focan,
- George Fountzilas,
- James Paul,
- Riccardo Rosso and
- Marco Venturini
- From the Units of Clinical Epidemiology and Medical Oncology, National Institute for Cancer Research, Genoa; Department of Medical Oncology, University and Istituto di Ricovero e Cura a Carattere Scientifico San Matteo, Pavia, Italy; Department of Oncology, Odense University Hospital, Odense, Denmark; Department of Oncology, CHC Clinique Saint Joseph, Liege, Belgium; Department of Medical Oncology, Papageorgiou Hospital, Thessaloniki, Greece; and CRUK Trials Unit, Beatson Oncology Centre, Glasgow, Scotland, United Kingdom
- Address reprint requests to Paolo Bruzzi MD, MPH, PhD, Unit of Clinical Epidemiology, National Institute for Cancer Research, Largo Rosanna Benzi 10, 16132 Genoa, Italy; e-mail: paolo.bruzzi{at}istge.it
Abstract
Purpose To assess the validity of objective response to chemotherapy as a surrogate end point for survival in metastatic breast cancer.
Patients and Methods We carried out a meta-analysis on individual data from 2,126 metastatic breast cancer patients who were enrolled onto 10 randomized trials comparing standard versus intensified epirubicin-containing chemotherapy.
Results The intensified chemotherapy was associated with a significantly higher tumor response rate compared with standard chemotherapy (pooled odds ratio for nonresponse, 0.60; 95% CI, 0.51 to 0.72). The intensified regimens also led to better (although not significant) survival (pooled odds ratio, 0.94; 95% CI, 0.86 to 1.04; P = .22). Tumor response was a highly significant predictor of survival (P < .0001). When tumor response was introduced in the Cox model, the hazard ratio in favor of experimental treatment changed from 0.94 to 1.005 (95% CI, 0.91 to 1.11; P = .92), indicating that no residual effect of the experimental treatment on survival was present once tumor response was adjusted for. This suggests that the overall survival benefit of intensified epirubicin was a result of the increase in response rate. The median survival time of patients with complete response and partial response was 28.8 months (95% CI, 25.4 to 45.3 months) and 21.3 months (95% CI, 19.2 to 22.4 months), respectively; whereas, the median survival time of patients with no response was 14.6 months (95% CI, 13.9 to 15.4 months).
Conclusion These results support the hypothesis that the achievement of an objective response to chemotherapy in metastatic breast cancer is associated with a true survival benefit. The potential role of objective response as a surrogate end point for survival in chemotherapy trials of metastatic breast cancer warrants further investigation.
INTRODUCTION
In advanced solid tumors, objective response to treatment is the standard end point of trials aimed at assessing the anticancer activity of new drugs and regimens. However, its role as a marker of a beneficial effect of treatment has been repeatedly questioned, both in clinical trials and in the individual patient.1,2 More specifically, it was shown that the longer survival of responders compared with nonresponders is not sufficient to demonstrate that treatments able to increase response rates are associated with survival benefits.3
The theory of surrogate end points4-6 provided the conceptual framework for assessing the presence of this potential benefit; a specific, intermediate end point or marker (in this case, objective response) can be considered a true surrogate indicator of another outcome of clinical interest (such as survival) when the effect of a treatment on the surrogate can reliably predict the effect of the treatment on the final clinical outcome. The demonstration of this property of a surrogate end point is defined as its validation, and it is specific to well-defined disease, clinical outcome, and treatment, in that extrapolation to different diseases, clinical outcomes, and treatments is not warranted, although in some instances, it may be plausible.
The validation of a surrogate end point implies the demonstration that the effect of a treatment on the clinical outcome of interest is mediated through and absorbed by the effect seen on the surrogate end point. This process requires4 the demonstration that an effect of treatment on the clinical outcome is present and that this effect disappears when the candidate surrogate is adjusted for. As a consequence, validation is best carried out using data from randomized trials. The necessary sample size, however, is larger than that required for assessing the efficacy of a treatment.
Therefore, validation of response to a given treatment as a surrogate of survival in a specific cancer condition requires the availability of large data sets of patients randomly assigned to the treatment of interest or to the control treatment, in whom both the surrogate and the clinical end point of interest have been measured. Because the necessary number of patients exceeds the size of most trials in metastatic patients, a meta-analysis based on individual patient data from all pertinent trials will be necessary in most instances.
Thus far, in solid tumors, investigations to validate objective response as a surrogate end point for survival were only carried out in advanced colorectal cancer for fluorouracil (FU) -based, experimental chemotherapy regimens, using a meta-analytic approach.7 The indirect evidence supporting a similar role for objective response to chemotherapy in metastatic breast cancer is compelling,8-12 but no analyses directly supporting this role have ever been published.
In the present study, we attempted to start a formal validation process of objective response as a surrogate of survival in metastatic breast cancer by taking advantage of the availability of individual patient data from trials assessing the role of epirubicin intensification in this disease. The issue of dose-intensified epirubicin-containing regimens in breast cancer patients was extensively evaluated at the National Cancer Research Institute of Genoa, Italy.13-17 In particular, two randomized studies evaluated the role of dose-intensified cyclophosphamide, epirubicin, and FU as first-line chemotherapy in metastatic breast cancer patients.16,17 We previously carried out a meta-analysis of published data, which included our two studies and six other trials comparing standard versus intensified epirubicin-containing regimens,16-23 focused on the increase in response rate associated with dose intensification. The pooled odds ratio of response (high dose-intensity arm v standard dose-intensity) was 1.58 (95% CI, 1.30 to 1.92). On the basis of these results, we decided to carry out a formal meta-analysis on individual data from the patients enrolled onto trials comparing standard versus intensified epirubicin-containing chemotherapy to validate objective response as a surrogate end point for survival.
PATIENTS AND METHODS
Data Source
For the purpose of the present meta-analysis, we performed a MEDLINE search as follows: epirubicin OR epidoxorubicin (Field: Title/Abstract) AND breast cancer (Field: Title/Abstract; Limits: Clinical trials). Trials were selected according to the following criteria: randomized clinical trial comparing standard versus intensified epirubicin-containing regimens. Moreover, references from retrieved original and review articles were scanned. Ten clinical trials were identified. The number of enrolled patients and regimens tested in these studies are listed in Table 1. The increase in the epirubicin dose-intensity was obtained by increasing the total dose per cycle18-20,22-25 by reducing the interval between cycles17,21 or by both increasing the dose and reducing the interval.16
Individual data from all enrolled patients were requested to principal investigators of the studies18,21-25 and to the company (Pharmacia, Milan, Italy) who had sponsored two of these studies19,20 and detained the databases. Items requested by the principal investigators included the study protocol and the computer files containing the individual patient records. Data of two studies16,17 were already on our database. For all of the remaining trials, data for all randomly assigned patients were provided by the principal investigator or the sponsor. The necessary information for each patient included study identification, treatment arm, date of randomization/registration, type of response, date of response (if responder), date of progression, date of last observation, and status of last observation (alive or dead).
Individual data from 2,259 metastatic breast cancer patients were collected and checked at the trial center of the National Cancer Institute of Genoa (Table 1). We excluded from the analysis the third arm of the French study (C500E100F500 for four cycles every 3 weeks: cyclophosphamide 500 mg/m2 day 1, epirubicin 100 mg/m2 day 1, and FU 500 mg/m2 day 1) because only a few cycles of chemotherapy were administered and the treatment was considered suboptimal for metastatic breast cancer patients. Notable imbalances in the number of randomly assigned patients are present in the trials of Brufman et al20 and Marschner et al.22 Both trials used a stratified randomization scheme where the stratification factors were center plus other factors, thus causing a large number of randomization lists, and it is well known that stratified randomization can result in imbalances in the number of patients assigned to each stratum, especially when the average number of patients per stratum is small. In the study by Marschner et al,22 the original article was based on 197 patients, whereas our analysis of this trial is based on the entire updated database of 269 patients with no exclusions. The final database used in the meta-analysis included 2,126 patients.
Validation of Tumor Response As a Surrogate of Survival
To validate objective response as a surrogate end point of survival in advanced breast cancer, it was necessary to demonstrate: (1) that the experimental treatment prolongs survival, (2) that the experimental treatment is associated with an increase in response rates, (3) that responders live longer than nonresponders, and (4) that the effect of treatment on survival disappears when response status is adjusted for. The last and crucial point implies that the effect of the experimental treatment on survival must be absorbed by its effect on objective response; as a consequence, survival among responders (or nonresponders) to the experimental treatment and to the control treatment should be equivalent, and no residual treatment effect should be observed.
In all trials, complete response (CR) was defined as the disappearance of all detectable tumor, and partial response (PR) was defined as a 50% or greater reduction in the tumor surface area, without appearance of new lesions. The minimum required response duration was 4 weeks in most of the trials. In particular, tumor response was assessed according to WHO criteria in eight trials and according to standard Eastern Cooperative Oncology Group criteria in one study.21 Only one study22 did not specify the tumor response assessment criteria that had been used. Stable disease was defined as a reduction of less than 50% or an increase of less than 25% in the tumor surface area, without new lesions. Progressive disease was defined as an increase of more than 25% in the tumor surface area or the appearance of any new lesion. The best overall response of each patient was considered in the analyses. The response rate was defined as the proportion of responders (complete or partial) among all patients. Survival time was considered from the day of random allocation to the day of death, regardless of the cause of death.
Statistical Methods
All the analyses were based on data from individual patients and were stratified for trial. The treatment effect on tumor response was quantified using the odds ratio, and its significance was assessed using the Mantel-Haenszel test. In analogy with other meta-analyses,7 to mirror the predicted effect of the experimental treatment on survival, odds of nonresponse were presented (proportion of nonresponses/proportion of responses). As a consequence, an odds ratio below unity indicates that the experimental treatment is associated with an increase in the probability of response. The treatment effect on survival was expressed in terms of hazard ratios; survival curves were estimated using the Kaplan-Meier method, and the statistical significance of the differences was assessed using the log-rank test. The prognostic effect of tumor response on survival was analyzed by a Cox model, with the response as a time-dependent covariate to eliminate the well-known bias caused by the fact that patients dying early cannot respond.26 Patients classified as responders but with missing date of response (51 patients, 5.6% of all responding patients) were arbitrarily assigned a time to response equal to the median response time over the whole cohort (2.5 months). Survival curves by tumor response were drawn by means of the Mantel-Byar (MB) method modified according to Simon and Makuch,27 with the landmark time set to the median time to response (2.5 months). To this purpose, the landmark method alone7 could not be used because of a large variability in the reported times to response. According to the modified MB method, the starting point to compute the survival curves is the landmark time; all the patients who died before the landmark were excluded. Patients with a response time of less than 2.5 months were included in the responders group, whereas all the other patients were included in the nonresponders group. Each patient obtaining a response was switched to the responders group at the time the response was recorded and became a censored observation in the nonresponders group. CIs for the MB estimates of median survival were estimated using the bootstrap technique. In the Cox model, which was aimed at evaluating whether treatment effect is mediated by tumor response, survival was first modeled as a function of treatment arm alone. Objective response was then introduced in the model as a time-dependent covariate to evaluate the modification in the coefficient (log hazard ratio) relating treatment arm to survival. To estimate surrogacy at the trial level (ie, the possibility to predict the treatment effect on survival based on its effect on response), as proposed by Buyse et al,28 two weighted regression lines were fitted using two different scales for treatment effects (the log hazard ratio of survival v the log odds ratio of response, and the difference of survival at median survival time v the difference of the percentage of response). The bootstrap method was used to estimate the 95% CIs of the coefficients of determination.
RESULTS
Effect of Treatment on Tumor Response
Two thousand seventy-seven patients had treatment and response data. The response rates for each trial and treatment arm are listed in Table 2. One thousand one hundred sixty-five patients (56%) showed no response, 688 patients (33%) achieved a PR, and 224 patients (11%) achieved a CR. An effect of the experimental treatment on response was observed in all trials, even though it was not always significant (Fig 1); overall, 37.9% and 50.2% of patients achieved an objective response in the standard and experimental groups, respectively. The pooled odds ratio showed a highly significant benefit of experimental intensified epirubicin over the standard treatment (odds ratio of nonresponse, 0.60; 95% CI, 0.51 to 0.72; P < .0001).
Effect of Treatment on Survival
Two thousand eighty-eight patients had treatment and survival data. The whole cohort with complete survival information had a median survival time of 16.3 months (range, 0 to 186 months). The survival benefits of intensified-dose chemotherapy on survival were much less obvious than the benefits on tumor response (Fig 2), with a 6% nonsignificant decrease in mortality (odds ratio, 0.94; 95% CI, 0.86 to 1.04; P = .22).
Effect of Tumor Response on Survival
The prognostic value of tumor response on survival was studied using a Cox model with response as a time-dependent covariate. Tumor response was confirmed to be a highly significant predictor of survival (P < .0001). The hazard ratio for PR compared with no response was 0.69 (95% CI, 0.62 to 0.77), and for CR compared with no response, the hazard ratio was 0.48 (95% CI, 0.40 to 0.57). Using the modified MB survival curves with a landmark set to the median response time (2.5 months), median survival time of patients achieving a PR was 21.3 months (95% CI, 19.2 to 22.4 months), and median survival time of patients achieving a CR was 28.8 months (95% CI, 25.4 to 45.3 months) compared with a median survival time of 14.6 months (95% CI, 13.9 to 15.4 months) for patients with no response (Fig 3).
Treatment Effect on Survival Adjusted for Tumor Response
In the Cox model, which used data from 2,055 patients with complete data on survival and response, the hazard ratio in favor of experimental treatment was 0.94 (95% CI, 0.86 to 1.04; P = .25). When tumor response was introduced in the model, the hazard ratio in favor of the experimental regimen became 1.005 (95% CI, 0.91 to 1.11; P = .92), indicating that no residual effect of the experimental treatment on survival, either positive or negative, was present once tumor response was adjusted for (Table 3). No significant variation in the hazard ratio for treatment was seen across categories of response (P for heterogeneity = .99), confirming that the prognosis was not affected by treatment within complete responders, partial responders, or nonresponders.
Prediction of Treatment Effect on Survival
The correlation between treatment benefits on survival and on tumor response was low, both between the log hazard ratio of survival versus the log odds ratio of response (R2 = 0.10; 95% CI, 0 to 0.43) and between the difference of survival (at median survival time) versus the difference of the percentage of response (R2 = 0.20; 95% CI, 0 to 0.65; Fig 4).
DISCUSSION
The validation of objective response to chemotherapy as a surrogate end point of survival in advanced solid tumors could have three potential implications that are strictly related but that deserve separate discussions. First, from a scientific perspective, it would provide a sounder basis for the efforts to find more active chemotherapy drugs and regimens because these could be assumed to prolong survival. The fact that such an effect (longer survival) is not easily detected in clinical trials can be easily explained if one takes into account the fact that, if objective response is truly a surrogate of survival, survival benefits involve only those patients who respond to the new regimen and would not respond to the old regimen (difference in response rates). The overall survival benefit is diluted and often may be negligible when the entire groups of patients treated with the old and new regimen are compared.
Second, one may think of efficacy trials that use objective response rate as the primary end point. This would make it possible to design smaller trials (because differences in response rates are bound to be much larger than differences in survival) and/or trials of shorter duration (because response to chemotherapy is detected within a few months of treatment start). This possibility is hindered by several problems, the most important of which is its limited applicability. In fact, validation of a surrogate end point requires the availability of data from randomized trials demonstrating the efficacy of the treatment under study, and it is treatment specific. As a consequence, extrapolation to therapies with a different mechanism of action, including different cytotoxic regimens, is not warranted. Furthermore, the development of more effective second-line regimens may decrease the relationship between response to first-line treatment and overall survival, thus depleting objective response of its validity as a surrogate end point. As a consequence, a prudent approach would be advisable in this setting, limiting the use of objective response as the primary end point to trials that assess the relative efficacy of different doses or modes of administration of the same drugs or that compare similar drugs with an identical mechanism of action. The weak correlation observed in this analysis between the effect of dose intensification on response rate and its effect on survival in the various trials, although expected (see preceding paragraph), further cautions against the use of objective response as the primary end point in clinical trials.
The third and most important implication of the demonstration that objective response to chemotherapy is a valid surrogate of survival would be its profound impact on the doctor-patient relationship. Today, oncologists, when discussing treatment options with patients with a metastatic solid tumor, have to face the moderate (and often frankly unsatisfactory) survival improvements that are observed in patient groups treated with the most active regimens. If response to chemotherapy is indeed a valid surrogate of survival, the difference in survival between responders and nonresponders is entirely caused by the effect of treatment; as a consequence, the average survival benefit equally shared by all treated patients becomes meaningless and should be replaced by the survival benefit experienced by responders versus nonresponders, conditional on the achievement of response. This different perspective may affect patients' choices because many patients with metastatic cancer who are reluctant to receive chemotherapy because of the marginal survival benefits may be willing to make an attempt in the presence of a non-negligible probability of response if it is shown that this response may translate into a substantial survival benefit.
To date, in solid tumors, investigations to validate objective response as a surrogate end point for survival were only carried out in advanced colorectal cancer. In a meta-analysis on original patients' data, experimental FU regimens were shown to be associated with a two-fold increase in response rate over standard FU regimens and with a 10% relative reduction in the odds of death. This survival difference was entirely a result of the effect of experimental FU regimens on response rate. As a consequence, the difference in survival between responders and nonresponders could be entirely considered as a beneficial treatment effect. However, treatment effects on response poorly predicted treatment effects on survival at the trial level, indicating that response could not be considered a valid surrogate for survival for the purpose of testing new treatments.
In breast cancer, several lines of evidence strongly suggest a similar role for objective response.7,12-15 A survival advantage was observed in several trials in which the absolute difference in response rate between the two treatment arms exceeded 10% to 15%; several meta-analyses of chemotherapy trials in advanced breast cancer indicated a survival advantage for more active treatments. Finally, long-term survivors of advanced breast cancer are almost exclusively patients who showed a complete response to first-line treatment for metastatic disease.10 Yet, no formal validation of objective response to chemotherapy as a surrogate end point of survival was ever published in breast cancer.
The present study, despite its weaknesses, provides evidence in support of objective response as a surrogate end point, suggesting that the intensification of epirubicin in first-line chemotherapy regimens in advanced breast cancer is associated with a survival advantage that is entirely a result of the effect of intensification on response rate. In this meta-analysis of 10 studies comparing dose-intensified epirubicin-containing regimens with the same regimens at standard dose-intensity, overall, an absolute 12% increase in response rate and a 6% relative reduction in the odds of death were observed. The survival difference completely disappeared when response was adjusted for, indicating that the survival benefit associated with response was independent of the activity of the regimen inducing the response. One weakness of this study is the lack of information on relevant prognostic factors in six studies, precluding the assessment of their confounding or modifying role in a multivariate setting. Another more crucial weakness of these analyses lies in the modest and not significant effect of the experimental treatment in survival comparisons. As a consequence, the first Prentice criterion is not fulfilled, raising questions on the whole validation process. This observation reflects the main criticism against the Prentice criteria largely discussed in the previous literature,28-30 which is focused on the fact that the requirement of a significant effect on the primary end point is overly stringent. However, these results, together with those reported by Buyse et al7 on colorectal cancer, provide strong evidence against the hypothesis that the longer survival observed in responders is a result of a third, unknown prognostic factor and does not indicate any survival benefit as a result of the treatment.3 According to this hypothesis, it was assumed that an active treatment, by inducing an objective response, might simply unmask patients with a better prognosis also in the absence of treatment. However, a corollary of this hypothesis is that survival of responders and nonresponders will decrease whenever the proportion of responses increases in association with the use of more active regimens. The similar survival observed within categories of response in our study between patients in more and less intense regimens, despite the non-negligible increases in response rates, strongly supports the notion that, in patients with advanced breast cancer, the achievement of an objective response by means of chemotherapy intensification is associated with a true increase in life expectancy.
These results do not provide direct and/or conclusive evidence on the general validity of objective response to chemotherapy as a surrogate of survival in advanced breast cancer. More specifically, on the basis of these results, it is not possible to conclude that objective response to any first-line chemotherapy (not to mention second-line chemotherapy) is associated with a survival benefit. However, these results provide a strong background for similar meta-analyses aimed at confirming the validity of objective response to different treatment modalities as a surrogate of survival. Because of the paucity of randomized trials comparing chemotherapy with no treatment, these analyses will have to be focused on studies comparing different types of chemotherapy with a different activity in advanced breast cancer. Should our results be consistently confirmed in these analyses, the overall evidence in support of the hypothesis that attainment of objective response in advanced breast cancer is associated with a survival benefit for the individual patient would become quite conclusive, with far-ranging implications in the management of individual patients and, to a lesser extent, in the design of clinical trials.
Authors’ Disclosures of Potential Conflicts of Interest
Although all authors have completed the disclosure declaration, the following authors or their immediate family members have indicated a financial interest. No conflict exists for drugs or devices used in a study if they are not being evaluated as part of the investigation. 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.
| Authors | Employment | Leadership | Consultant | Stock | Honoraria | Research Funds | Testimony | Other |
|---|---|---|---|---|---|---|---|---|
| Lucia Del Mastro | Pharmacia (B) | |||||||
| Riccardo Rosso | Pharmacia (A) | |||||||
| Marco Venturini | Pharmacia (B) |
Dollar Amount Codes (A) <$10,000 (B) $10,000-99,999 (C) ≥$100,000 (N/R) Not Required
Response odds ratios (OR) for experimental versus standard treatment. O, observed; N, number of patients; E, expected; VAR, variance.
Survival hazard ratios (HR) for experimental versus standard treatment. O, observed; N, number of patients; E, expected; VAR, variance.
Survival curves overall and according to response. NR, no response; R, response; PR, partial response; CR, complete response.
Treatment effect on survival versus treatment effect on response (circles' area proportional to the patients number).
Randomized Studies Comparing Standard Dose-Intensity Versus Intensified Dose-Intensity of Epirubicin-Containing Chemotherapy Included in This Meta-Analysis
Response Rate for Each Experimental Arm
Treatment Effect on Survival Adjusted for Tumor Response
Acknowledgments
We thank all the investigators who collected and provided the data used in this study.
Footnotes
-
Supported by grants from the Associazione Italiana per la Ricerca sul Cancro (grant No. 0301) and from the Ministero della Salute (Ricerca Corrente 2002-2004).
Presented at the IV Congresso Nazionale di Oncologia Medica, Torino, Italy, September 28-October 1, 2002 and at The 26th Annual San Antonio Breast Cancer Symposium, Antonio, TX, December 3-6, 2003.
Authors' disclosures of potential conflicts of interest are found at the end of this article.
- Received February 17, 2004.
- Accepted December 23, 2004.
