PATIENTS AND METHODS

The principal eligibility criterion was a histologically proven primary breast cancer of stage T1a-3aN+M₀. Primary local treatment was a modified radical mastectomy (Patey technique) with en bloc axillary dissection with at least six identifiable lymph nodes in the specimen. It was mandatory to obtain the patient’s informed consent before entering her onto the trial. Patients were required to be younger than 65 years and to present with a Karnofsky index of at least 60. During the recruitment period, the study was also opened for older patients if they were in good condition.

The study had a 2 × 2 factorial design. Patients were randomized to receive either three or six cycles of CMF and to receive either 2 years of hormonal therapy with TAM or no hormonal therapy. This resulted in four treatment arms: (A) 3 × CMF; (B) 3 × CMF + TAM; (C) 6 × CMF; and (D) 6 × CMF + TAM. A comparison of the treatment effect of three with six cycles of CMF can be obtained by combining patients of treatment groups A and B and comparing them with the combined patients of groups C and D. To evaluate the effect of TAM, patients in B plus D have to be compared with those in A plus C. Additionally, and only in such a 2 × 2 factorial design, the interactive effect of cytotoxic and hormonal therapy can be investigated. According to the statements of the National Cancer Institute consensus conference,⁵ in December 1986, the protocol was modified, with premenopausal patients being randomized only between treatment arms A and C.

Chemotherapy was administered according to the modified Bonnadonna CMF regimen, which consisted of 500 mg/m² cyclophosphamide, 40 mg/m² methotrexate, and 600 mg/m² fluorouracil administered intravenously on day 1 and 8 of a 4-week treatment period. Hormonal treatment consisted of a daily dose of 3 × 10 mg TAM orally over 2 years, starting after the third cycle of CMF.

Time between successive cycles of chemotherapy was extended up to 6 weeks if WBC (< 2500/μL) or platelets (< 100,000/μL) were insufficient. After 6 weeks, a dose reduction of 50% was recommended if these values were still to small (WBC < 3000/μL or platelets < 100,000/μL). In the case of an infection, chemotherapy was delayed. Liver and renal disease also caused reduction of methotrexate and cyclophosphamide or termination of chemotherapy.

As potential prognostic factors, the following characteristics were determined at time of primary diagnosis: patient’s age, menopausal status, nodal status, tumor location, tumor size, tumor grade according to Bloom and Richardson,⁶ histologic tumor type, and estrogen and progesterone receptor status. Methods for determination of these factors have been described previously.^3,7 Registration and randomization was performed centrally by phone, usually at the same time. The center was used as a stratification criterion, and within each center, we performed a block randomization. Although it was allowed shortly after the operation, we encouraged randomization as late as possible, ideally after three cycles of basic chemotherapy, to increase the compliance. With this approach, we investigated whether it is necessary to give three additional cycles of CMF to patients who tolerate a short-time basic chemotherapy of three cycles. Obviously, blinding was not possible, and the investigators were aware of the treatment.

Patients were followed up at regular intervals to ensure detection of any kind of recurrence at the earliest time possible. Examinations were performed every 3 months during the first 2 years after operation, every 4 months during the following 3 years, every 6 months in years 6 and 7, and in annual intervals from year eight and on. Patients were observed until the middle of 1997, providing a median follow-up time of about 9 years. Not all patients followed the observation schedule planned in the study protocol. At the end of the observation period, documentation of follow-up visits was missing for more than 2 years for several patients. Therefore, information on the actual survival status of patients with incomplete follow-up was requested from the corresponding registration offices, which increased the median follow-up time with respect to OS to 10 years. This information was used only for the calculation of OS. For calculation of time to event, only the last available information from the clinical center was used.

Recurrence was defined as locoregional or distant (metastases in distant sites or contralateral or secondary tumor). The first event of failure was classified as an isolated locoregional recurrence (appearance of local or regional failure before the diagnosis of distant metastases or death), as distant failure (with or without a simultaneous locoregional recurrence), or death without previous event.

EFS (named recurrence-free survival in the first analysis³ but identical definition), which was defined as time from mastectomy to the first occurrence of locoregional or distant recurrence, contralateral tumor, secondary tumor, or death, and OS (time from operation to death) had been stated as the primary end points. Quality of life was stated as a secondary criterion, but only Karnofsky Index was determined. However, Karnofsky Index will not be analyzed because it is not considered as a useful measure for this criterion. Side effects of chemotherapy were documented in detail and will be presented, although they were not explicitly stated as outcome measures.

For univariate analyses, EFS and OS rates were calculated according to the Kaplan-Meier method.⁸ Event-specific rates for the first event of failure after mastectomy were estimated using the methodology of cumulative incidence rates.^9,10 The relative risks (RRs) between different groups defined by treatment or prognostic factors with corresponding 95% confidence intervals (CIs) were determined using univariate Cox regression models.¹¹ The P values were based on the log-rank test.¹⁰ A simultaneous assessment of the effects of different prognostic factors was performed within a multiple regression analysis using the Cox model.⁹ From this model, estimates of RRs with corresponding 95% CIs were calculated. Factors occuring on more than two levels were coded using dummy variables to estimate the RRs between the different levels separately. P values were based on Wald tests.⁹ In the first multivariate analysis,³ factors that had P values less than .05 in the univariate analyses were included. The model was stratified with regard to randomization to TAM. This final model was not changed for the current analysis, only the regression coefficients were re-estimated.

Interactive effects of chemotherapy and hormonal treatment were estimated and tested in the subpopulation of patients randomized to all four treatment arms. This analysis was not adjusted for any other factor. Furthermore, treatment effects were estimated in subgroups that were selected to be in accordance with the IBCSG analysis⁴ or with the current discussion on treatment-covariate interaction in the literature. These analyses were not specified in the protocol. Except for the eight patients who were excluded because no information was available (Fig 1), the analyses were performed on an intent-to-treat basis.

The sample size of the study was originally planned to total approximately 1,500 patients. This large number was chosen to detect a treatment difference of 10% in terms of 5-year survival probabilities with a power of 95% already after a short follow-up. For this purpose, the effective sample size, ie, the expected number of events, should be 240. Accrual of the trial was stopped mainly because of administrative reasons (slow accrual and the termination of financial support because of the end of a government sponsored clinical trial program) and not on the basis of an interim analysis.³

RESULTS

From July 1984 to December 1989, 481 patients were randomized, of whom seven were excluded because of the total lack of cooperation of three clinical centers. One woman had only undergone a breast conserving surgery, which was considered a major protocol deviation at that time. There were 262 eligible patients, with a treatment preference for one of the four arms, who were registered and entered onto the nonrandomized part of the covering Comprehensive Cohort Study.⁷ The number of patients eligible but not randomized in the study is certainly much larger; exact figures are unknown.

Figure 1 shows the study profile concerning compliance to chemotherapy treatment and follow-up. Previously described³ details on CMF and TAM compliance show that a clear distinction in the amount of chemotherapy between the two treatment groups was achieved. Of the 184 patients randomized on TAM, 120 patients received it for at least 1.5 years, 19 patients for less than 1 year, and seven patients refused TAM. Information on duration was unknown for 26 patients. Four patients randomized to the no-TAM group took TAM.

The distribution of patient characteristics is comparable between the two CMF groups (Table 1). Approximately 12% of the patients were 40 years of age or younger, 42% were premenopausal, and 57% of the patients had between one and three positive lymph nodes. Median tumor size was 25 mm, and approximately 7% of the tumors were larger than 50 mm. The distribution of patient characteristics with respect to hormonal treatment indicates that 51% of the patients treated with TAM had less than four positive nodes compared with 61% of the patients from the no-TAM group. All other characteristics were well balanced.³

On the basis of the first event of failure, for all patients the cumulative incidence rate of locoregional recurrence at 2 years was 8.2% (95% CI, 6.1% to 11.1%, and the rate of distant failure was 16.7% (95% CI, 13.6% to 20.5%). At 10 years, these cumulative rates increased to 19.9% (95% CI, 16.3% to 24.1%) for locoregional recurrence and to 41.3% (95% CI, 36.9% to 46.3%) for distant failure. So far, 81 locoregional recurrences, 168 distant recurrences, 11 secondary tumors (three breast and eight others), and 11 deaths without previous event have been documented as first events. This results in 271 events for EFS. Including the information from the registration offices, 226 patients were known to have died.

In univariate analyses, duration of chemotherapy did not show any effect on EFS (RR of 6 × CMF v 3 × CMF: 0.96; 95% CI, 0.76 to 1.22; P = .75) or OS (RR of 6 × CMF v 3 × CMF: 0.94; 95% CI, 0.72 to 1.22); P = .63).

The estimated effect of TAM was a bit larger, although not significant for either EFS (RR of TAM v no-TAM: 0.93; 95% CI, 0.71 to 1.21; P = .58) or OS (RR of TAM v no-TAM: 0.85; 95% CI, 0.63 to 1.13; P = .26). The corresponding estimates of EFS and OS according to duration of chemotherapy are shown in Fig 2, and estimates according to hormonal treatment are shown in Fig 3. Number of involved lymph nodes, progesterone receptor, tumor grading, and tumor size were significant prognostic factors for both EFS and OS. ER showed a weak but nonsignificant effect on both survival criteria.

In the multivariate analysis, the effect of TAM was somewhat stronger and nearly reached significance for OS (Table 2). Number of lymph nodes and progesterone receptors exhibited strong prognostic effects; the effects of tumor size and grade were of borderline importance (Tables 2 and 3⇓).

The results for an interactive effect of hormonal treatment and the duration of chemotherapy are listed in Table 4. There is some indication of a weak interactive effect for EFS; however, this is not significant and the CIs of the estimated effects are wide. The corresponding analysis for OS gives no indication of an interactive effect between the two treatments.

Separate analyses of treatment effect by menopausal status do not indicate the existence of a strong interaction with treatment. Investigation of the effect of TAM in the two subgroups defined by the ER status showed for EFS a trend towards a larger effect in ER-positive patients (RR: 0.69; 95% CI, 0.48 to 0.99) compared with ER-negative patients (RR: 0.95; 95% CI, 0.64 to 1.41), whereas the estimated effect for TAM on OS was nearly identical in the two ER subgroups. In the ER subgroups, the effect of duration of chemotherapy was nearly identical to the overall effect.

DISCUSSION

Our trial was the first to adress the important question of reducing six cycles of CMF to three cycles in node-positive breast cancer patients. The results, based on 9 years follow-up for EFS and 10 years for OS, are similar to the estimated effects after approximately 5 years follow-up.³ In multivariate analysis, the estimated RR in favor of the longer duration is only 0.95 for EFS (95% CI, 0.74 to 1.21) and 0.90 for OS (95% CI, 0.69 to 1.18). If the trial would have been planned to show equivalence between three and six cycles of CMF, an often used criterion would be whether the 90% CI for the RR lies between 0.8 and 1.25. For EFS, the 90% CI ranges from 0.77 to 1.17 and is, therefore, outside of the equivalence interval. This indicates that formally we could not show equivalence between three and six cycles of CMF. The current analysis provides further evidence that duration of CMF may be reduced to three cycles only, especially when taking into consideration the increased percentage of patients with short-term toxicity if chemotherapy is given for six cycles. Details on toxicity were published in the first report of this study.³ This report also gives details about the good treatment compliance, which implied that we achieved a sufficiently clear distinction in the amount of chemotherapy between the two treatment groups.

An advantage of the shorter treatment duration was also published in analyses combining quantity and quality of life in a single measure (quality-adjusted life-years and quality-adjusted time without symptoms and toxicity) in another trial comparing three and six cycles of CMF in premenopausal patients from the IBCSG.¹² In the first report of this study,⁴ investigating the usual criteria EFS and OS only, the authors concluded that three courses are not sufficient compared with a longer treatment duration. However, the advantage was restricted to patients younger than 40 years. In 1167 patients 40 years of age or older, the estimated effect of three versus six cycles on EFS was almost one. In the 308 patients younger than 40 years, a strong effect in favor of six cycles (RR approximately 0.75) was observed. In our study, approximately 42% of the patients were premenopausal, and there were only 50 patients younger than 40 years. For the latter group, crossing of the EFS rates was observed. This may have been solely caused by chance because this subgroup is too small for a sensible analysis. We could also not confirm the tendency for a larger advantage of six cycles of CMF in ER-negative patients. In our study, the estimates of the RRs for EFS were nearly identical in the two subgroups defined by the ER status and only slightly different for OS (0.95 in ER-negative patients and 0.84 in ER-positive patients). Despite the restriction to premenopausal patients in the IBCSG trial, the ER distributions were very similar, with approximately 30% of patients with concentrations less than 10 femtomole (fmol)/mg. In the analysis, we followed our earlier approach using a cut point of 20 fmol/mg to define receptor positivity. One possible other reason for the differences may be because of the different doses and schedules of CMF treatment used by the IBCSG,⁴ who gave cyclophosphamide 100 mg/m² orally on days 1 to 14. Recently, Goldhirsch et al¹³ discussed widely varied doses and schedules of CMF. Based on a discussion of several trials, they recommend ’classical’ CMF where oral cyclophosphamide is given on days 1 to 14 and intravenous methotrexate and fluorouracil on days 1 and 8. However, no clear evidence is available that our ’modified’ CMF is inferior. Furthermore, 29% of the patients in the IBCSG trial had breast conserving surgery, whereas all patients in our study had a mastectomy. However, according to the current knowledge this should have no influence on the results. Besides the IBCSG and our study, there are at least two additional trials, one of the German Adjuvant Breast Cancer Group and one of a British group, investigating the issue of three versus six cycles of chemotherapy.

To investigate whether toxic effects of chemotherapy are outweighed by an improvement in EFS and OS, Gelber et al¹⁴ conducted a meta-analysis of quality-adjusted survival in postmenopausal women who were all treated with TAM. Approximately 4,000 patients from nine randomized trials were summarized to compare chemoendocrine therapy with endocrine therapy alone. The small advantages in EFS and OS rates for patients treated with chemotherapy were balanced by acute toxic side effects, and chemoendocrine therapy did not provide more quality-adjusted survival than TAM alone.

Based on the results so far, we expect more definitive conclusions concerning the reduction of chemotherapy to three cycles for patients older than 40 years by a combined analysis of studies comparing three versus six cycles, which would supplement and extend the recent meta-analyses by the Early Breast Cancer Trialists’ Colloborative Group (EBCTCG).¹⁵ The effects of TAM observed on our study are, although not significant, in reasonable agreement to the corresponding meta-analyses of the EBCTCG.¹⁶

APPENDIX

The following were participants in the GBSG Study Group: Study Coordination: G. Bastert, Heidelberg, H. Bojar, Düsseldorf, K. Hübner, Frankfurt, H. Scheurlen, Heidelberg, M. Schumacher, Freiburg, P. R. G.: Prinicipal Investigators and Randomizing Centers: Marienhospital Essen-Altenessen, R. I. A. Neumann, R. P. DeDyker; Universitätsklinik Göttingen, H. Rauschecker, W. Gatzemeier; St Josef-Krankenhaus Sterkrade Oberhausen, G. Ott; Universitäts-Frauenklinik Homburg/Saar, G. P. Breitbach, D. Mink; Stădtische Frauenklinik Esslingen, A. Warecka-Speichermann, J. Maier; Amalie-Sjeveking-Krankenhaus Hamburg, W. Czopnik; Kreiskrankenhaus Herford, R. Bentler, G. Job; Stădtisches Kliniken Kassel, G. Hopf; Kreiskrankenhaus Kronach, J. Bühler, C. Beyerle; H.-G.-Walther-Krankenhaus Lichtenfels, P. Lehmann; St Lukas-Klinik Solingen, D. Mancarella; Elisabeth-Krankenhaus Straubing, S. Stahl-Kuschel, E. Regenbrecht; Krankenhaus St-Josef Wuppertal-Elberfeld, S. Kürcher, B. Heien-Conrads; Kreiskrankenhaus Rotthalmünster, L. Kronpaβ; Krankenhaus St Josef Regensburg, U. Prasser, G. Ulrich, A. Köppl; St Franziskus-Hospital Bielefeld, J. Jünemann; Zentralkrankenhaus Bremen, H. Franz, B. Ötting-Kipke; Krankenhaus der Barmberzigen Brüder Trier, H. H. Hennekeuser. B. Rendenbach; Further centers: Knappschaftskrankenhaus Bardenberg Würselen, St Katharinen-Hospital Freehen, Stüdtisohen Krankenhaus München-Neuperiach. Krankenhaus Aalen, St Josefs-Krankenhaus Adenau, St Rochus-Hospital Castrop-Rauxel, St Josef-Krankenhaus Essen, Kreiskrankenhaus Groβ-Umstadt, Stădtisches Krankenhaus Idar-Oberstein. Borromäus-Hospital Leer. St Hildegardis-Krankenhaus Mainz, Kreis- und Stadtkrankenhaus Marktredwitz, Stădtisches Krankenhaus Pirmasens, Stadtkrankenhaus Rüsselsheim, DRK-Krankenhaus Saarlouis, Stădtisches Krankenhaus Sindelfingen, Stadtkrankenhaus Traunstein. Paracel-sus-Krankenhaus Ruit Ostfildern, St Vinzenz- und Elisabeth-Hospital Mainz, Verbandskrankenhaus Schwelm, Klinik Dr Opitz Regensburg. St Elisabeth-Klinik Saarlouis, and Kreiskrankenhaus Annweller.