Factors Correlated With Progression-Free Survival After High-Dose Chemotherapy and Hematopoietic Stem Cell Transplantation for Metastatic Breast Cancer FREE

Breast cancer that has metastasized beyond regional lymph nodes or has recurred after primary treatment (designated as advanced breast cancer in this article) is an incurable disease.¹ High-dose chemotherapy followed by autologous hematopoietic stem cell transplantation (autotransplantation) is increasingly used to treat women with advanced breast cancer. Breast cancer is the most common indication for hematopoietic stem cell transplantation in North America.² Despite widespread use, the appropriateness and benefits of autotransplantation have been debated.^3- 5 Many phase 1 and 2 studies report efficacy in women with advanced breast cancer,^6- 17 and 1 small published randomized trial reports better survival rates than with standard-dose therapy.¹⁸ In the latter trial, 90 patients were randomized to receive either 2 cycles of high-dose (2.4 g/m²) cyclophosphamide, 35 to 45 mg/m² of mitoxantrone, and 2.5 g/m² of etoposide or 6 to 8 cycles of conventional-dose (600 mg/m²) cyclophosphamide, 12 mg/m² of mitoxantrone, and 1.4 mg/m² of vincristine. This trial has been criticized because the 2 groups were not balanced for all prognostic factors (ie, sites of metastases and estrogen-receptor status) and the conventional-dose arm used an uncommon regimen and had poorer-than-expected survival rates.¹⁹ Inclusion criteria for phase 1 and 2 studies differ, and numbers of subjects are usually small. In a recent overview of the use of autotransplantation in treating breast cancer, we reported data from 3500 women with advanced breast cancer treated at centers reporting to the Autologous Blood and Marrow Transplant Registry (ABMTR).² Three-year progression-free survival (PFS) rates ranged from 7% to 32% depending on response to conventional-dose chemotherapy prior to autotransplantation.

Identifying women most likely to benefit or very unlikely to benefit from autotransplantation for advanced breast cancer is important for patients, for physicians considering treatment options, and for investigators developing new trials. Factors predicting treatment failure after conventional therapy have been well established.^20- 26 Few multivariate analyses of such factors have been published for women receiving autotransplants.^13,15- 16 We studied 1188 women with advanced breast cancer who received autotransplants at 63 centers to determine factors associated with progression or death (treatment failure).

The ABMTR is a voluntary organization of more than 200 institutions performing autotransplantations primarily in North, Central, and South America. Centers provide data on consecutive patients receiving autotransplants to a Medical College of Wisconsin statistical center. The ABMTR defines an autotransplant patient as an individual receiving treatment with a sufficiently high chemotherapy dose to warrant autologous bone marrow– or blood-derived hematopoietic stem cell support. Whether a particular treatment regimen requires stem cell support is decided by the treating institution.

Data collection was initiated by the ABMTR in 1992. Retrospective data collection occurred for women receiving autotransplants between 1989 and 1992 and prospectively thereafter. Institutions that participate register consecutive patients receiving autotransplants for all disease indications. According to data from the Centers for Disease Control and Prevention Hospital Surveys,^27- 28 about half of autotransplantations performed in North America have ABMTR registration.

The ABMTR collects data at 2 levels: registration and research. Registration data include disease type, age, patient sex, pretransplantation disease stage and response to chemotherapy, diagnosis date, graft classification (bone marrow– and/or blood-derived stem cells), high-dose conditioning treatment, disease progression and survival following transplantation, development of a new malignancy, and cause of death. Requests for information on progression or death for registered patients occur every 6 months. All ABMTR teams contribute registration data. Research data are collected on comprehensive report forms submitted for consecutive registered patients in transplant centers with the required data management support to submit the research information, including pretransplantation and posttransplantation clinical information such as tumor size and pathology, sites of disease, menopausal status, hormone receptor status, all breast cancer treatments before and after transplantation, clinical status (including cardiac, pulmonary, renal, and liver function) before and after transplantation, doses of high-dose chemotherapy, blood or marrow graft treatment, and sites of posttransplantation progression. Demographics and survival rates of women with advanced breast cancer in the registration and research databases are similar.

This analysis uses ABMTR research data reported for 1188 consecutive women with advanced (metastatic and locally recurrent) breast cancer who received transplants in 63 centers between January 1, 1989, and January 31, 1995. Metastatic disease excludes disease of the regional lymph nodes (ipsilateral axillary and internal mammary lymph nodes) but includes the ipsilateral supraclavicular lymph nodes.¹ Locally recurrent disease denotes disease occurring in the breast and/or regional lymph nodes after a disease-free interval (DFI). Women with disease recurrence solely in the contralateral breast (n<100) were excluded because of difficulty in distinguishing recurrence from a new primary site of disease (surival was similar to that of women with chemotherapy-responsive disease). One hundred seven women were excluded because of incomplete data for sites of metastases or date of diagnosis of metastatic disease. The survival rate of these women was not different from that of the 1188 included in the analysis (log-rank test for difference in survival probability, P = .58). Women were considered to have hormone receptor–positive tumors if either estrogen-receptor or progesterone-receptor assays were positive; those reported as having borderline receptor levels (n = 36) were grouped with the receptor-negative patients. Women in whom metastases were found within 1 month of diagnosis of breast cancer were grouped with women with metastases at first presentation. Women with pleural or parenchymal lung disease were considered a single group.

The primary outcome was treatment failure, the inverse of PFS. Events were death, disease progression, and in women with a complete response, breast cancer recurrence. Women surviving without progression or recurrence were censored at last follow-up. Univariate probabilities of survival and PFS were calculated using the Kaplan-Meier method. Variables were tested in univariate analysis for their association with treatment failure using the Cox proportional hazards regression model; using the time-dependent covariate method, these analyses also were used to examine whether the proportional hazards assumption was met. Optimal cut points for categorizing continuous variables (including DFI) were determined using Martingale residual plots.²⁹ A forward stepwise selection method with a significance level of .05 was used to select variables for the multivariate model. Covariates in the final multivariate model were tested for proportional hazards using the time-dependent covariate method and for first order interactions. Results of tests for "center effects" (intercenter differences unexplained by known covariates) were not statistically significant.³⁰

Studying effects of posttransplantation maneuvers on outcome, such as posttransplantation tamoxifen therapy and radiation therapy, must account for bias introduced by deaths occurring before intended treatments were administered. For example, women who died before planned posttransplantation tamoxifen and/or radiation treatments could be administered might be considered to be part of the no-treatment group even though such treatment had been planned. This bias would artificially increase the proportion of adverse events in the nontreatment group if all patients were considered from time of transplantation and was overcome by only studying effects of posttransplantation tamoxifen (or in <1% of women, other hormonal agents) and radiation therapy in women surviving without disease recurrence longer than 6 months posttransplantation. Tamoxifen therapy was analyzed in 4 groups: women who were hormone receptor–positive and received tamoxifen pretransplantation only, posttransplantation only, at both times, and at neither time. These 4 groups were compared simultaneously with women with hormone receptor–negative tumors, adjusting for other variables found to be significant in multivariate analysis. Prophylactic posttransplantation radiation therapy (radiation administered to involved breasts or lymph nodes for local control) was similarly studied in women surviving for longer than 6 months posttransplantation.

Because of multiple comparisons, only P values <.01 were considered significant; P values of .01 to .05 were considered marginally significant and presented to show trends.

The distribution of patient-, disease-, treatment-, and transplantation-related characteristics of the 1188 women are presented in Table 1. Median age was 44 years (range, 18-70 years). Median follow-up after transplantation was 29½ months. The 3-year probabilities of survival and PFS were 31% (95% confidence interval [CI], 28%-34%) and 13% (95% CI, 10%-16%), respectively (Figure 1).

Eight factors were statistically significant (P<.01) and 4 marginally significant (P = .05-.01) in univariate analyses of association with treatment failure. Significant variables were breast cancer stage at diagnosis, hormone receptor status, use of adjuvant chemotherapy, initial DFI, response to pretransplantation chemotherapy, pretransplantation Karnofsky performance score, and number and sites of metastases. Marginally significant variables were age, use of radiation therapy prior to transplantation, type of high-dose therapy used, and year of transplantation.

Factors significant in the final multivariate model of risk of treatment failure are shown in Table 2. All variables statistically significant in univariate analyses were significant in the multivariate model (see "Methods" section). Of the marginally significant variables, only age remained in the final model. Women older than 45 years had a relative hazard of treatment failure of 1.17 (95% CI, 1.02-1.33) vs younger women (P = .02). A Karnofsky performance score of less than 90% pretransplantation was associated with increased risk of treatment failure (relative hazard, 1.27; 95% CI, 1.07-1.51; P = .005). Hormone receptor–negative tumors had a relative hazard of treatment failure of 1.31 (95% CI, 1.15-1.51) vs hormone receptor–positive tumors (P<.001).

There was an interaction between the effect of prior adjuvant chemotherapy and initial DFI (interval from diagnosis of breast cancer to detection of metastatic disease). First, women whose first presentation was with metastatic disease were not eligible for adjuvant therapy. Additionally, the effect of prior adjuvant therapy on posttransplantation treatment failure differed in women with initial DFIs of no longer than 18 months vs longer than 18 months. These 2 variables were therefore combined in the final model into a single covariate with the following categories: metastatic breast cancer at diagnosis, no adjuvant therapy for initial diagnosis of localized breast cancer, adjuvant therapy as part of initial treatment with a DFI of no longer than 18 months, and adjuvant therapy with a DFI longer than 18 months. Women not receiving adjuvant therapy for initially localized disease that later recurred had the same risk of treatment failure after autotransplantation as women with metastases at diagnosis, regardless of the duration of their initial DFI. Women receiving adjuvant chemotherapy followed by a DFI longer than 18 months had a higher risk of treatment failure than those not receiving adjuvant therapy (relative hazard, 1.31; 95% CI, 1.10-1.56; P = .002) although not as high as those with a shorter (≤18 months) DFI after adjuvant chemotherapy (relative hazard, 1.99; 95% CI, 1.62-2.43; P<.001 for comparison with no adjuvant therapy, P<.001 for comparison with a DFI >18 months after adjuvant therapy).

Number and sites of metastases were also important factors in PFS. Four important prognostic groups were determined in the final model. Women were included in the reference group and had similar risks of treatment failure if they had 1 or 2 sites of metastases outside the liver or central nervous system (CNS). Women with metastases in the CNS, liver, or in 3 or more organs of any kind had a poor prognosis.

Risk of treatment failure correlated strongly with pretransplantation response to chemotherapy. Women with a partial response to chemotherapy had a relative hazard of 1.65 (95% CI, 1.36-1.99; P<.001), and women with resistant disease had a relative hazard of 1.87 (95% CI, 1.54-2.27; P<.001) vs women in complete remission but were not significantly different from each other (P = .09). Women with indeterminate sensitivity (due to bone-only disease, single-site disease excised or irradiated pretransplantation, or an untested response to chemotherapy) had a risk of treatment failure similar to the reference group. Figure 2 presents Kaplan-Meier estimates of PFS according to chemotherapy sensitivity pretransplantation, number and sites of metastases, adjuvant chemotherapy and DFI, and number of adverse prognostic factors. The 3-year probability of PFS was 43% (95% CI, 27%-61%) for the 38 women with no adverse risk factors and 4% (95% CI, 2%-8%) for the 343 with more than 3 adverse risk factors.

Effect of posttransplantation tamoxifen therapy was analyzed in women surviving without progression 6 months posttransplantation. After adjustment for other variables in the final model described above, posttransplantation tamoxifen therapy was associated with decreased treatment failure in women with hormone receptor–positive disease. Use of tamoxifen pretransplantation did not influence posttransplantation PFS, regardless of whether tamoxifen was given posttransplantation. Use of radiation as part of planned posttransplantation treatment was not associated with improved outcome (Table 3).

In this study of 1188 women receiving autotransplants for advanced breast cancer, use of posttransplantation tamoxifen in women with hormone receptor–positive tumors significantly reduced risk of treatment failure. Several other factors were significant predictors of treatment failure in multivariate analysis, including age older than 45 years; Karnofsky performance score less than 90; metastases in the CNS (approaching significance), liver, or involving 3 or more sites; hormone receptor–negative tumors; use of adjuvant chemotherapy; initial DFI of no longer than 18 months in women receiving adjuvant chemotherapy; and poor response to chemotherapy pretransplantation. Surprisingly, there was little difference in risk of treatment failure between women with partial response and no response to chemotherapy pretransplantation. From this study, it is not possible to state whether pretransplantation conventional-dose therapy affected the natural progression of the disease or just identified women who would do poorly with autotransplantation.

The survival of women with advanced breast cancer varies widely, regardless of treatment given. Prior studies attempting to define patient, disease, and treatment factors predicting progression or death following therapy are summarized in Table 4.^{8,13,15- 16,20- 26,31} Interpretation of these studies and their sometimes conflicting results is difficult. Variations among studies likely relate to small study population sizes and differing selection criteria. Also, in most studies, statistical techniques were not used to adjust for important interactions between factors and changes in effect of factors over time.²⁹ In this study, a large population of consecutive patients receiving autotransplants was analyzed, interactions among factors were examined, and changing effects of factors on treatment failure over time were considered. Women receiving autotransplants are no doubt a selected group of women with advanced breast cancer; however, we believe this study accurately determines factors affecting outcome of women receiving this therapy.

There are several important negative findings from this study. Interval from diagnosis of metastases to autotransplantation, prior hormonal therapy, prior anthracycline therapy, use of growth factors to enhance marrow recovery, and source of stem cells did not affect outcome. Several factors significantly correlated with treatment failure in univariate analysis but not after adjustment for other prognostic factors in multivariate analysis. These included stage of breast cancer at diagnosis, prior radiation therapy, high-dose therapy regimen, and most recent calendar year studied. Correlations among variables may account for some differences among published studies. Outcomes were similar in the data provided by 63 teams after adjustment for patient characteristics.

The only intervention found to reduce treatment failure after autotransplantation was the subsequent use of tamoxifen; it should therefore be recommended for all women with hormone receptor–positive tumors. Failure to determine a superior high-dose chemotherapy regimen suggests that women should be treated with the least toxic autotransplantation regimens (unless they volunteer to participate in a clinical trial). The absence of an effect on treatment failure of interval from diagnosis of metastases to autotransplantation in this study may indirectly support a recently reported benefit on survival of delaying autotransplantation until recurrence of disease in women achieving a complete remission after standard-dose chemotherapy.³¹

The question of superiority of high-dose vs standard-dose therapy for advanced breast cancer is not addressed by this study. However, the data from this study and from the studies reviewed in Table 4 illustrate the importance of nontreatment factors in determining outcome. Age, hormone receptor status, DFI, sites of disease, and chemotherapy responsiveness are all important predictors of outcome, regardless of the treatment administered. Selection biases in uncontrolled studies could lead to substantial outcome differences that have little or nothing to do with treatment effect.^32- 33 The need for large, randomized controlled trials is clear, and 1 such trial was recently completed in the United States.³⁴ In contrast to the prior report from South Africa,¹⁸ the US study showed no benefit from high-dose chemotherapy. Results of the multivariate analysis of the current data set suggest that stratification for important prognostic factors to allow valid analysis of subgroups or limiting the eligible population to those who are more likely to benefit from high-dose chemotherapy should be considered in formulating the design of such trials. The data in this analysis suggest that some women are very unlikely to benefit from receiving autotransplants; women with resistant disease, CNS metastases, 3 or more metastatic sites, and those who had received adjuvant chemotherapy all had PFS rates of less than 10% at 3 years posttransplantation (Figure 2, A, B, and C). Such women should probably not be considered for autotransplantation except in the context of clinical trials designed to test novel regimens or approaches that might improve this poor prognosis. On the other hand, women with limited sites of metastases and complete response to conventional chemotherapy might be more likely to benefit from dose-intensification to eliminate minimal residual disease. The current study predicted better transplantation outcome in such women, but their outcome may also be good with conventional chemotherapy; controlled comparisons are needed. There were few such women in the 2 reported randomized trials.^18,34

In this study of a large number of women consecutively treated with autotransplantation for advanced breast cancer, we determined patient-, disease- and treatment-related factors associated with treatment failure. The large number and complex interaction of prognostic factors determined in this study highlight the need for careful statistical analysis of large numbers of patients to determine prognostic factors and impact of new therapies in treating women with advanced breast cancer. The design of new clinical trials should consider data available from information resources such as the ABMTR and similar organizations.