Despite advances in therapies for breast cancer, improvement in survival for patients with recurrent or metastatic breast cancer has been difficult to establish. The objective of the current study was to determine whether the survival of women with recurrent breast cancer has improved from 1974 to 2000.
The authors analyzed the survival experience of 834 women who developed recurrent breast cancer between November 1974 and December 2000. All patients had been treated previously with adjuvant anthracycline-based protocols. Patients were divided into five consecutive groups based on year of breast cancer recurrence, and survival was compared across the five groups. Because some prognostic variables were divided unevenly divided among the cohorts, a multivariate model was created to determine the association of year of recurrence and survival after accounting for other prognostic factors.
In the unadjusted analysis, there was a statistically significant improvement in survival across the five groups, and the more recent cohorts had longer survival (P < 0.001). Other variables that predicted longer survival after breast cancer recurrence included smaller initial tumor size, lower stage of disease, fewer lymph nodes involved, longer disease-free interval, estrogen receptor–positive tumors, and nonvisceral dominant site of disease recurrence. In the multivariate analysis, which adjusted for these prognostic factors, year of recurrence was associated with a trend toward improved survival, with a 1% reduction in risk for each increasing year.
In 2003, approximately 212,600 new cases of breast cancer will be diagnosed in the United States, and 39,800 women can be expected to die of metastatic disease.1 Despite advances in supportive care and the development of numerous, active chemotherapy regimens and hormone therapies, improvement in survival for patients with recurrent or metastatic breast cancer has been difficult to establish. Overall breast cancer mortality rates have been declining: From 1995 to 1999, breast cancer death rates decreased by 3.2% annually.1 Nonetheless, it remains unclear whether the decrease in the death rate is due to greater proportions of patients who were diagnosed with early-stage disease or due to improvements in treatment.
Therapeutic advances clearly have been making an impact on the survival of women with early-stage breast cancer. Two early adjuvant chemotherapy trials that started in the 1970s demonstrated that adjuvant chemotherapy resulted in improved disease-free survival. The first trial demonstrated the superiority of 2 years of adjuvant melphalan compared with placebo, and the second trial demonstrated improved survival for patients treated with 1 year of adjuvant cyclophosphamide, methotrexate, and 5-fluorouracil.2, 3 Because most adjuvant trials have had insufficient statistical power to address issues of survival adequately, the Early Breast Cancer Trialists' Collaborative Group has performed meta-analyses on all known randomized trials.4, 5 In successive survival analyses, these aggregate data have established the benefit of adjuvant chemotherapy and endocrine therapy in women with lymph node–positive and lymph node–negative disease, the superiority of polychemotherapy compared with single-agent regimens, and the superiority of anthracycline-containing regimens and has established the optimal duration of both adjuvant chemotherapy and tamoxifen therapy. More recently, a large randomized clinical trial (Cancer and Leukemia Group B trial 9344) has demonstrated that the addition of paclitaxel to an anthracycline-based adjuvant regimen results in improved disease-free and overall survival.6
Despite the clear advances that are being made in the adjuvant treatment of breast cancer, a favorable effect of new therapies on the survival of patients with metastatic disease has not been established. However, several studies have suggested that newer therapies are improving survival rates. In patients with anthracycline-refractory disease, it was shown that single-agent docetaxel prolonged survival compared with an older regimen of mitomycin plus vinblastine.7 Slamon et al. reported improved survival for women with HER-2-overexpressing tumors who were treated with chemotherapy plus trastuzumab compared with women who were treated with chemotherapy alone.8 Similarly, it was found that women who were treated with combination capecitabine and docetaxel had prolonged overall survival compared with women who were treated with either agent alone.9 Those trials suggest that newer therapies may be having a favorable impact on the survival of women with metastatic breast cancer.
Although those trials have shown the survival impact of different chemotherapy regimens, the impact of new therapies on overall survival rates for women with metastatic disease has not been established clearly. Improvement in survival over time would suggest that, in aggregate, new therapies are helping women with recurrent disease live longer. Therefore, the current study was undertaken to evaluate whether the survival of women with recurrent breast cancer treated at The University of Texas M. D. Anderson Cancer Center (Houston, TX) has improved over the past 25 years.
MATERIALS AND METHODS
The current study reports on a group of 834 patients who experienced recurrent breast cancer between November 1974 and December 2000. This group represents all patients who initially were treated on The University of Texas M. D. Anderson Cancer Center Institutional Review Board–approved adjuvant chemotherapy protocols between 1974 and 1994 and who subsequently developed recurrent disease before 2001. All patients were entered prospectively into an electronic database and were followed for evidence of disease recurrence. Survival data were updated as of April 2002. The median follow-up period for surviving patients after recurrence is 9.3 years. There have been 688 deaths among the 834 patients. The major study objective was to evaluate the association between the year that disease recurrence was detected and subsequent survival.
All patients with recurrent breast cancer had been treated prior to recurrence on one of five consecutive anthracycline-based adjuvant chemotherapy protocols that accrued patients from 1974 to 1994. Each protocol enrolled women with Stage II and III breast cancer; the most recent protocol's eligibility requirements also included women who had Stage I disease with adverse prognostic features (negative estrogen receptor status, lymphovascular invasion, high labeling index, or high nuclear grade). The details of each protocol have been published previously.10–15 Of the women who were included in the study, 116 patients were treated on the first protocol, 133 women were treated on the second adjuvant protocol, 165 women were treated on the third adjuvant protocol, 140 women were treated on the fourth protocol, and 281 women were treated on the fifth adjuvant protocol.
In all protocols, extensive pretreatment staging studies were done and were very similar across the trials. The evaluation included a complete physical examination, laboratory evaluation, chest X-ray, skeletal films, liver imaging, bone scan, and mammograms of the remaining and/or contralateral breast. In the first two protocols, liver imaging was done by nuclear scan; whereas, in the last three protocols, liver imaging could be done by either ultrasound or nuclear scan. Patient follow-up was specified in each protocol and was very similar across the five protocols. In the first 2 protocols, patients were followed at 4-month intervals with chest X-rays, liver scans, and bone scans. The first protocol also required skeletal X-rays every 4 months. The last 3 protocols followed patients with chest X-rays, bone scans, and liver ultrasound at 4-month intervals for the first 2 years, then every 6 months for 1 year, then yearly. In all five studies, a yearly mammogram of the remaining and contralateral breast was performed.
To determine whether breast cancer survival has improved over the past 25 years, patients initially were divided into 5 groups by year of recurrence. Recurrence was defined as including locally recurrent disease as well as distant metastases. From 1974 to 1979, 93 patients developed breast cancer recurrence; from 1980 to 1984, 216 patients had disease recurrence; from 1985 to 1989, 235 patients had disease recurrence; from 1990 to 1994, 185 patients had disease recurrence; and from 1995 to 2000, 106 patients had disease recurrence. Survival was measured from time of recurrence until death and was compared between the five groups. The dominant site of recurrence was defined as follows: patients were considered to have soft tissue–dominant disease if they had local recurrences to the skin, chest wall, lymph nodes, or breast (two patients with distant skin metastases also were included); patients were considered to have bone-dominant disease if they had metastases to the bone with or without soft tissue involvement; and patients were considered to have visceral-dominant disease if they had metastases to organs with or without bone or soft tissue involvement. Prognostic variables were then studied to determine the effect of each variable on survival from the time of recurrence. Because some variables with prognostic importance were distributed unevenly over the five groups, a multivariate analysis was performed to investigate the independent effect of time of recurrence on survival after adjusting for other prognostic factors. In particular, because all patients who developed recurrent disease between 1974 and 2000 were included, patients who were diagnosed with recurrent disease in the more recent years, by definition, could have longer disease-free intervals. We addressed this potential bias by using multivariate analysis to evaluate the association of survival with the year of recurrence while incorporating covariates for disease characteristics.
The chi-square test was used to investigate the independence between two categorical variables. Wilcoxon (or Kruskal–Wallis) rank sum tests were used to compare groups of independent continuous variables. Survival from time of recurrence to last follow-up was estimated using the Kaplan–Meier product-limit method. The two-sided log-rank test was used to test the association between patient variables and survival. Multivariate analysis was done using a Cox proportional hazards regression model to determine the association of time of recurrence with survival, after adjusting for the role of other factors.16 To investigate the association of recurrence date with survival, January 1, 1974, was treated as baseline, and the time difference between the recurrence date and January 1, 1974, was calculated as the recurrence interval instead of considering the actual recurrence date. To address the confounding disease-free interval, the analysis was stratified using 5 approximately equal groups based on the length of disease-free interval (≤ 17 months, 18–27 months, 28–40 months, 41–68 months, and > 68 months). All P values presented are two-sided. Disease-free interval was defined as the interval from the date of first treatment to the date of first disease recurrence. Survival was computed from the date of first disease recurrence to the date of death or last follow-up. Statistical analyses were carried out using SAS software (Version 8.02; SAS Institute, Cary, NC) and Splus 6 (Insightful; Seattle, WA).
Patients were divided into 5 groups according to their recurrence date: 1974–1979, 1980–1984, 1985–1989, 1990–1994, and 1995–2000. The distributions of patient characteristics for these patient groups are presented in Table 1. Patient characteristics were not distributed evenly across the five groups. The median ages of the groups ranged from 46 years to 51 years. Patients who developed recurrent disease in 1990–1994 were slightly younger than patients in the other groups. Menopausal status differed across groups, with the group of patients who developed recurrent disease from 1990 to 1994 having a higher percentage of patients who were premenopausal at diagnosis. However, no clear trends over time were seen for menopausal status of patients.
Clear time trends in the proportion of patients with unknown estrogen receptor status were observed. In the earlier groups, many patients did not have assays for estrogen receptor status performed: Ninety-one percent of patients who developed recurrent disease between 1974 and 1979 and 47% of patients who developed recurrent disease between 1980 and 1984 had unknown estrogen receptor status. In the more recent groups, the percentage of patients with unknown estrogen receptor status was more stable at 19%, 15%, and 13%, although the proportions of patients with unknown estrogen receptor status still were declining slowly. Among patients with known estrogen receptor status, no difference in the proportion of estrogen receptor–positive tumors was detected between the 5 cohorts (P = 0.61). This finding suggests that no real difference exists in the distribution of this marker over the years.
Initial tumor stage, tumor size, number of lymph nodes, disease-free interval, and dominant site of disease recurrence all differed among the groups, with patients who developed recurrent disease in more recent years having more favorable prognostic features. Table 1 demonstrates a trend over time for stage of disease; patients with recurrent disease in more recent years tended to have earlier stage tumors. This same trend was seen for tumor size and the number of lymph nodes involved, with patients who developed recurrent disease more recently having had smaller primary tumors and less lymph node involvement. The disease-free interval also differed significantly among the five groups, with the more recent groups tending to have had a longer disease-free interval. Finally, there were differences between the groups in dominant site of disease recurrence. Patients who developed recurrent disease in more recent years were more likely to have soft tissue disease compared with patients who developed recurrent disease in earlier years.
Survival from the time of recurrence was compared between the five groups. In this unadjusted analysis, survival was significantly longer in the more recent groups. The median survival was 15 months, 17 months, 22 months, 27, and 58 months in the groups who developed recurrent disease during 1974–1979, 1980–1984, 1985–1989, 1990–1994, and 1995–2000, respectively: The 5-year overall survival rates were 10%, 14% 22% 29%, and 44% across the five groups, respectively. However, as discussed above, the more recent groups were confounded by more favorable profiles of prognostic factors. Survival from the time of recurrence is shown in Figure 1. Survival curves stratified by disease-free interval are shown in Figure 2.
Because the distribution of patient characteristics differed across the years of the study, patient and tumor characteristics were evaluated to determine the role of potential risk factors on survival from time of recurrence. Table 2 summarizes the association of patient and tumor characteristics with survival. Patients with fewer lymph nodes involved, smaller tumors, and lower stage tumors at the time of primary diagnosis tended to have longer survival after disease recurrence. Menopausal status was not associated with survival after disease recurrence. Dominant site of disease recurrence was associated strongly with survival. Patients with soft tissue recurrences had an estimated 5-year survival of 41% compared with 23% for patients with bone metastases, and 13% for patients with visceral metastases (P < 0.01). Estrogen receptor status of the primary tumor also was associated with survival after recurrence: patients with initial tumors that were estrogen receptor positive had longer survival. The results presented in Table 2 were not adjusted for the year of disease recurrence.
Table 2. Association of Prognostic Factors with Survival
Total no. of patients
No. of deaths
Median survival (mos)
Five yr overall survival
Log-rank P value
Dominant site of disease recurrence
Size of primary tumor (cm)
No. of involved lymph nodes
Estrogen receptor status
To investigate the independent effect of each prognostic factor as well as year of recurrence on survival, multivariate models were fitted that considered dominant site of disease recurrence, primary tumor size, number of involved lymph nodes, disease-free interval, and year of recurrence. Stage was omitted due to its close correlation with tumor size and lymph node involvement. Because one-third of patients had unknown estrogen receptor status, and because the proportion of patients with unknown status was so high in the early years of the study, results for models that included this term are not reported (but were similar to the reported models).
Table 3 displays the results of the multivariate analysis. Hazard ratios < 1.0 indicate a protective effect, i.e., a lower risk of death, and hazard ratios > 1.0 indicate an increased risk of death. The analysis was stratified within 5 approximately equal-sized groups based on length of disease-free interval: < 17 months, 18–27 months, 28–40 months, 41–68 months, > 68 months. After adjusting for other variables, the number of lymph nodes involved, the size of the primary tumor, and the site of disease recurrence all were associated independently with survival from the time of recurrence. A dominant site of soft tissue recurrence, in contrast to bone or visceral disease, was associated with longer survival. A greater number of involved lymph nodes, visceral metastases, and primary tumors measuring > 2 cm were associated with shorter survival after recurrence. Results suggested that, in this group of patients, primary tumors measuring > 5 cm had little additional impact on survival once metastasis occurred. This finding is compatible with the single-factor results shown in Table 2.
Table 3. Multivariate Analysis of Factors Predicting Survival from Time of Recurrence
The primary objective of the current study was to determine whether survival for patients with recurrent breast cancer improved over the years 1974–2000. In the model summarized in Table 3, the hazard ratio associated with recurrence year was 0.99 (95% confidence interval, 0.97–1.0), suggesting a reduction of 1% in the risk of death with an increase of 1 year in the recurrence date. The statistical significance of including the term in the regression model did not achieve statistical significance at the 0.05 level (P = 0.09). Because of confounding between the length of disease-free interval and the date of recurrence, other modeling strategies were attempted, including considering disease-free interval as a continuous covariate in the model and use of alternate stratification levels. In an unstratified model that included disease-free interval as a covariate, the reduction in risk associated with a 1-year change in recurrence year was estimated at 2% (P = 0.04).
In the current data set, we have observed the effect of year of diagnosis of recurrent disease on overall survival for patients with recurrent disease diagnosed in the years 1974–2000. Patients who developed recurrent breast cancer in more recent years generally had improved survival. Using statistical adjustment techniques, we estimate that there has been a reduction in risk of approximately 1% for each increasing year, although the change was not highly statistically significant.
In addition, the current study has confirmed the importance of disease-free interval, the number of involved axillary lymph nodes, the site of disease recurrence, and estrogen receptor status as independent predictors of survival in patients with metastatic breast cancer in a large population of 834 patients. There have been very few studies of this size that have been able to adjust for the independent prognostic value of each factor. Clark et al. have reported on prognostic variables in patients with recurrent breast cancer.17 Our findings differ from those of Clark et al. in that the current study suggests that primary tumor size > 2 cm is associated with an increased risk of death after recurrence.
We recognize that the current study has limitations. Some degree of bias was present in the unadjusted analysis due to the weighting of more favorable prognostic factors in the groups that developed recurrent disease in more recent years. The preponderance of earlier stage disease, smaller tumors, lymph node–negative disease, and longer disease-free interval in the later groups are attributable partially to the restricted nature of the data set, in that only patients with disease diagnosed and who received adjuvant chemotherapy between the years 1974 and 1994 were included. Thus, patients with recurrent disease in the early years necessarily had short disease-free intervals and other factors associated with high risk of early recurrence, whereas patients with recurrent disease in the final year were guaranteed to have disease-free intervals of at least 5 years. The length of disease-free interval is known as an indicator of disease aggressiveness and subsequent survival. However, after adjusting for these variables in the multivariate analysis, a trend toward improving survival still was seen.
The changing distribution of patient characteristics also may be a reflection of changing standards of care for the adjuvant treatment of patients with breast cancer, because women with earlier stage breast cancer have been more likely to receive adjuvant chemotherapy in recent years. It also may reflect the increasing use of screening mammography and the resulting diagnosis at earlier stage of the disease. The advantage of using this study design is that we have captured a patient population with uniform adjuvant treatment and follow-up. We felt that this advantage outweighed the difficulties of unequal distribution of prognostic factors, because each prognostic variable could be adjusted for in a multivariate analysis. However, the reliability of statistical adjustment techniques is limited in the presence of confounded data; therefore, the estimate of risk reduction associated with recurrence year is uncertain. Results nevertheless are consistent with literature reports of an improved survival picture for patients with metastatic breast cancer.
The interpretation of the current study also is limited by the problems inherent in analyzing survival data over time. Stage migration and improved diagnostic testing leading to earlier ascertainment of metastases have made the determination of survival trends problematic. For instance, improvements in radiographic techniques and more precise imaging can change apparent stage distribution. Tiny metastatic lesions, which may have been detected with modern imaging in the 1990s, may have been missed in a patient who was diagnosed in the 1970s: Therefore, some patients who were diagnosed in the 1970s may have been misclassified with early-stage disease. This would result in an apparent poorer prognosis among patients diagnosed in the 1970s. In addition, the improvements in radiographic imaging may have resulted in earlier diagnosis of recurrent disease. This would result in lead-time bias, giving a misleading appearance of improved survival.
Although the possibility of bias due to stage migration and earlier ascertainment of recurrence cannot be ruled out, this study was designed to minimize these biases. We selected a group of patients who had been treated at a single institution with adjuvant, anthracycline-based protocols. All patients, regardless of the year of initial diagnosis or year of disease recurrence, had similar initial screening procedures and follow-up, minimizing ascertainment bias. In fact, the patients treated in the earlier years had slightly more intensive follow-up. This trend would result in a bias away from seeing any effect of year of recurrence on survival. In addition, during these same years, the efficacy of our adjuvant chemotherapy regimens did not change to any significant degree, because anthracyclines were a part of all of our regimens, and none of the patients in this study received adjuvant taxanes. Although some proportion of the reduced risk of death shown in this study may be due to stage migration and ascertainment bias, the design of this study, with uniform screening and follow-up of all patients, makes it unlikely that these biases were large contributors to the improvement in survival. The effect that we have demonstrated is suggestive of a true trend toward improved survival in patients with recurrent breast cancer. In addition, these findings are consistent with recently presented data. Chia et al. studied over 2000 women in a population-based analysis of survival for women with metastatic breast cancer who were diagnosed from 1991 to 2001.18 Those authors found an improvement in survival for women who developed metastatic disease in the most recent cohort, after adjusting for known prognostic factors. A large study of over 68,000 patients who were treated at 1 of 20 French Comprehensive Cancer Centers also demonstrated an improvement in survival from 1980 to 1999 among patients with metastatic disease.19 Our findings also are consistent with the national trend of a 3% annual decline in breast cancer mortality.1
Over the years, many new therapies have been approved that can palliate metastatic breast cancer. Figure 3 shows the number of new drugs approved by the United States Food and Drug Administration from the 1950s through the 1990s. A marked increase over time in the number of drugs available to treat breast cancer is evident. In particular, many effective new therapies were approved in the 1990s, including the taxanes, vinorelbine, capecitabine, Herceptin, aromatase inhibitors, and pamidronate. With the introduction of these new drugs over the past 30 years, postrecurrence treatment has changed. In the patient population in this study, there was no prespecified sequence of treatments after disease recurrence. However, the proliferation of new and active drugs in the 1990s, including cytotoxic agents and endocrine therapies, appears to have altered the survival of these patients in a favorable way.
In conclusion, the results of the current analysis suggest that the survival of women with recurrent breast cancer has been improving over the past several decades. The strengths of this study include a large patient population, extensive follow-up (median, 9.3 years), and uniform initial staging and follow-up testing. Given the inherent difficulty of determining changes in survival over time, we encourage interpreting these results with caution. These findings, although not conclusive, suggest that breast cancer survival has been improving. We present these data to encourage further research to clarify whether advances in therapy have translated into improvements in survival for women with recurrent breast cancer.