Postmastectomy Radiation and Mortality in Women With T1-2 Node-Positive Breast Cancer
http://www.100md.com
《临床肿瘤学》
the Department of Therapeutic Radiology, Yale University School of Medicine, New Haven, CT
ABSTRACT
PATIENTS AND METHODS: Using data from the Surveillance, Epidemiology, and End Results program, we identified 18,038 women with T1-2 node-positive invasive breast cancer who were treated with mastectomy between 1988 and 1995. The relationship between PMRT and mortality was determined using proportional hazards multivariate modeling and propensity score matched case-control analysis.
RESULTS: Median follow-up was 8.1 years. Only 2,648 women (15%) received PMRT. After adjusting for covariates, PMRT use was not associated with mortality (hazard ratio [HR] = 0.96; 95% CI, 0.90 to 1.03). However, the interaction term for PMRT use and number of involved regional lymph nodes was significant (P = .002), suggesting that, above a certain threshold of involved nodes, a mortality benefit from PMRT may exist. Adjusted analysis stratified by number of involved nodes revealed that patients with seven or more involved nodes treated with PMRT experienced a significant reduction in all-cause (HR = 0.84; 95% CI, 0.76 to 0.93) and cause-specific mortality (HR = 0.86; 95% CI, 0.77 to 0.96). Propensity score matched case-control analysis confirmed that PMRT was associated with reduced mortality only in the subset of patients with seven or more involved nodes (HR = 0.81; 95% CI, 0.73 to 0.91 for all-cause mortality; and HR = 0.82; 95% CI, 0.72 to 0.93 for cause-specific mortality).
CONCLUSION: For women with T1-2 breast cancer, PMRT is associated with a 15% to 20% relative reduction in mortality for patients with seven or more involved regional lymph nodes.
INTRODUCTION
In the late 1990s, three randomized trials concluded that, for women with stage II to III breast cancer, PMRT reduced the risk of locoregional recurrence from roughly 30% to 10% and produced an absolute survival benefit of 10% at 10 years.4-6 Some clinicians have used these trials to justify routine PMRT for all women with node-positive breast cancer. However, three large cohort studies found that women with primary tumors ≤ 5 cm and only one to three involved axillary nodes (T1-2 N1) experienced locoregional recurrence rates of only 6% to 13% after treatment with mastectomy and systemic chemotherapy.7-10 For such patients, the potential mortality benefit from PMRT would be minimal.
As a result, recent consensus statements from the National Institutes of Health, American Society of Clinical Oncology, and American Society for Therapeutic Radiology and Oncology have emphasized the need for further study of PMRT for T1-2 N1 breast cancer.10-12 Unfortunately, a recent intergroup trial designed to address this issue was closed prematurely because of poor patient accrual.13 For the foreseeable future, it is unlikely that any level I evidence on this subject will be generated.
In lieu of randomized data, retrospective review of high-quality, population-based data may provide insights into the relationship between PMRT and survival. In a cohort of women with T1-2 node-positive primary breast cancer reported by the Surveillance, Epidemiology, and End Results (SEER) program, we sought to explore the relationship between PMRT and survival and how this relationship varies with the number of involved regional lymph nodes.
PATIENTS AND METHODS
Radiation Use
Use of radiotherapy within the initial treatment course, typically defined as treatments received within 4 to 6 months of initial diagnosis, was abstracted by local tumor registries and reported to SEER. Only those patients coded as receiving external-beam irradiation or no irradiation were included in this study.
Outcomes
The primary end point was all-cause mortality, and the secondary end point was cause-specific mortality. Mortality and cause of death data were collected at the local SEER registry by matching patients to state vital statistics and, in some cases, were supplemented by data from the National Center for Health Statistics. Follow-up time was calculated from the month and year of initial diagnosis and is currently available through December 2000. Vital status and date of last contact were available for all patients.
Covariates
Our analyses were adjusted for age, race, SEER registry, year of diagnosis, primary tumor size, primary tumor laterality (left v right), primary tumor location (inner v outer or central), grade, estrogen receptor and progesterone receptor status, number of involved nodes, type of mastectomy, and histology coded according to the International Classification of Diseases for Oncology Second Edition.14 Cut points for categorical variables including age, tumor size, tumor location within the breast, percentage of positive nodes, and number of sampled nodes were chosen based on prior studies.8,9,15-17 Number of positive nodes was divided into groups of one, two, three, four, five to six, seven to nine, and 10 or more nodes to produce a comparable number of patients in each group.
Bivariate Statistics
Pearson’s {chi}2 test was used to determine unadjusted associations between PMRT use and categorical covariates. Unadjusted associations between PMRT use, covariates, and mortality were determined using the Kaplan-Meier log-rank test and Cox proportional hazards regression.
Proportional Hazards Model
Multivariate Cox proportional hazards regressions tested whether PMRT was an independent predictor of mortality and whether the effect sizes for associations were modified by number of involved nodes using an interaction term for PMRT and number of involved nodes. Given their linear relationship with mortality, the covariates of age, tumor size, number of involved nodes, and year of diagnosis were entered into the multivariate model as continuous variables. Covariates lacking linear relationships with mortality were entered as categorical (dummy) variables in multivariate analysis, with missing values as a dummy category.
Number of sampled nodes and percentage of involved regional nodes were not included in the multivariate model as a result of colinearity with the number of involved regional nodes. Instead, a subset analysis was performed on the subgroup of patients with ≥ 20% positive nodes based on a previous recursive partitioning analysis suggesting that this cutoff is the most significant predictor of locoregional recurrence.18
Propensity Score Modeling
The propensity score, defined as the probability of receiving radiation, was calculated for each patient using a multivariate logistic regression model adjusted for all relevant covariates.19 All models were evaluated using the Hosmer and Lemeshow goodness-of-fit test, where P > .05 indicates an appropriate model. Cases, who were defined as those patients receiving PMRT, were then matched to controls, who were defined as those patients not receiving PMRT, on the basis of the propensity score using a greedy 5->1 digit-matching algorithm.20 The distribution of covariates in cases and controls was compared using the paired t test for continuous variables, McNemar {chi}2 for categoric variables, and the Wilcoxon test for ordinal variables. The association of PMRT and mortality was determined using the Kaplan-Meier log-rank test and Cox proportional hazards regression.
All statistical analyses were two-tailed with an {alpha} = .05 and were conducted using SAS version 8.02 (SAS Institute, Cary, NC). This study was exempt from review by the Yale University School of Medicine Human Investigation Committee.
RESULTS
PMRT and Clinicopathologic Covariates
PMRT use was associated with adverse prognostic factors, such as large tumor size, ductal histology, high grade, multiple positive regional lymph nodes, and ≥ 20% positive nodes (Table 1). Of those patients with one to three involved nodes, only 8% received PMRT compared with 20% of patients with four to six involved nodes and 31% of patients with seven or more positive nodes (P < .0001).
PMRT and Survival: Proportional Hazards Model
With a median follow-up of 8.1 years for patients alive at last contact, unadjusted analysis showed that patients who received PMRT had increased risk of death, with a 10-year survival rate of 55% in the no radiation group versus 49% in the PMRT group (all-cause mortality: HR = 1.17; 95% CI, 1.10 to 1.25; P < .0001; cause-specific mortality: HR = 1.38; 95% CI, 1.28 to 1.49; P < .0001). Covariates associated with poor all-cause mortality included increased tumor size (HR = 1.24 per cm; 95% CI, 1.21 to 1.26; P < .0001) and increased number of involved nodes (HR = 1.055 per involved node; 95% CI, 1.051 to 1.058; P < .0001; Table 2).
After multivariate analysis, however, PMRT was no longer associated with all-cause mortality (HR = 0.97; 95% CI, 0.90 to 1.03; P = .3; Table 3) or cause-specific mortality (HR = 0.97; 95% CI, 0.90 to 1.05; P = .5). Furthermore, the interaction between PMRT use and number of involved regional lymph nodes was significant for all-cause mortality (P = .002) and for cause-specific mortality (P = .0005), which suggests that a mortality benefit from PMRT might be present at least in certain patients with more than a threshold number of involved nodes. When stratified by number of involved regional nodes, those patients with seven or more lymph nodes treated with PMRT experienced a significant reduction in mortality (all-cause mortality HR for seven or more involved nodes = 0.84; 95% CI, 0.76 to 0.93; P = .0005; cause-specific HR = 0.86; 95% CI, 0.77 to 0.96; P = .007; Fig 1).
In the subset of 7,572 patients with complete information regarding grade, estrogen receptor and progesterone receptor status, and primary tumor location, PMRT was again associated with reduced mortality only for those patients with seven or more positive nodes (HR = 0.77; 95% CI, 0.66 to 0.90; P = .0009). Finally, in a separate subset multivariate analysis conducted on the 8,637 patients with ≥ 20% involved nodes, PMRT was also associated with reduced mortality (HR = 0.90; 95% CI, 0.84 to 0.98; P = .009).
PMRT and Survival: Propensity Score Model
To confirm that the associations observed in proportional hazards multivariate modeling were not a result of residual confounding from covariates, a propensity score matched case-control analysis was conducted. First, the sample was divided into one of the following three subsets: one to three involved nodes, four to six involved nodes, and seven or more involved nodes. Second, the propensity score was calculated for each patient using multivariate logistic regression modeling of PMRT use (goodness of fit, P = .21 for the one to three involved nodes model, P = .30 for the four to six involved nodes model, and P = .09 for the seven or more involved nodes model). For each nodal strata, no significant differences in baseline covariates existed between cases and controls (Table 4).
For patients with seven or more involved regional nodes, 10-year overall survival rate improved from 34% in untreated controls to 42% in the PMRT group (HR = 0.81; 95% CI, 0.73 to 0.91; P = .0003; Table 5 and Fig 2). Similarly, 10-year cause-specific survival improved from 44% in the control group to 53% in the PMRT group (HR = 0.82; 95% CI, 0.72 to 0.93; P = .002). PMRT use was not associated with reduced mortality for patients with one to six involved nodes (Table 5 and Fig 2).
DISCUSSION
Consistent with our findings, previous reports have shown that the risk of isolated locoregional recurrence also increases with number of involved nodes. For example, a study of 2,016 women treated on Eastern Cooperative Oncology Group trials with mastectomy and cyclophosphamide, methotrexate, and fluorouracil chemotherapy without PMRT found that, for women with primary tumors less than 5 cm, the risk of isolated locoregional recurrence was 8% for one to three involved nodes, 15% for four to seven involved nodes, and 20% for eight or more involved nodes.9 Thus, the threshold identified in this study corresponds to a risk for isolated locoregional recurrence approaching 20%.
This study also highlights differences in axillary sampling between existing randomized data and standard practice in the United States. For example, roughly 75% of the patients in the Danish 82b and 82c PMRT trials had less than 10 axillary lymph nodes sampled, compared with only 14% in our study.4,6 Previous observational studies have shown that sampling fewer than 10 axillary nodes more than doubles the risk of subsequent locoregional failure.7-9 Thus, although the Danish trials concluded that all node-positive women experience a substantial mortality benefit from PMRT, these results may not be directly applicable to patients in the United States whose risk for locoregional recurrence is reduced by more extensive axillary sampling.4,6 To account for the issue of axillary sampling, some investigators have proposed that percentage of involved axillary nodes may be better than number of involved axillary nodes in identifying those patients at substantial risk for locoregional recurrence.18 Although not designed to compare the prognostic significance of these two variables, our study does confirm that patients with ≥ 20% involved nodes, a previously reported threshold, experience a substantial mortality benefit from PMRT.
In 1988, a seminal patterns of failure analysis recommended routine PMRT for patients with four or more involved axillary nodes, given their relatively high risk of local-regional recurrence.21 Nevertheless, only 26% of such patients in our study received PMRT, suggesting that PMRT was surprisingly underused between 1988 and 1995. Future studies are clearly indicated to determine whether or not utilization rates of PMRT have increased to acceptable norms after publication of the three PMRT randomized trials between 1997 and 1999.4-6
One important limitation of this study is possible underreporting of PMRT use. Previous studies have indicated that the radiotherapy variable in the SEER registry is 72% to 94% sensitive and 99% specific.22-26 Underreporting of PMRT, however, would have actually biased results toward the null hypothesis, which is unlikely in our analysis given the significance of the association identified between PMRT and survival. This data set also does not include certain pathologic factors, such as margin status and lymph-vascular invasion, suggesting that the effect size of PMRT may need to be refined in further studies.
When interpreting observational data, appropriate adjustment for potential confounders is mandatory to determine the effect of an intervention. As suggested by Rubin,19 proportional hazards multivariate modeling may produce errant results when there is little overlap between the intervention and untreated groups. To address this potential bias, we conducted a case-control analysis in which PMRT cases were matched to untreated controls using the propensity score, which represents a composite variable that describes all measured confounders. Both the multivariate regression model and matched case-control model produced similar results, suggesting that measured confounders did not account for the observed treatment effect in patients with seven or more involved nodes. However, unmeasured confounders, most importantly comorbid illness, cannot be accounted for in the SEER data and, thus, may result in selection bias. Nevertheless, the strong association between PMRT and cause-specific survival observed in this study implies that PMRT improves overall survival by reducing breast cancer deaths and, thus, suggests that selection bias caused by comorbid illness does not explain the observed relationship between PMRT and survival.
In summary, this study provides additional evidence that, in the community setting, PMRT produces a substantial reduction in mortality for patients at high risk for locoregional recurrence. However, for patients with one to three involved nodes, no mortality benefit from PMRT was noted, and routine use of PMRT is not supported. The decision to offer PMRT for such patients must be individualized based on clinicopathologic risk factors and patient preference.
Authors’ Disclosures of Potential Conflicts of Interest
NOTES
Supported by National Institutes of Health/National Institute of General Medical Sciences Medical Scientist Training Grant GM07205 (G.L.S.).
Presented in part at the 86th Annual Meeting of the American Radium Society, Napa, CA, May 2, 2004.
Authors’ disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Surveillance, Epidemiology, and End Results: SEER public-use data (1973-2000). http://www.seer.cancer.gov/publicdata/
2. Surveillance, Epidemiology, and End Results: SEER*Stat 5. www.seer.cancer.gov/seerstat
3. Jemal A, Tiwari RC, Murray T, et al: Cancer statistics, 2004. CA Cancer J Clin 54:8-29, 2004
4. Overgaard M, Hansen PS, Overgaard J, et al: Postoperative radiotherapy in high-risk premenopausal women with breast cancer who receive adjuvant chemotherapy: Danish Breast Cancer Cooperative Group 82b Trial. N Engl J Med 337:949-955, 1997
5. Ragaz J, Jackson SM, Le N, et al: Adjuvant radiotherapy and chemotherapy in node-positive premenopausal women with breast cancer. N Engl J Med 337:956-962, 1997
6. Overgaard M, Jensen MB, Overgaard J, et al: Postoperative radiotherapy in high-risk postmenopausal breast-cancer patients given adjuvant tamoxifen: Danish Breast Cancer Cooperative Group DBCG 82c randomised trial. Lancet 353:1641-1648, 1999
7. Katz A, Strom EA, Buchholz TA, et al: Locoregional recurrence patterns after mastectomy and doxorubicin-based chemotherapy: Implications for postoperative irradiation. J Clin Oncol 18:2817-2827, 2000
8. Woodward WA, Strom EA, Tucker SL, et al: Locoregional recurrence after doxorubicin-based chemotherapy and postmastectomy: Implications for breast cancer patients with early-stage disease and predictors for recurrence after postmastectomy radiation. Int J Radiat Oncol Biol Phys 57:336-344, 2003
9. Recht A, Gray R, Davidson NE, et al: Locoregional failure 10 years after mastectomy and adjuvant chemotherapy with or without tamoxifen without irradiation: Experience of the Eastern Cooperative Oncology Group. J Clin Oncol 17:1689-1700, 1999
10. Taghian AG, Jeong JH, Anderson S, et al: Pattern of loco-regional and distant failure in patients with breast cancer treated with mastectomy and chemotherapy (+/- tamoxifen) without radiation: Results from five NSABP randomized trials. Int J Radiat Oncol Biol Phys 51:106a-107a, 2001 (suppl 3S1)
11. Recht A, Edge SB, Solin LJ, et al: Postmastectomy radiotherapy: Clinical practice guidelines of the American Society of Clinical Oncology. J Clin Oncol 19:1539-1569, 2001
12. Harris JR, Halpin-Murphy P, McNeese M, et al: Consensus Statement on postmastectomy radiation therapy. Int J Radiat Oncol Biol Phys 44:989-990, 1999
13. Buchholz TA, Strom EA, Perkins GH, et al: Controversies regarding the use of radiation after mastectomy in breast cancer. Oncologist 7:539-546, 2002
14. Percy C, Van Holten V, Muir C: International Classification of Diseases for Oncology (ed 2). Geneva, Switzerland, WHO, 1990
15. Early Breast Cancer Trialists’ Collaborative Group: Polychemotherapy for early breast cancer: An overview of the randomised trials—Early Breast Cancer Trialists’ Collaborative Group. Lancet 352:930-942, 1998
16. Katz A, Strom EA, Buchholz TA, et al: The influence of pathologic tumor characteristics on locoregional recurrence rates following mastectomy. Int J Radiat Oncol Biol Phys 50:735-742, 2001
17. Gaffney DK, Tsodikov A, Wiggins CL: Diminished survival in patients with inner versus outer quadrant breast cancers. J Clin Oncol 21:467-472, 2003
18. Katz A, Buchholz TA, Thames H, et al: Recursive partitioning analysis of locoregional recurrence patterns following mastectomy: Implications for adjuvant irradiation. Int J Radiat Oncol Biol Phys 50:397-403, 2001
19. Rubin DB: Estimating causal effects from large data sets using propensity scores. Ann Intern Med 127:757-763, 1997
20. Parsons LS: Reducing bias in a propensity score matched-pair sample using greedy matching techniques. Proc Ann SAS Users Group Int Conference 214-226, 2001
21. Fowble B, Gray R, Gilchrist K, et al: Identification of a subgroup of patients with breast cancer and histologically positive axillary nodes receiving adjuvant chemotherapy who may benefit from postoperative radiotherapy. J Clin Oncol 6:1107-1117, 1988
22. Malin JL, Kahn KL, Adams J, et al: Validity of cancer registry data for measuring the quality of breast cancer care. J Natl Cancer Inst 94:835-844, 2002
23. Virnig BA, Warren JL, Cooper GS, et al: Studying radiation therapy using SEER-Medicare-linked data. Med Care 40:IV-49-IV-54, 2002 (suppl 8)
24. Du X, Freeman JL, Goodwin JS: Information on radiation treatment in patients with breast cancer: The advantages of the linked Medicare and SEER data—Surveillance, Epidemiology and End Results. J Clin Epidemiol 52:463-470, 1999
25. Ballard-Barbash R, Potosky AL, Harlan LC, et al: Factors associated with surgical and radiation therapy for early stage breast cancer in older women. J Natl Cancer Inst 88:716-726, 1996
26. Brown ML, Hankey BF, Ballard-Barbash R: Measuring the quality of breast cancer care. Ann Intern Med 133:920, 2000(Benjamin D. Smith, Grace )
ABSTRACT
PATIENTS AND METHODS: Using data from the Surveillance, Epidemiology, and End Results program, we identified 18,038 women with T1-2 node-positive invasive breast cancer who were treated with mastectomy between 1988 and 1995. The relationship between PMRT and mortality was determined using proportional hazards multivariate modeling and propensity score matched case-control analysis.
RESULTS: Median follow-up was 8.1 years. Only 2,648 women (15%) received PMRT. After adjusting for covariates, PMRT use was not associated with mortality (hazard ratio [HR] = 0.96; 95% CI, 0.90 to 1.03). However, the interaction term for PMRT use and number of involved regional lymph nodes was significant (P = .002), suggesting that, above a certain threshold of involved nodes, a mortality benefit from PMRT may exist. Adjusted analysis stratified by number of involved nodes revealed that patients with seven or more involved nodes treated with PMRT experienced a significant reduction in all-cause (HR = 0.84; 95% CI, 0.76 to 0.93) and cause-specific mortality (HR = 0.86; 95% CI, 0.77 to 0.96). Propensity score matched case-control analysis confirmed that PMRT was associated with reduced mortality only in the subset of patients with seven or more involved nodes (HR = 0.81; 95% CI, 0.73 to 0.91 for all-cause mortality; and HR = 0.82; 95% CI, 0.72 to 0.93 for cause-specific mortality).
CONCLUSION: For women with T1-2 breast cancer, PMRT is associated with a 15% to 20% relative reduction in mortality for patients with seven or more involved regional lymph nodes.
INTRODUCTION
In the late 1990s, three randomized trials concluded that, for women with stage II to III breast cancer, PMRT reduced the risk of locoregional recurrence from roughly 30% to 10% and produced an absolute survival benefit of 10% at 10 years.4-6 Some clinicians have used these trials to justify routine PMRT for all women with node-positive breast cancer. However, three large cohort studies found that women with primary tumors ≤ 5 cm and only one to three involved axillary nodes (T1-2 N1) experienced locoregional recurrence rates of only 6% to 13% after treatment with mastectomy and systemic chemotherapy.7-10 For such patients, the potential mortality benefit from PMRT would be minimal.
As a result, recent consensus statements from the National Institutes of Health, American Society of Clinical Oncology, and American Society for Therapeutic Radiology and Oncology have emphasized the need for further study of PMRT for T1-2 N1 breast cancer.10-12 Unfortunately, a recent intergroup trial designed to address this issue was closed prematurely because of poor patient accrual.13 For the foreseeable future, it is unlikely that any level I evidence on this subject will be generated.
In lieu of randomized data, retrospective review of high-quality, population-based data may provide insights into the relationship between PMRT and survival. In a cohort of women with T1-2 node-positive primary breast cancer reported by the Surveillance, Epidemiology, and End Results (SEER) program, we sought to explore the relationship between PMRT and survival and how this relationship varies with the number of involved regional lymph nodes.
PATIENTS AND METHODS
Radiation Use
Use of radiotherapy within the initial treatment course, typically defined as treatments received within 4 to 6 months of initial diagnosis, was abstracted by local tumor registries and reported to SEER. Only those patients coded as receiving external-beam irradiation or no irradiation were included in this study.
Outcomes
The primary end point was all-cause mortality, and the secondary end point was cause-specific mortality. Mortality and cause of death data were collected at the local SEER registry by matching patients to state vital statistics and, in some cases, were supplemented by data from the National Center for Health Statistics. Follow-up time was calculated from the month and year of initial diagnosis and is currently available through December 2000. Vital status and date of last contact were available for all patients.
Covariates
Our analyses were adjusted for age, race, SEER registry, year of diagnosis, primary tumor size, primary tumor laterality (left v right), primary tumor location (inner v outer or central), grade, estrogen receptor and progesterone receptor status, number of involved nodes, type of mastectomy, and histology coded according to the International Classification of Diseases for Oncology Second Edition.14 Cut points for categorical variables including age, tumor size, tumor location within the breast, percentage of positive nodes, and number of sampled nodes were chosen based on prior studies.8,9,15-17 Number of positive nodes was divided into groups of one, two, three, four, five to six, seven to nine, and 10 or more nodes to produce a comparable number of patients in each group.
Bivariate Statistics
Pearson’s {chi}2 test was used to determine unadjusted associations between PMRT use and categorical covariates. Unadjusted associations between PMRT use, covariates, and mortality were determined using the Kaplan-Meier log-rank test and Cox proportional hazards regression.
Proportional Hazards Model
Multivariate Cox proportional hazards regressions tested whether PMRT was an independent predictor of mortality and whether the effect sizes for associations were modified by number of involved nodes using an interaction term for PMRT and number of involved nodes. Given their linear relationship with mortality, the covariates of age, tumor size, number of involved nodes, and year of diagnosis were entered into the multivariate model as continuous variables. Covariates lacking linear relationships with mortality were entered as categorical (dummy) variables in multivariate analysis, with missing values as a dummy category.
Number of sampled nodes and percentage of involved regional nodes were not included in the multivariate model as a result of colinearity with the number of involved regional nodes. Instead, a subset analysis was performed on the subgroup of patients with ≥ 20% positive nodes based on a previous recursive partitioning analysis suggesting that this cutoff is the most significant predictor of locoregional recurrence.18
Propensity Score Modeling
The propensity score, defined as the probability of receiving radiation, was calculated for each patient using a multivariate logistic regression model adjusted for all relevant covariates.19 All models were evaluated using the Hosmer and Lemeshow goodness-of-fit test, where P > .05 indicates an appropriate model. Cases, who were defined as those patients receiving PMRT, were then matched to controls, who were defined as those patients not receiving PMRT, on the basis of the propensity score using a greedy 5->1 digit-matching algorithm.20 The distribution of covariates in cases and controls was compared using the paired t test for continuous variables, McNemar {chi}2 for categoric variables, and the Wilcoxon test for ordinal variables. The association of PMRT and mortality was determined using the Kaplan-Meier log-rank test and Cox proportional hazards regression.
All statistical analyses were two-tailed with an {alpha} = .05 and were conducted using SAS version 8.02 (SAS Institute, Cary, NC). This study was exempt from review by the Yale University School of Medicine Human Investigation Committee.
RESULTS
PMRT and Clinicopathologic Covariates
PMRT use was associated with adverse prognostic factors, such as large tumor size, ductal histology, high grade, multiple positive regional lymph nodes, and ≥ 20% positive nodes (Table 1). Of those patients with one to three involved nodes, only 8% received PMRT compared with 20% of patients with four to six involved nodes and 31% of patients with seven or more positive nodes (P < .0001).
PMRT and Survival: Proportional Hazards Model
With a median follow-up of 8.1 years for patients alive at last contact, unadjusted analysis showed that patients who received PMRT had increased risk of death, with a 10-year survival rate of 55% in the no radiation group versus 49% in the PMRT group (all-cause mortality: HR = 1.17; 95% CI, 1.10 to 1.25; P < .0001; cause-specific mortality: HR = 1.38; 95% CI, 1.28 to 1.49; P < .0001). Covariates associated with poor all-cause mortality included increased tumor size (HR = 1.24 per cm; 95% CI, 1.21 to 1.26; P < .0001) and increased number of involved nodes (HR = 1.055 per involved node; 95% CI, 1.051 to 1.058; P < .0001; Table 2).
After multivariate analysis, however, PMRT was no longer associated with all-cause mortality (HR = 0.97; 95% CI, 0.90 to 1.03; P = .3; Table 3) or cause-specific mortality (HR = 0.97; 95% CI, 0.90 to 1.05; P = .5). Furthermore, the interaction between PMRT use and number of involved regional lymph nodes was significant for all-cause mortality (P = .002) and for cause-specific mortality (P = .0005), which suggests that a mortality benefit from PMRT might be present at least in certain patients with more than a threshold number of involved nodes. When stratified by number of involved regional nodes, those patients with seven or more lymph nodes treated with PMRT experienced a significant reduction in mortality (all-cause mortality HR for seven or more involved nodes = 0.84; 95% CI, 0.76 to 0.93; P = .0005; cause-specific HR = 0.86; 95% CI, 0.77 to 0.96; P = .007; Fig 1).
In the subset of 7,572 patients with complete information regarding grade, estrogen receptor and progesterone receptor status, and primary tumor location, PMRT was again associated with reduced mortality only for those patients with seven or more positive nodes (HR = 0.77; 95% CI, 0.66 to 0.90; P = .0009). Finally, in a separate subset multivariate analysis conducted on the 8,637 patients with ≥ 20% involved nodes, PMRT was also associated with reduced mortality (HR = 0.90; 95% CI, 0.84 to 0.98; P = .009).
PMRT and Survival: Propensity Score Model
To confirm that the associations observed in proportional hazards multivariate modeling were not a result of residual confounding from covariates, a propensity score matched case-control analysis was conducted. First, the sample was divided into one of the following three subsets: one to three involved nodes, four to six involved nodes, and seven or more involved nodes. Second, the propensity score was calculated for each patient using multivariate logistic regression modeling of PMRT use (goodness of fit, P = .21 for the one to three involved nodes model, P = .30 for the four to six involved nodes model, and P = .09 for the seven or more involved nodes model). For each nodal strata, no significant differences in baseline covariates existed between cases and controls (Table 4).
For patients with seven or more involved regional nodes, 10-year overall survival rate improved from 34% in untreated controls to 42% in the PMRT group (HR = 0.81; 95% CI, 0.73 to 0.91; P = .0003; Table 5 and Fig 2). Similarly, 10-year cause-specific survival improved from 44% in the control group to 53% in the PMRT group (HR = 0.82; 95% CI, 0.72 to 0.93; P = .002). PMRT use was not associated with reduced mortality for patients with one to six involved nodes (Table 5 and Fig 2).
DISCUSSION
Consistent with our findings, previous reports have shown that the risk of isolated locoregional recurrence also increases with number of involved nodes. For example, a study of 2,016 women treated on Eastern Cooperative Oncology Group trials with mastectomy and cyclophosphamide, methotrexate, and fluorouracil chemotherapy without PMRT found that, for women with primary tumors less than 5 cm, the risk of isolated locoregional recurrence was 8% for one to three involved nodes, 15% for four to seven involved nodes, and 20% for eight or more involved nodes.9 Thus, the threshold identified in this study corresponds to a risk for isolated locoregional recurrence approaching 20%.
This study also highlights differences in axillary sampling between existing randomized data and standard practice in the United States. For example, roughly 75% of the patients in the Danish 82b and 82c PMRT trials had less than 10 axillary lymph nodes sampled, compared with only 14% in our study.4,6 Previous observational studies have shown that sampling fewer than 10 axillary nodes more than doubles the risk of subsequent locoregional failure.7-9 Thus, although the Danish trials concluded that all node-positive women experience a substantial mortality benefit from PMRT, these results may not be directly applicable to patients in the United States whose risk for locoregional recurrence is reduced by more extensive axillary sampling.4,6 To account for the issue of axillary sampling, some investigators have proposed that percentage of involved axillary nodes may be better than number of involved axillary nodes in identifying those patients at substantial risk for locoregional recurrence.18 Although not designed to compare the prognostic significance of these two variables, our study does confirm that patients with ≥ 20% involved nodes, a previously reported threshold, experience a substantial mortality benefit from PMRT.
In 1988, a seminal patterns of failure analysis recommended routine PMRT for patients with four or more involved axillary nodes, given their relatively high risk of local-regional recurrence.21 Nevertheless, only 26% of such patients in our study received PMRT, suggesting that PMRT was surprisingly underused between 1988 and 1995. Future studies are clearly indicated to determine whether or not utilization rates of PMRT have increased to acceptable norms after publication of the three PMRT randomized trials between 1997 and 1999.4-6
One important limitation of this study is possible underreporting of PMRT use. Previous studies have indicated that the radiotherapy variable in the SEER registry is 72% to 94% sensitive and 99% specific.22-26 Underreporting of PMRT, however, would have actually biased results toward the null hypothesis, which is unlikely in our analysis given the significance of the association identified between PMRT and survival. This data set also does not include certain pathologic factors, such as margin status and lymph-vascular invasion, suggesting that the effect size of PMRT may need to be refined in further studies.
When interpreting observational data, appropriate adjustment for potential confounders is mandatory to determine the effect of an intervention. As suggested by Rubin,19 proportional hazards multivariate modeling may produce errant results when there is little overlap between the intervention and untreated groups. To address this potential bias, we conducted a case-control analysis in which PMRT cases were matched to untreated controls using the propensity score, which represents a composite variable that describes all measured confounders. Both the multivariate regression model and matched case-control model produced similar results, suggesting that measured confounders did not account for the observed treatment effect in patients with seven or more involved nodes. However, unmeasured confounders, most importantly comorbid illness, cannot be accounted for in the SEER data and, thus, may result in selection bias. Nevertheless, the strong association between PMRT and cause-specific survival observed in this study implies that PMRT improves overall survival by reducing breast cancer deaths and, thus, suggests that selection bias caused by comorbid illness does not explain the observed relationship between PMRT and survival.
In summary, this study provides additional evidence that, in the community setting, PMRT produces a substantial reduction in mortality for patients at high risk for locoregional recurrence. However, for patients with one to three involved nodes, no mortality benefit from PMRT was noted, and routine use of PMRT is not supported. The decision to offer PMRT for such patients must be individualized based on clinicopathologic risk factors and patient preference.
Authors’ Disclosures of Potential Conflicts of Interest
NOTES
Supported by National Institutes of Health/National Institute of General Medical Sciences Medical Scientist Training Grant GM07205 (G.L.S.).
Presented in part at the 86th Annual Meeting of the American Radium Society, Napa, CA, May 2, 2004.
Authors’ disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
1. Surveillance, Epidemiology, and End Results: SEER public-use data (1973-2000). http://www.seer.cancer.gov/publicdata/
2. Surveillance, Epidemiology, and End Results: SEER*Stat 5. www.seer.cancer.gov/seerstat
3. Jemal A, Tiwari RC, Murray T, et al: Cancer statistics, 2004. CA Cancer J Clin 54:8-29, 2004
4. Overgaard M, Hansen PS, Overgaard J, et al: Postoperative radiotherapy in high-risk premenopausal women with breast cancer who receive adjuvant chemotherapy: Danish Breast Cancer Cooperative Group 82b Trial. N Engl J Med 337:949-955, 1997
5. Ragaz J, Jackson SM, Le N, et al: Adjuvant radiotherapy and chemotherapy in node-positive premenopausal women with breast cancer. N Engl J Med 337:956-962, 1997
6. Overgaard M, Jensen MB, Overgaard J, et al: Postoperative radiotherapy in high-risk postmenopausal breast-cancer patients given adjuvant tamoxifen: Danish Breast Cancer Cooperative Group DBCG 82c randomised trial. Lancet 353:1641-1648, 1999
7. Katz A, Strom EA, Buchholz TA, et al: Locoregional recurrence patterns after mastectomy and doxorubicin-based chemotherapy: Implications for postoperative irradiation. J Clin Oncol 18:2817-2827, 2000
8. Woodward WA, Strom EA, Tucker SL, et al: Locoregional recurrence after doxorubicin-based chemotherapy and postmastectomy: Implications for breast cancer patients with early-stage disease and predictors for recurrence after postmastectomy radiation. Int J Radiat Oncol Biol Phys 57:336-344, 2003
9. Recht A, Gray R, Davidson NE, et al: Locoregional failure 10 years after mastectomy and adjuvant chemotherapy with or without tamoxifen without irradiation: Experience of the Eastern Cooperative Oncology Group. J Clin Oncol 17:1689-1700, 1999
10. Taghian AG, Jeong JH, Anderson S, et al: Pattern of loco-regional and distant failure in patients with breast cancer treated with mastectomy and chemotherapy (+/- tamoxifen) without radiation: Results from five NSABP randomized trials. Int J Radiat Oncol Biol Phys 51:106a-107a, 2001 (suppl 3S1)
11. Recht A, Edge SB, Solin LJ, et al: Postmastectomy radiotherapy: Clinical practice guidelines of the American Society of Clinical Oncology. J Clin Oncol 19:1539-1569, 2001
12. Harris JR, Halpin-Murphy P, McNeese M, et al: Consensus Statement on postmastectomy radiation therapy. Int J Radiat Oncol Biol Phys 44:989-990, 1999
13. Buchholz TA, Strom EA, Perkins GH, et al: Controversies regarding the use of radiation after mastectomy in breast cancer. Oncologist 7:539-546, 2002
14. Percy C, Van Holten V, Muir C: International Classification of Diseases for Oncology (ed 2). Geneva, Switzerland, WHO, 1990
15. Early Breast Cancer Trialists’ Collaborative Group: Polychemotherapy for early breast cancer: An overview of the randomised trials—Early Breast Cancer Trialists’ Collaborative Group. Lancet 352:930-942, 1998
16. Katz A, Strom EA, Buchholz TA, et al: The influence of pathologic tumor characteristics on locoregional recurrence rates following mastectomy. Int J Radiat Oncol Biol Phys 50:735-742, 2001
17. Gaffney DK, Tsodikov A, Wiggins CL: Diminished survival in patients with inner versus outer quadrant breast cancers. J Clin Oncol 21:467-472, 2003
18. Katz A, Buchholz TA, Thames H, et al: Recursive partitioning analysis of locoregional recurrence patterns following mastectomy: Implications for adjuvant irradiation. Int J Radiat Oncol Biol Phys 50:397-403, 2001
19. Rubin DB: Estimating causal effects from large data sets using propensity scores. Ann Intern Med 127:757-763, 1997
20. Parsons LS: Reducing bias in a propensity score matched-pair sample using greedy matching techniques. Proc Ann SAS Users Group Int Conference 214-226, 2001
21. Fowble B, Gray R, Gilchrist K, et al: Identification of a subgroup of patients with breast cancer and histologically positive axillary nodes receiving adjuvant chemotherapy who may benefit from postoperative radiotherapy. J Clin Oncol 6:1107-1117, 1988
22. Malin JL, Kahn KL, Adams J, et al: Validity of cancer registry data for measuring the quality of breast cancer care. J Natl Cancer Inst 94:835-844, 2002
23. Virnig BA, Warren JL, Cooper GS, et al: Studying radiation therapy using SEER-Medicare-linked data. Med Care 40:IV-49-IV-54, 2002 (suppl 8)
24. Du X, Freeman JL, Goodwin JS: Information on radiation treatment in patients with breast cancer: The advantages of the linked Medicare and SEER data—Surveillance, Epidemiology and End Results. J Clin Epidemiol 52:463-470, 1999
25. Ballard-Barbash R, Potosky AL, Harlan LC, et al: Factors associated with surgical and radiation therapy for early stage breast cancer in older women. J Natl Cancer Inst 88:716-726, 1996
26. Brown ML, Hankey BF, Ballard-Barbash R: Measuring the quality of breast cancer care. Ann Intern Med 133:920, 2000(Benjamin D. Smith, Grace )