Postmastectomy radiation therapy for lymph node-negative, locally advanced breast cancer after modified radical mastectomy

Analysis of the NCI Surveillance, Epidemiology, and End Results database




The role of postmastectomy radiotherapy (PMRT) for lymph node-negative locally advanced breast carcinoma (T3N0M0) after modified radical mastectomy (MRM) with regard to improvement in survival remains an area of controversy.


The 1973–2004 National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) database was examined for patients with T3N0M0 ductal, lobular, or mixed ductal and lobular carcinoma of the breast who underwent MRM, treated from 1988-2003. Patients who were men, who had positive lymph nodes, who survived ≤6 months, for whom breast cancer was not the first malignancy, who had nonbeam radiation, intraoperative or preoperative radiation were excluded. The average treatment effect of PMRT on mortality was estimated with a propensity score case-matched analysis.


In all, 1777 patients were identified; 568 (32%) patients received PMRT. Median tumor size was 6.3 cm. The median number of lymph nodes examined was 14 (range, 1–49). Propensity score matched case-control analysis showed no improvement in overall survival with the delivery of PMRT in this group. Older patients, patients with ER− disease (compared with ER+), and patients with high-grade tumors (compared with well differentiated) had increased mortality.


The use of PMRT for T3N0M0 breast carcinoma after MRM is not associated with an increase in overall survival. It was not possible to analyze local control in this study given the limitations of the SEER database. The impact of potential improvement in local control as it relates to overall survival should be the subject of further investigation. Cancer 2008. © 2008 American Cancer Society.

T3N0M0 breast cancer represents a rare disease entity, occurring in approximately 2% of all breast cancers.1 Postmastectomy radiation therapy (PMRT) for locally advanced and lymph node-positive breast cancer has been studied in large randomized trials2, 3 and has been found to definitively improve locoregional control and overall survival. However, the impact on survival of PMRT after modified radical mastectomy with lymph node dissection for locally advanced breast cancer that is node-negative (T3N0M0 [T3: tumor >5 cm in greatest dimension, N0: no regional lymph node metastases, M0: no distant metastases]) remains an area of controversy.

The large Danish 82b2 and 82c3 randomized trials, which together randomized over 3000 patients with stage II or III breast cancer to adjuvant chemotherapy or tamoxifen with or without PMRT, revealed a significantly improved locoregional control rate and overall survival rate. The Danish 82b trial noted an improvement in overall survival at 10 years (54% versus 45%, P <.001) in premenopausal women with high-risk disease who had received CMF (cyclophosphamide, methotrexate, fluorouracil) chemotherapy, whereas the Danish 82c trial noted a somewhat similar improvement in postmenopausal women with high-risk disease who had received tamoxifen (TAM) (45% vs 36%, P = .03).

When the subsets of patients with lymph node-negative disease were examined a decreased incidence of locoregional failure persisted with the addition of PMRT. Although overall survival appeared similar for postmenopausal patients (82c, 10-year survival of 56% for radiotherapy [RT] + TAM vs 55% for TAM alone), there was a significant benefit in overall survival for premenopausal patients (82b, 10-year survival of 82% for RT + CMF vs 70% for CMF alone). These studies, however, were criticized for their relatively low number of lymph nodes examined (median of 7), which increased the benefit of chest wall and regional lymph node radiation, as the extent of nodal dissection has been shown to be relevant to both local recurrence risk and overall survival.4 Helinto et al.1 also found an overall survival benefit (P = .03) and locoregional control benefit (P <.001) to the addition of PMRT for pT3N0M0 cancers. Jagsi et al.5 found an increased locoregional recurrence rate after mastectomy associated with tumors >2 cm versus ≤2 cm, supporting the possible role of adjuvant RT in larger tumors. With lymphovascular invasion and premenopausal status, a tumor >2 cm in size had a 40.6% chance of recurring postmastectomy.5

However, in a multiinstitutional retrospective review of patients with T3N0 disease who did not undergo PMRT, Floyd et al.6 found a low isolated locoregional failure rate overall in patients with 5 cm or greater tumor size and node-negative disease (7.6% at 5 years), an increased risk for isolated locoregional failure (21% at 5 years) for tumors with lymphatic vessel invasion, and that disease-free survival (defined as freedom from isolated locoregional, distant, simultaneous locoregional and distant recurrence, or death from disease) was 87% at 5 years, and overall survival was 83% at 5 years. Finally, a review of the National Surgical Adjuvant Breast and Bowel Project (NSABP) randomized trials7 reviewing 313 women with T3N0M0 disease found an isolated locoregional recurrence rate of only 7%, and the authors recommended that PMRT not be used routinely in these patients. Notably, there were an additional 10% of patients who developed locoregional failure simultaneous with distant failure at 10 years, suggesting that with better systemic therapy (and elimination of micrometastasis) that isolated local failure could be as high as 17% for T3N0 patients at 10 years.

In spite of this controversy, the National Comprehensive Cancer Network (NCCN) guidelines favor radiation for patients with tumors greater than 5 cm with negative lymph nodes.8 Given the lack of randomized trial evidence and relatively small number of patients with T3N0M0 disease, we investigated the National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) database. The SEER database is useful for investigating the effect of demographics, stage, surgery type, and radiation use in rare tumors or rare clinical situations such as T3N0M0 breast cancer. However, unlike randomized trial data, observational data regarding the efficacy of treatment versus nontreatment is confounded by selection bias. To adjust for this selection bias, Rosenbaum and Rubin9, 10 proposed a propensity score matching method to estimate the average treatment effect with observational datasets. Smith et al.11 performed an analysis of the SEER data and found a survival benefit to PMRT for T1-2 breast cancer with 7 or more positive lymph nodes using a propensity score matched case-control analysis. This report describes a similar investigation involving patients with T3N0M0 disease.


Selection of Patients

After waiver from the Yale University School of Medicine Human Investigation Committee, the NCI 1973–2004 SEER database12 was examined. Patients diagnosed from 1973–1987 were not included in this analysis because they did not have adequate information regarding pathologic staging and lymph node dissection. Patients treated in 2004 were excluded to allow for sufficient follow-up. Histology and site of disease are coded in SEER according to the International Classification of Diseases for Oncology, Edition 3 (ICD-O-3).13 Patients treated from 1988 to 2003 with ductal (ICD-O-3 morphology code 8500), lobular (8520), or mixed ductal and lobular carcinoma (8522) of the breast (ICD-O-3 site code c50.0-c50.9) with T3N0M0 disease who underwent MRM were analyzed. There were 191 patients in SEER with T4N0 breast cancer who otherwise fit our selection criteria who were not included in this analysis. Patients who were men, patients for whom breast cancer was not the first malignancy, who had positive lymph nodes, who had nonbeam radiation, or had intraoperative or preoperative radiation were excluded from analysis. Nonbeam radiation was defined as any kind of brachytherapy, unsealed source, intraoperative radiation not otherwise specified, or radiation not otherwise specified. Patients who survived 6 months or less were not included in the analysis to allow inclusion only of patients who survived long enough to undergo RT and remove potential selection bias, which potentially would favor those who received RT.

Propensity Score Matching

Large observational datasets are useful for generating hypotheses or for investigating clinical areas where randomized trials are not forthcoming. However, given the ease with which statistical analysis can be performed with modern statistical software, caution must be undertaken in preparing conclusions. The use of a propensity score, described by Rosenbaum and Rubin,9 is a method for adjusting the statistical conclusions of large datasets by collapsing multiple confounding covariates into a single predictive variable (in our case, the likelihood of a patient receiving PMRT vs no PMRT, based on potentially confounding covariates). Using the propensity score to match cases and controls by their likelihood of receiving treatment reduces selection bias, and allows for an estimation of treatment effect based on balanced populations in the absence of randomized data. This is particularly important in our study, for example, given the likely clinical bias for treating younger women with RT as opposed to much older women, and the confounding effect this would have on predicting survival given the application of PMRT.

Propensity score and average treatment effect was calculated using the 2005 updated version of the psmatch2 module of STATA (College Station, Tex).14 Covariates used in the estimation of propensity score were age as a categorical variable (<50, 50–59, ≥60), number of lymph nodes examined (≤5, 6–10, 11–15, 16–20, >20), tumor size (5.1–7.5 cm, 7.6–10.0 cm, 10.1–12.5 cm, 12.6–15 cm, >15 cm), year diagnosed (≤1998, >1998), SEER registry, estrogen (ER) and progesterone receptor (PR) status (negative, positive, unknown, or test not performed), and histology. Mahalanobis metric matching was used to generate matched controls and calculate average treatment effect for all patients and for subsets of high-risk patients. As suggested by Rosenbaum and Rubin,15 we used a caliper size one-quarter of the standard deviation of the propensity scores. Covariate distribution was checked for balance using the t-test for equality of means between the PMRT and non-PMRT propensity score matched control group. Average treatment effect was bootstrapped for 50 iterations and the standard error for the average treatment effect was calculated. RT dose, technical descriptions, and chemotherapy information are not available within the SEER database.

Mortality, Overall and Cause-specific Survival

Hazard ratios for overall and cause-specific (death by breast cancer) survival were evaluated using univariate and multivariate Cox proportional hazard analysis. Survival curves were generated by the method of Kaplan-Meier. The average treatment effect of radiation on overall mortality was calculated as above with propensity score matched case-control analysis. All statistical analysis was performed with STATA/SE 9.2.


Patient Characteristics

In all, 1777 patients were detected. Patients excluded are listed in Table 1. Median follow up was 57 months (range, 7–203). Median age was 59 (range, 20–101). In all, 48.5% of patients had ER+ tumors, 40.2% had PR+ tumors, and 25.7% of patients had tumors that were both ER− and PR−. Her2 status was not recorded in SEER. In all, 568 patients (32%) received postoperative radiation therapy. The median tumor size was 6.3 cm. Median number of lymph nodes examined was 14 (range, 1–49). 45.9% of patients had poorly differentiated or anaplastic tumors. 79.5% of patients were white, 13.0% were black, and 6.7% were Asian or Pacific Islanders. Patient characteristics are listed in Table 2.

Table 1. Excluded Patients
Exclusion criteriaNo. excluded (No. remaining)
  1. SEER indicates Surveillance, Epidemiology, and End Results.

  2. All patients (669,144) with breast malignancy, 1973-2004 SEER database.

Size <50 mm or size undefined468,880 (200,264)
Positive lymph node recorded56,028 (144236)
Nonmodified radical mastectomy113,554 (30,682)
Not first malignancy4,011 (26,671)
Size data missing from SEER22,254 (4417)
Not recorded as stage IIB in years diagnosed 1988–20031,787 (2630)
Unknown number of nodes examined140 (2490)
Radiation type not external beam or sequence of radiation unknown or not postoperative163 (2327)
Histology not invasive ductal carcinoma or invasive lobular carcinoma or mixed ductal and lobular carcinoma510 (1817)
Survival ≤6 mo30 (1787)
Male10 (1777)
Table 2. Patient Characteristics
CategoryNo PMRT (%)PMRT (%)P
  • SEER indicates Surveillance, Epidemiology, and End Results; ER, estrogen receptor; PR, progesterone receptor.

  • *

    Cases were collected by SEER for these registries before 1992 but have been excluded from the limited-use file that was analyzed for this study.

Entire cohort (N = 1777)1209 (68.0)568 (32.0) 
Race  .036
 White956 (67.7)457 (32.3) 
 Black173 (74.9)58 (25.1) 
 American Indian/Alaskan Native7 (77.8)2 (22.2) 
 Asian or Pacific Islander70 (58.8)49 (41.2) 
 Other or Unknown3 (60)2 (40) 
Age, y  <.001
 20–299 (40.9)13 (59.1) 
 30–3988 (61.5)55 (38.5) 
 40–49222 (61.3)140 (38.7) 
 50–59244 (62.9)144 (37.1) 
 60–69249 (67.1)122 (32.9) 
 70+397 (80.9)94 (19.1) 
Size of tumor, cm  .33
 5.1–7.5903 (68.4)418 (31.6) 
 7.6–10.0247 (66.8)123 (33.2) 
 10.1–12.528 (59.6)19 (40.4) 
 12.6–15.019 (76.0)6 (24.0) 
 >15.012 (85.7)2 (14.3) 
No. of nodes dissected, median=13  .60
 1–576 (69.1)34 (30.9) 
 6–10260 (65.5)137 (34.5) 
 11–15384 (68.3)178 (31.7) 
 16–20269 (70.8)111 (29.2) 
 21+220 (67.1)108 (32.9) 
Histology  .001
 Invasive ductal carcinoma866 (70.2)367 (29.8) 
 Invasive lobular carcinoma241 (60.5)157 (39.5) 
 Mixed invasive ductal and lobular carcinoma102 (69.9)44 (30.1) 
Grade of tumor  <.001
 Well differentiated77 (65.8)40 (34.2) 
 Moderately differentiated319 (70.0)137 (30.0) 
 Poorly differentiated or anaplastic518 (63.5)298 (36.5) 
 Unknown295 (76.0)93 (24.0) 
Year of diagnosis  <.001
 1988–1990187 (88.2)25 (11.8) 
 1991–1993200 (81.0)47 (19.0) 
 1994–1996206 (76.0)65 (24.0) 
 1997–1999196 (63.4)113 (36.6) 
 2000–2002338 (58.2)75 (41.8) 
 200382 (52.2)75 (47.8) 
ER status  <.001
 Not performed32 (80)8 (20) 
 Positive557 (64.6)305 (35.4) 
 Negative313 (61.5)196 (38.5) 
 Borderline or unknown307 (83.9)59 (16.1) 
PR status  <.001
 Not performed49 (80.3)12 (19.7) 
 Positive457 (64.0)257 (36.0) 
 Negative386 (62.1)236 (37.8) 
 Borderline or unknown317 (83.4)63 (16.6) 
SEER registry (first diagnosis year data reported to SEER)  <.001
 San Francisco/Oakland (1973)119 (62.0)73 (38.0) 
 Connecticut (1973)120 (88.2)16 (11.8) 
 Metropolitan Detroit (1973)154 (71.3)62 (28.7) 
 Hawaii (1973)33 (62.3)20 (37.7) 
 Iowa (1973)97 (60.6)63 (39.4) 
 New Mexico (1973)31 (68.9)14 (31.1) 
 Seattle/Puget Sound (1973)72 (57.1)54 (42.9) 
 Utah (1973)42 (82.3)9 (17.6) 
 Metropolitan Atlanta (1975)86 (79.6)22 (20.4) 
 Alaska* (1992)2 (66.7)1 (33.3) 
 San Jose/Monterey* (1992)33 (58.9)23 (41.1) 
 Los Angeles* (1992)189 (75.9)60 (24.1) 
 Rural Georgia* (1992)5 (100)0 (0) 
 Greater California, excluding San Francisco/Oakland, Los Angeles, San Jose/Monterey (2000)100 (57.5)74 (42.5) 
 Kentucky (2000)20 (58.8)14 (41.2) 
 Louisiana (2000)55 (61.8)34 (38.2) 
 New Jersey (2000)51 (63.7)29 (36.2) 

Unadjusted Overall and Cause-Specific Survival Analysis

In unadjusted univariate analysis (Table 3), 5- and 10-year overall survival was 79% and 61% for no PMRT versus 84% and 71% for PMRT (P = .008) by the method of Kaplan-Meier (Fig. 1). In addition, in unadjusted univariate analysis of the entire patient cohort (all patients regardless of PMRT) using Cox proportional hazards analysis, having fewer lymph nodes dissected (P = .007), decreasing age as a categorical (Fig. 2), and continuous variable (P <.001), having an ER−/PR− tumor (P = .012), or being of African American race (P = .026) predicted poorer survival.

Figure 1.

Unadjusted Kaplan-Meier survival curves for patients receiving postmastectomy radiotherapy (PMRT) versus no PMRT.

Figure 2.

Unadjusted Kaplan-Meier curves for survival by age.

Table 3. Unadjusted 5-Year and 10-Year Overall Survival Rates by Kaplan-Meier and Univariate and Multivariate Cox Proportional Hazard Analysis for Hazard of Death
 % 5-Year OS* [SE]% 10-Year OS* [SE]Univariate hazard ratio [SE]PMultivariate model hazard ratio [SE]P
  • SEER indicates Surveillance, Epidemiology, and End Results; PMRT, postmastectomy radiotherapy; ER, estrogen receptor; PR, progesterone receptor.

  • *

    Overall survival.

  • Not applicable.

  • Not included in multivariate model.

  • §

    Index value.

  • ||

    Too soon or too few events to calculate.

  • Calculated as continuous variable, hazard ratio is per 1 year or per 1 lymph node.

Women80.6% [1.1%]62.9% [1.6%] 
PMRT   .008  
 Yes84.1% [1.8%]70.7% [3.3%]0.75 [0.09] 0.93 [0.11].54
 No79.3% [1.3%]60.7% [1.8%]1§ 1§ 
Race   .06  
 White81.5% [1.2%]63.4% [1.8%]1§ 1§ 
 Black71.3% [3.5%]56.4% [4.6%]1.33 [0.17] 1.36 [0.20].03
 American Indian/Alaskan Native76.2% [14.8%]57.1% [19.9%]1.17 [0.68] 0.55 [0.41].53
 Asian or Pacific Islander87.0% [3.4%]69.7% [6.6%]0.79 [0.17] 0.89 [0.64].61
 Other or Unknown|||||| || 
Age, y   <.00011.048 [0.004]<.001
 20–2985.9% [7.6%]85.9% [7.6%]1§   
 30–3987.7% [3.0%]78.1% [4.4%]0.84 [0.51]   
 40–4988.4% [1.8%]81.4% [2.7%]0.77 [0.46]   
 50–5986.8% [1.9%]73.5% [3.2%]1.02 [0.60]   
 60–6981.6% [2.3%]66.0% [3.5%]1.46 [0.86]   
 70+67.4% [2.4%]36.4% [3.1%]3.26 [1.89]   
Size of tumor, cm   .89  
 5.1–7.581.2% [1.2%]63.8% [1.8%]1§  
 7.6–10.078.8% [2.4%]61.5% [3.7%]1.08 [0.12]  
 10.1–12.576.2% [7.7%]47.5% [12.4%]1.19 [0.35]  
 12.6 –15.075.4% [9.8%]58.2% [13.2%]1.19 [0.42]  
 >15.092.9% [6.9%]51.6% [23.4%]0.78 [0.45]  
No. of nodes dissected   .0070.987 [0.007].09
 1–565.9% [6.0%]37.9% [8.8%]1§   
 6–1078.5% [2.4%]58.2% [3.8%]0.67 [0.13]   
 11–1584.0% [1.7%]61.0% [3.0%]0.57 [0.11]   
 16–2079.2% [2.3%]68.1% [3.1%]0.52 [0.10]   
 20+82.3% [2.3%]68.6% [3.3%]0.50 [0.10]   
Histology   .50  
 Invasive ductal carcinoma79.4% [1.3%]62.4% [1.9%]1§ 1§ 
 Invasive lobular carcinoma84.3% [2.1%]65.9% [3.6%]0.89 [0.10] 0.91 [0.12].47
 Mixed ductal and lobular carcinoma81.0% [3.8%]57.4% [6.6%]1.06 [0.18] 1.36 [0.25].09
Grade   .063  
 Well differentiated87.5% [3.9%]60.8% [9.0%]1§ 1§ 
 Moderately differentiated83.0% [2.1%]62.5% [3.6%]1.29 [0.33] 1.28 [0.33].34
 Poorly differentiated or anaplastic76.2% [1.7%]62.4% [2.4%]1.55 [0.38] 1.65 [0.42].048
 Unknown84.7% [2.0%]65.0% [2.9%]1.22 [0.31] 1.16 [0.30].57
Year of diagnosis   .030.98 [0.02].26
 1988–199078.7% [2.8%]60.5% [3.4%]1§   
 1991–199379.2% [2.6%]57.5% [3.2%]1.08 [0.14]   
 1994–199680.0% [2.4%]66.1% [3.3%0.81 [0.12]   
 1997–199985.7% [2.0%]§,*0.68 [0.12]   
 2000–2002||||1.02 [0.16]   
 2003||||0.54 [0.28]   
ER status   .27  
 Not performed76.9% [7.3%]62.7% [9.6%]0.90 [0.27] 0.56 [0.28].25
 Positive85.3% [1.4%]61.7% [2.7%]0.81 [0.09] 0.70 [0.12].03
 Negative74.9% [2.1%]65.2% [2.8%]1§ 1§ 
 Borderline or unknown78.5% [2.3%]62.2% [3.0%]0.93 [0.12] 1.11 [0.46].79
PR status   .21  
 Not performed79.0% [5.8%]56.3% [9.2%]0.97 [0.24] 1.13 [0.45].77
 Positive86.0% [1.5%]62.4% [2.8%]0.79 [0.09] 0.91 [0.14].55
 Negative76.1% [1.9%]64.9% [2.7%]1§ 1§ 
 Borderline or unknown78.6% [2.3%]62.6% [3.0%]0.93 [0.11] 0.65 [0.26].29
Primary site   .61  
 Nipple (c50.0)77.4% [11.9%]25.8% [15.4%]1§ 1§ 
 Central portion or breast (subareolar) (c50.1)79.8% [3.6%]70.8% [4.8%]0.59 [0.24] 0.73 [0.31].45
 Upper inner quadrant of breast (c50.2)77.2% [3.6%]56.3% [5.6%]0.75 [0.30] 0.99 [0.41].98
 Lower inner quadrant of breast (c50.3)74.4% [6.8%]53.0% [9.7%]0.79 [0.36] 0.79 [0.37].62
 Upper outer quadrant of breast (c50.4)83.6% [1.7%]61.2% [2.9%]0.58 [0.23] 0.82 [0.33].62
 Lower outer quadrant of breast (c50.5)82.1% [5.3%]74.3% [7.1%]0.46 [0.21] 0.59 [0.28].26
 Axillary tail of breast (c50.6)|||||| 0.95 [1.03].96
 Overlapping lesion of breast (c50.8)78.8% [2.2%]65.3% [3.1%]0.61 [0.24] 0.91 [0.37].82
 Breast, NOS (c50.9)80.6% [2.6%]64.5% [4.0%]0.58 [0.23] 0.72 [0.29].42
SEER registry site   .35  
 San Francisco/Oakland81.9% [2.9%]62.1% [4.6%]1§ 1§ 
 Connecticut83.9% [3.3%]61.4% [5.1%]0.94 [0.18] 0.76 [0.15].17
 Metropolitan Detroit77.7% [3.0%]64.7% [3.9%]1.02 [0.18] 0.90 [0.16].57
 Hawaii83.8% [5.3%]72.0% [7.1%]0.76 [0.22] 0.71 [0.23].29
 Iowa78.5% [3.4%]60.3% [4.7%]0.96 [0.18] 0.84 [0.16].37
 New Mexico69.5% [7.1%]57.2% [8.9%]1.31 [0.37] 1.46 [0.42].19
 Seattle/Puget Sound83.8% [3.5%]61.8% [5.5%]0.85 [0.18] 0.81 [0.17].32
 Utah78.4% [6.1%]65.1% [8.0%]1.00 [0.28] 1.08 [0.31].80
 Metropolitan Atlanta75.2% [4.4%]63.2% [5.7%]1.01 [0.21] 0.83 [0.18].39
 Alaska|||||| || 
 San Jose/Monterey78.1% [5.9%]42.6% [13.9%]1.10 [0.31] 1.21 [0.35].51
 Los Angeles83.4% [2.6%]63.8% [4.4%]0.85 [0.15] 0.84 [0.16].37
 Rural Georgia|||||| || 
 Greater California, excluding San Francisco/Oakland, Los Angeles, and San Jose/Monterey||||0.47 [0.15] 0.54 [0.19].08
 Kentucky||||1.53 [0.66] 1.77 [0.80].20
 Louisiana||||1.47 [0.46] 1.26 [0.42].49
 New Jersey||||0.80 [0.32] 0.89 [0.34].75

In the multivariate Cox proportional hazards model of predictors of overall survival (Table 3), PMRT no longer remained a significant predictor (P = .54). However, other factors that remained significant as predictors for worsened survival overall included increasing age (P <.001), black race (P = .03), and having a poorly differentiated or anaplastic tumor (P = .048). Also in the multivariate model, having ER+ disease in comparison to ER− disease was correlated with improved survival (P = .03) for all patients with T3N0 breast cancer.

Tumor location within the breast was recorded in SEER and included in univariate and multivariate analysis. There was no significant difference in terms of overall survival based on tumor location. In general, tumors were most frequently located in the upper outer quadrant of the breast.

Univariate and multivariate Cox proportional hazards analysis was performed (data not reported) for cause-specific survival (death due to breast cancer). In univariate analysis and in the multivariate model, black race (P = .008), having a poorly differentiated or anaplastic tumor (P = .008), and having increasing age (P = .001) predicted greater hazard for breast cancer death in our cohort of patients with T3N0 disease. PMRT use was not a significant predictor of cause-specific survival in univariate or multivariate analysis for the entire cohort or for high-risk subsets, listed in Table 4 using a bootstrapped propensity score analysis for average treatment effect on cause specific survival (Table 4). More recent year of diagnosis predicted for decreased hazard of breast cancer death (P = .019) in univariate analysis but not the multivariate model (P = .26) for all patients with T3N0 breast cancer.

Table 4. Bootstrapped Propensity Score Case-Matched Average Treatment Effect (Mahalanobis Metric-Matching Within Calipers Defined By 25% of the Standard Deviation of the Estimated Propensity Scores) for All Patients and for Different High-Risk Subgroups of Patients.
Patient GroupAverage treatment effect of PMRT risk of overall mortality [SE]PAverage treatment effect of PMRT risk of cancer specific mortality [SE]P
  1. PMRT indicates postmastectomy radiation therapy; SE, standard error of the mean; ER, estrogen receptor.

All patients−5.3% [6.4].43−0.9% [6.8%].90
Patients with 1-13 lymph nodes examined−13.5% [11%].22−9.6% [7.4%].20
Age, y, ≤50−2.3% [10.4%].83−4.5% [9.3%].63
Age, y, ≤65−3.2% [8.3%].70−1.6% [6.0%].79
ER−−10.4% [11.1%].35−6.2% [8.6%].47
Poorly differentiated or anaplastic−8.1% [8.6%].35+1.6% [11.0%].88
Black+40% [36%].26+40% [37%].29
Invasive lobular carcinoma−3.1% [13.0%].81−6.2% [8.6%].47
Tumor size >7.5 cm−11.1% [23.1%].63−11.1% [15%].46

Propensity Score Estimation and Average Treatment Effect

The propensity score was successfully estimated. There was no difference in overall survival and mortality between the PMRT and non-PMRT matched control group, either by survival analysis (data not reported) or by bootstrapped average treatment effect on mortality (Table 4). This analysis was performed for all patients and for high-risk subsets, and no significant average treatment effect of PMRT was found.


An improvement in overall survival remains the ultimate goal of anticancer therapy. Adjuvant and primary radiation therapy aims to improve overall survival by treating tumor and areas at risk with locoregional therapy. In this vein, the Early Breast Cancer Trialists' Collaborative Group meta-analysis suggested that for a 20% absolute reduction in 5-year local recurrence, there is a corresponding 5% reduction in 15-year breast cancer mortality.4 It has been suggested that T3N0M0 breast tumors may represent a distinct clinical and biological entity, with the ability to grow locally without metastasizing to regional lymph nodes.6 The question then arises, when is a T3N0M0 breast cancer optimally treated with mastectomy with or without systemic therapy and when does a breast cancer require additional local adjuvant therapy, specifically radiotherapy? Some authors have suggested that if patients with T3N0M0 breast cancer have a less than 10% absolute risk of 5-year isolated local recurrence, as has been suggested by prior studies,6, 7 then these patients might not require routine PMRT. However, as noted previously, there has been identification of higher-risk patients who would theoretically benefit from radiotherapy.1, 5–7 Identifying these high-risk subgroups and individualizing treatment on a case-by-case basis may then form the most rational treatment.

Our study of a large observational cohort was able to confirm the existence of higher-risk subgroups of patients with T3N0M0 invasive ductal or invasive lobular breast carcinoma. Patients with high-grade tumors, ER− tumors, older age, and black race were suggested to have a higher hazard for death. In addition, invasive lobular carcinoma and higher-grade tumors predicted for PMRT use, whereas older age predicted for PMRT omission. These conclusions make intuitive sense based on prior literature indicating race, grade of tumor, age, and ER status being important for prognosis and outcome of patients with breast cancer,3, 5, 6, 16 and supports the validity of the SEER database in our investigation.

Unadjusted multivariate survival analysis was unable to show a benefit or detriment to the inclusion or omission of PMRT. Given the many factors that go into a decision to offer PMRT, and each patient's own bias regarding radiotherapy, we attempted a propensity score matched analysis to try and reduce this selection bias in the absence of randomized data. Our analysis was also unable to find a definite survival benefit to PMRT, in all patients, or in selected high-risk subgroups. The lack of overall survival benefit in our study is significant in light of a similar SEER study that was performed on the cohort of patients with T1-2, node-positive breast cancer11 that did find an overall survival benefit in patients with 7 or more involved regional lymph nodes. However, the sample size of the population in that study was much larger (18,038 patients) given the more frequent nature of T1-2, node-positive disease.

The lack of a perceived reduction in mortality in our study may have been due to the relatively short median follow-up (57 months), or possibly because the benefit is too small to be detected by the analysis of our cohort. When this analysis was repeated using breast cancer death as the endpoint, the results were similar. Unfortunately, it was not possible to measure locoregional recurrence as an endpoint, as these data are not accurately recorded in SEER. The proportion of patients with local failure and local failure plus distant disease was significant in prior studies, and possibly as systemic therapy improves there will be a significant impact of controlling local disease in this subgroup leading to improvement in survival. The lack of survival benefit in patients who received PMRT versus no PMRT is consistent with previously published data from the NSABP.7

The Early Breast Cancer Trialists' Collaborative Group (EBCTCG) meta-analysis of trials for PMRT4 showed that the survival benefit to PMRT appeared significant after 5 years of follow up. In all, 838 patients in our original study cohort were followed for more than 5 years. Based on the EBCTCG analysis, the proportional mortality benefit from radiotherapy is likely similar across stages and nodal status. Therefore, we expect an approximate 15% to 20% relative reduction in the risk of death. To detect a 15% to 20% relative reduction in the risk of death, our study would have required 530-1020 patients with more than 5 years of follow up, and therefore it is likely possible that our study was underpowered to detect such a benefit. Nonetheless, absence of a detected benefit is still a valuable observation, indicating that the relative benefit from PMRT in patients with T3N0 is not appreciably larger than expected.

The EBCTCG also investigated the effects of chemotherapy on recurrence and 15-year survival,17 and was able to detect a significant advantage of anthracycline-based chemotherapy over CMF. The breast cancer death rate ratio between anthracycline-based therapy versus CMF was 0.84 (SE 0.03) (2P < .00001) but drew from a group of 14,000 women to make this discovery.

This study was limited by the retrospective, nonrandomized and observational nature of the SEER database, along with the limitations created by variables that were not included in the database. Patients who received PMRT were selected based on clinician, institution, and patient preference. The propensity score matched analysis attempted to adjust for this selection bias, but was unable to adjust for unrecorded variables, and therefore the potential for a hidden bias remains. In addition, the SEER database did not record rates of locoregional recurrence, and also did not record medical comorbidities, medical interventions (including chemotherapy or hormonal therapy), surgical margin data or radiation dose or technique (including whether CT planning was used, boost treatment was given, or whether internal mammary lymph nodes were irradiated). These factors that are not recorded in SEER are important in predicting overall survival and locoregional control.

In addition, limiting our statistical analysis is the inability to take into account unrecorded confounding variables, such as comorbid illness, radiation dose and technique, hormonal therapy, chemotherapy, and histopathologic characteristics such as lymphovascular invasion and Her2+ status. Therefore, whether or not PMRT improves survival in patients with or without the above characteristics remains unclear.


Based on the analysis of a large observational cohort of patients, there does not appear to be an overall survival benefit to the use of PMRT in patients with T3N0M0 breast cancer. Patients with high-grade tumors, African American patients, and patients with ER− tumors should be considered to be at higher risk of overall and breast cancer specific mortality. An incremental improvement in survival with the use of PMRT was not detected, although whether there are subpopulations that are at higher risk that may benefit remains to be seen. The benefit in terms of locoregional control was not evaluated in this study, and based on other studies there is certainly a role for PMRT in this group to improve local-regional control.