Effectiveness of Radiation Therapy for Older Women With Early Breast Cancer

Transcription

1 Effectiveness of Therapy for Older Women With Early Breast Cancer Benjamin D. Smith, Cary P. Gross, Grace L. Smith, Deron H. Galusha, Justin E. Bekelman, Bruce G. Haffty Background: Recent clinical trials have questioned the necessity of breast radiation therapy for older women with early breast cancer. However, the effectiveness of radiation therapy for older women in the community setting has not been addressed. Methods: We used the Surveillance, Epidemiology, and End Results (SEER) Medicare database from January 1, 1992, through December 31, 1999, to identify 8724 women aged 7 years or older treated with conservative surgery for small, lymph node negative, estrogen receptor positive (or unknown receptor status) breast cancer. We used a proportional hazards model to test whether radiation therapy was associated with a lower risk of a combined outcome, defined as a second ipsilateral breast cancer reported by SEER and/ or a subsequent mastectomy reported by Medicare claims. All statistical tests were two-sided. Results: therapy, compared with no radiation therapy, was associated with a lower risk of the combined outcome (hazard ratio =.19, 95% confidence interval =.14 to.28). therapy was associated with an absolute risk reduction of 4. events per 1 women at 5 years (i.e., from 5.1 events without radiation therapy to 1.1 with radiation therapy) and 5.7 events per 1 persons at 8 years (i.e., from 8. events without radiation therapy to 2.3 with radiation therapy) ( P <.1, log-rank test). therapy was most likely to benefit those aged 7 79 years without comorbidity (number needed to treat [NNT] to prevent one event = 21 to 22 patients) and was least likely to benefit those aged 8 years or older with moderate to severe comorbidity (NNT = 61 to 125 patients). Conclusion: For older women with early breast cancer, radiation therapy was associated with a lower risk of a second ipsilateral breast cancer and subsequent mastectomy. Patients aged 7 79 years with minimal comorbidity were the most likely to benefit, and older patients with substantial comorbidity were least likely to benefit. [J Natl Cancer Inst 26;98:681 9] For many years, breast radiation therapy after conservative surgery has been a standard of care for patients with early breast cancer (1 4 ). However, a recent randomized trial from the Cancer and Leukemia Group B (trial C9343) explored whether breast radiation therapy was necessary in those aged 7 years and older with small, estrogen receptor (ER) positive (or unknown ER status), clinically lymph node negative breast cancer treated with conservative surgery and tamoxifen (5 ). Although radiation therapy conferred a statistically significant reduction in the risk of local regional relapse, the absolute risk reduction was only three events per 1 persons at 5 years of follow up. Given the small absolute benefit of radiation therapy, coupled with its cost (6 ) and morbidity (5 ), new clinical guidelines have recommended omission of breast radiation therapy for women who meet the entry criteria for trial C9343 (7 ). Although meaningful differences may exist between patients who enroll in clinical trials and those in the general population (8 11 ), the effectiveness of breast radiation therapy in the community setting has not been addressed (12 ). Identification of those who are most and least likely to benefit from breast radiation therapy would allow clinicians to tailor their treatment recommendations appropriately. To address these critical issues, we identified patients in the Surveillance, Epidemiology, and End Results (SEER) Medicare database who would have been eligible for the C9343 trial. We investigated whether radiation therapy was associated with a lower risk of a second breast cancer event i.e., a combined outcome defined as a second ipsilateral breast cancer as reported by SEER and/or a subsequent mastectomy as reported by Medicare claims. P ATIENTS AND METHODS Data Source The National Cancer Institute s SEER Medicare database tracks incident malignancies in Medicare beneficiaries, linking tumor-specific variables coded by local SEER registries to Medicare claims. During the study period of January 1, 1992, through December 31, 1999, data were available from 11 tumor registries representing approximately 14% of the United States population (13 ). Study Sample and Outcomes From 1992 through 1999, women aged 7 years or older with breast cancer were identified in the database. The following women were excluded because they would not have met entry criteria for trial C9343: tumor size not reported or greater than 2. cm ( patients excluded); no invasive component (6679 patients excluded); histology not consistent with epithelial origin (1355 patients excluded); clinical stage T4 (54 patients excluded); ER-negative status (6454 patients excluded); involved re - gional lymph nodes ( patients excluded); distant metastasis Affi liations of authors: Departments of Therapeutic Radiology (BDS, BGH) and Internal Medicine (CPG, GLS, DHG), Yale University School of Medicine, New Haven, CT; Department of Oncology, Memorial Sloan-Kettering Cancer Center (JEB), New York, NY; Department of Oncology, University of Medicine and Dentistry New Jersey-Robert Wood Johnson Medical School, Piscataway, NJ (BGH). Correspondence to: Benjamin D. Smith, MD, Department of Therapeutic Radiology, Yale School of Medicine, 333 Cedar Street, New Haven, CT ( bensmith@alumni.rice.edu ). See Notes following References. DOI: 1.193/jnci/djj186 The Author 26. Published by Oxford University Press. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org. Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26 ARTICLES 681

2 at presentation (3332 patients excluded); stage not reported (2611 patients excluded); not treated with breast-conserving surgery ( patients excluded); history of prior malignancy (176 patients excluded); bilateral disease (51 patients excluded); and disease not pathologically confirmed (1797 women excluded). Thus, patients met all clinical entry criteria (patients could also be excluded for more than one reason; a total of 9413 patients were excluded solely because of treatment with mastectomy). Within this clinical group, patients with a second primary cancer diagnosed within 9 months after the index primary were excluded (435 patients) because billing records could not discriminate between procedures performed for the index cancer versus the second cancer. Patients with inadequate Medicare records (641 patients without Part A and B Medicare coverage and 3547 patients without fee-for-service coverage from 12 months before diagnosis to 9 months after diagnosis) were also excluded, leaving 8932 patients with adequate baseline Medicare claims (patients could, again, be excluded for more than one reason). Cancer-specific treatments received within 9 months of initial diagnosis were considered to represent treatment for the index cancer (14 ). A sensitivity analysis tested whether inclusion of treatments received within only 6 months or up to 12 months after initial diagnosis would alter the study results. The follow-up period began 9 months after diagnosis and continued through December 31, 22. The primary outcome a second breast cancer event was a combined outcome defined as a second ipsilateral, pathologically confirmed, invasive breast cancer reported by SEER and/or a subsequent mastectomy reported by Medicare claims. The secondary outcome was risk of repeat breast-conserving surgery reported by Medicare claims. Breastconserving surgery was not included in the primary outcome because it may be performed to rule out cancer and is, therefore, not specific for local relapse. Medicare claims do not consistently report laterality for breast surgical procedures. Because laterality is important in establishing a second ipsilateral versus contralateral event, we excluded two patients with unknown laterality at initial diagnosis and 26 patients reported by SEER as experiencing any contralateral breast event in the follow-up period. The final cohort therefore included 8724 patients aged 7 years or older who were treated with conservative surgery for small, lymph node negative, invasive breast cancer that was ER positive (or of unknown ER status). A sensitivity analysis tested whether inclusion of patients with contralateral breast events would alter the study results. Treatment-Related Variables Initial surgery was determined from the site-specific surgery variable in SEER and from Medicare claims (breast-conserving surgery: International Classification of Diseases-9 [ICD-9] Procedure Codes 85.2, 85.21, 85.22, 85.23, or 85.25; Current Procedural Terminology Codes 1911, 1912, 19125, 1916, or 19162; mastectomy: ICD-9 Procedure Codes 85.41, 85.42, 85.43, 85.44, 85.45, 85.46, 85.47, or 85.48; or Current Procedural Terminology Codes 1918, 19182, 192, 1922, or 1924) (15 19 ). The most aggressive surgical procedure reported by SEER or Medicare within 9 months of diagnosis was considered the definitive surgery. Treatment with radiation therapy was obtained from SEER data and Medicare claims (ICD-9 Procedure Codes or 92.29; ICD-9 Diagnosis Codes V58., V66.1, or V67.1; Current Procedural Terminology Codes or ; or Revenue Center Codes 33 or 333) (14,16 2 ). Patients were considered to have received radiation therapy if either SEER or Medicare reported treatment with radiation therapy within 9 months of diagnosis. Treatment with chemotherapy within 9 months of diagnosis was determined from Medicare claims (ICD-9 Procedure Code 99.25; ICD-9 Diagnosis Codes V58.1, V66.2, or V67.2; Current Procedural Terminology Codes ; Health Care Procedure Coding System Codes J9 J9999 or Q83 Q85; or Revenue Center Codes 331, 332, or 335) (16 19,21,22 ). The method of axillary lymph node assessment was determined from SEER data (pathologic assessment if at least one lymph node was sampled, otherwise a clinical assessment). Other Variables Patient characteristics included age at diagnosis, race (white, black, white Hispanic, Asian/Pacific Islander, or other/unknown), year of diagnosis, marital status (widowed, married, single, separated/divorced, or unknown), SEER registry, urban versus rural residence (big metropolitan, metropolitan, urban, less urban, or rural), median income of census tract or zip code (census tract data were used if available, otherwise zip code data were used) (23 ), percentage of adults in census tract or zip code with less than 12 years education (census tract data were used if available, otherwise zip code data were used) (23 ), and number of physician visits on separate days during a prediagnosis interval from 12 months to 1 month. Distance from the patient s zip code to the nearest radiotherapy facility was calculated from radiotherapy facility zip codes provided by the American College of Radiology (ZIPFind Deluxe 5., Xionetic Technologies, Inc., Bozeman, MT). A modified Charlson comorbidity index (24 26 ) was calculated from Part A and Part B Medicare claims during a prediagnosis interval from 12 months to 1 month. To enhance specificity, Part B diagnosis codes were included only if they appeared more than once over a period of more than 3 days or in Part A claims as well (27,28 ). Tumor characteristics included size, location within the breast (inner quadrant, outer quadrant or central, overlapping, or unknown) (29 ), grade (well differentiated, moderately differentiated, poorly or undifferentiated, or unknown), histology (ductal, lobular, or other) (3 ), ER status (positive, borderline, negative, or unknown, as reported by SEER), progesterone receptor status (positive, borderline, negative, or unknown, as reported by SEER), and laterality (right or left). Hospital characteristics included teaching status (yes or no), which was self-reported in the Hospital Cost Report Information Systems files available for 1996, 1998, and 2 (31 ). Data regarding surgical margin status and adjuvant endocrine therapy were not reported by SEER Medicare. All variables are summarized in Appendix A. Statistical Analysis Because the risk of local relapse at 5 and 8 years is a commonly reported endpoint in breast cancer clinical trials (3 5,32,33 ), the unadjusted risk of each outcome at 5 and 8 years was estimated by use of the Kaplan Meier method. The Kaplan Meier method was selected to provide estimates that could be directly compared with the results of trial C9343. Unadjusted associations between radiation therapy and outcomes were tested by use of the log-rank test. Patients were censored at the time of 682 ARTICLES Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26

3 death, loss of Medicare Part A or B coverage, or loss of fee-forservice coverage. If none of these events occurred, patients were censored at the end of the follow-up period, December 31, 22. The hazard functions of the groups receiving radiation therapy or no radiation therapy were approximately parallel, indicating that the proportional hazards assumption was satisfied. Therefore, the adjusted relationship between breast radiation therapy and local relapse was determined with a Cox proportional hazards model adjusted for patient, tumor, treatment, and hospital covariates that were considered statistically significant at a P value of less than or equal to.25 in unadjusted analyses. Clinically relevant covariates including age at diagnosis, tumor size, histology, ER status, comorbidity, and treatment with chemotherapy were included in the final model, regardless of their unadjusted P values. Covariates lacking linear relationships with the outcome were entered as categorical (dummy) variables, with missing values as a dummy category. The model was stratified by SEER region and the year of diagnosis to account for variations with geography and time. Prespecified interaction terms were used to determine whether the effect of radiation therapy differed with respect to age, comorbidity, tumor size, histology, and ER status. The adjusted number needed to treat (NNT) indicates the number of patients who require breast radiation therapy to prevent one local relapse and accounts for the competing risk of death from any cause. The adjusted NNT was calculated by dividing the unadjusted NNT by the survival probability point estimate (34 ). The unadjusted NNT is the reciprocal of the absolute risk reduction estimated with the Kaplan Meier method. This method assumes that radiation therapy does not affect overall survival through 8 years of follow-up, an assumption that is supported by the results of trial C9343 and other clinical trials (3 5,32 ). For example, if radiation therapy conferred a 4% absolute improvement (unadjusted) in the risk of local relapse at 5 years and if 5% of patients were alive at 5 years, then the unadjusted NNT would be 25 patients (i.e., 1/.4), and the adjusted NNT would be 5 patients (i.e., 25/.5). All statistical analyses were two-sided with an α value of less than or equal to.5 and were conducted with SAS version 9.1 (SAS Institute, Cary, NC). The Yale School of Medicine Institutional Review Board approved this study and granted a waiver of informed consent. Comparison Cohort To determine whether the benefit associated with radiation therapy was sensitive to the eligibility criteria for trial C9343, we formed a comparison cohort consisting of patients who failed to meet the entry criteria for trial C9343 because their age was years, their tumor size was cm, or their tumor was ER negative. For each subgroup in the comparison cohort, the unadjusted risk of a second breast cancer event and the absolute risk reduction associated with radiation therapy were calculated by use of the Kaplan Meier method. R ESULTS Baseline Characteristics Of the 8724 patients identified, median age was 77 years (interquartile range [IQR] = 73 to 82 years), 7842 (9%) were white, and 365 (4%) were black. Median tumor size was 1. cm (IQR =.8 to 1.5 cm), 6353 (73%) had ductal histology, and 1789 (21%) had unknown ER status. Comorbidity was absent in 5666 (65%), mild in 1916 (22%), moderate to severe in 884 (1%), and unknown in 258 (3%). A total of 636 patients (73%) received breast radiation therapy and 255 (3%) received chemotherapy. Treatment with radiation therapy was associated with younger age, lack of comorbid illness, treatment with chemotherapy, and pathologic axillary lymph node assessment ( Table 1 ). Concordance between SEER and Medicare claims was high for both radiation therapy ( κ =.79) and surgery ( κ =.9). Unadjusted Risk of Primary and Secondary Outcomes Median follow-up was 5. years (IQR = 3.5 to 7. years). Of the 8724 patients included in the analysis, 84 patients (1.%) experienced a second ipsilateral breast cancer reported by SEER Table 1. Baseline characteristics of patients in the Surveillance, Epidemiology, and End Results (SEER) Medicare database who would have been eligible for the C9343 trial Covariate No. No radiation therapy, % therapy, %* P Entire cohort Median age 83 (78 87) 76 (73 79) <.1 Race White Black Asian/Pacific Islander White Hispanic Other Comorbidity index < Unknown Median tumor size 1. (.7 1.5) 1. (.8 1.5).41 Histology Ductal <.1 Lobular Other Estrogen receptor status Positive <.1 Borderline 25 Unknown Progesterone receptor status Positive <.1 Borderline Negative Unknown Chemotherapy Yes <.1 No Axillary lymph node assessment Clinical <.1 Pathologic Location of initial surgery Community hospital Teaching hospital * therapy was defined as any billing code for radiation therapy present in Medicare claims within 9 months of diagnosis or any treatment with radiation therapy reported by SEER data. P value from two-sided Pearson s chi-square test, unless otherwise specified. All statistical tests were two-sided. Data are years (interquartile range). P value from two-sided Wilcoxon rank-sum test. Data are centimeters (interquartile range). In concert with SEER Medicare guidelines, cell sizes with an absolute number of cases of less than five have been suppressed. Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26 ARTICLES 683

4 data, and 166 patients (1.9%) underwent subsequent mastectomy as reported by Medicare claims. therapy was associated with a reduced risk for a second ipsilateral breast cancer ( P <.1, Fig. 1, A, and Table 2 ), subsequent mastectomy (P <.1, Fig. 1, B, and Table 2 ), and the combined outcome of either a second ipsilateral breast cancer and/or subsequent mastectomy (i.e., a second breast cancer event) ( P <.1, Fig. 1, C, and Table 2 ). At 5 years, the risk of a second breast cancer event was 5.1% (95% confidence interval [CI] = 4.1 to 6.2) in patients receiving no radiation therapy and 1.1% (95% CI =.79 to 1.4) in patients treated with radiation therapy ( Table 2 ). At 8 years, the risk of a second breast cancer event increased to 8.% (95% CI = 6.2 to 9.8) in patients receiving no radiation therapy and 2.3% (95% CI = 1.7 to 2.9) in patients treated with radiation therapy ( Table 2 ). Of the 8724 patients included in the analysis, 794 (9%) patients underwent repeat breast-conserving surgery. therapy was not associated with risk of this outcome ( P =.83) (Fig. 1, D ). Adjusted Analysis After adjusting for patient, tumor, treatment, and hospital characteristics, breast radiation therapy remained associated with a reduced risk of a second breast cancer event (the combined outcome) (hazard ratio [HR] =.19, 95% CI =.14 to.28; P <.1; Table 3 ). therapy was also associated with a Fig. 1. Association of radiation therapy with outcomes. Patients were at risk for all outcomes beginning 9 months after diagnosis. A ) Second ipsilateral breast cancer reported by Surveillance, Epidemiology, and End Results (SEER). This outcome was defined as a second ipsilateral, pathologically confirmed, invasive breast cancer reported by SEER data. B ) Subsequent mastectomy reported by Medicare claims. This outcome is defined as a subsequent mastectomy reported by Medicare claims. C ) Second breast cancer event defined as a second ipsilateral, pathologically confirmed, invasive breast cancer reported by SEER data or as a subsequent mastectomy reported by Medicare claims. D ) Repeat breast-conserving surgery as reported by Medicare claims. This outcome is defined as repeat breast-conserving surgery reported by Medicare claims. RT = radiation therapy. Error bars = 95% confidence intervals. * P values were calculated from a two-sided log-rank test. A. Second ipsilateral breast cancer reported by SEER 2 No 15 P <.1* % Number at Risk: No RT RT 2,364 6,36 2,78 6,179 1,455 4,82 reduced risk for each component of the combined outcome (for a second ipsilateral breast cancer reported by SEER, HR =.13, 95% CI =.7 to.23; P <.1; and for subsequent mastectomy reported by Medicare, HR =.21, 95% CI =.14 to.31; P <.1) ( Table 4 ). The effect size associated with radiation therapy was stable regardless of whether the model was unadjusted, adjusted only for age, or fully adjusted (Table 4 ). Breast radiation therapy was associated with a large benefit for those with lobular histology ( P interaction =.2), with an 8- year second breast cancer event risk of 16% (95% CI = 5.5 to 27) in patients receiving no radiation therapy, compared with a risk of.38% (95% CI =. to 1.1) in patients treated with radiation therapy (adjusted HR =.18, 95% CI =.2 to.14; P <.1). Besides radiation therapy, other covariates associated with a second breast cancer event included being a member of the black race compared with being white (HR = 2.22, 95% CI = 1.18 to 4.21; P =.1), having a widowed marital status compared with having a married marital status (HR =.69, 95% CI =.49 to.97; P =.3), and having a progesterone receptor negative status compared with having a positive status (HR = 1.49, 95% CI = 1. to 2.22; P =.5). Age, comorbidity, tumor size, and ER status were not associated with risk of a second breast cancer event, and the interaction of radiation therapy with these variables was not statistically significant. In a sensitivity analysis, the adjusted relationship between radiation therapy and risk of a second breast cancer event was not altered by 78 2, , C. Second breast cancer event (either A. and/or B.) D. Repeat breast conserving surgery reported by Medicare claims 2 No 2 No P <.1 * P =.83 * % Years B. Subsequent mastectomy reported by Medicare claims 2 % % ,364 6,36 5 P <.1* 2 2,11 6,9 4 1,327 4,496 Years ,53 No , Years Years 8 1 Number at Risk: No RT 2,364 2,6 1, ,364 1,965 1, RT 6,36 6,9 4,494 2,5 1, ,36 5,84 4,17 2, ARTICLES Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26

5 Table 2. Outcomes in the Surveillance, Epidemiology, and End Results (SEER) Medicare cohort and in the C9343 trial therapy, No. events per 1 persons (95% CI) No radiation therapy, No. events per 1 persons (95% CI) Absolute risk reduction, No. events per 1 persons (95% CI) * Outcome 5 years 8 years 5 years 8 years P 5 years 8 years SEER Medicare cohort Second ipsilateral breast cancer.28 (.13 to.43).86 (.51 to 1.2) 2.5 (1.7 to 3.2) 3.6 (2.4 to 8.4) < (1.4 to 2.9) 2.7 (1.5 to 4.) Subsequent mastectomy.93 (.65 to 1.2) 2.2 (1.6 to 2.7) 3.9 (3. to 4.8) 6.6 (5. to 8.3) <.1 3. (2. to 3.9) 4.5 (2.7 to 6.3) Any second breast cancer event 1.1 (.79 to 1.4) 2.3 (1.7 to 2.9) 5.1 (4.1 to 6.2) 8. (6.2 to 9.8) <.1 4. (2.9 to 5.1) 5.7 (3.8 to 7.6) C9343 trial Mastectomy 1 ( to 2) 2 ( to 3).15 1 Local regional recurrence 1 ( to 2) 4 (2 to 7) <.1 3 * All numbers have been rounded to two significant digits. As a result, the absolute risk reduction may not precisely equal the difference in event risks. P value from two-sided log-rank tests, indicating comparison of the survival curves for radiation therapy versus no radiation therapy. Patients were at risk for all outcomes beginning 9 months after diagnosis. Defined as a second ipsilateral, pathologically confirmed, invasive breast cancer reported by SEER. Defined as a subsequent mastectomy appearing in Medicare claims. De fined as a second ipsilateral, pathologically confirmed, invasive breast cancer reported by SEER or a subsequent mastectomy appearing in Medicare claims. The Cancer and Leukemia Group B C9343 trial included women aged 7 years with clinical stage I, estrogen receptor-positive or unknown breast cancer treated with breast-conserving surgery and 5 years of tamoxifen. Patients were randomly assigned to breast radiation therapy versus no radiation therapy (5 ). inclusion of patients who experienced a contralateral breast cancer or by inclusion of treatments received within 6 or 12 months of diagnosis. Number Needed to Treat Among women who met the entry criteria for trial C9343, the absolute risk reduction associated with radiation therapy increased over time. At 5 years, the absolute risk reduction was 4. events (95% CI = 2.9 to 5.1) per 1 patients (i.e., from 5.1 events without radiation therapy to 1.1 events with radiation therapy). At 8 years, the absolute risk reduction increased to 5.7 events (95% CI = 3.8 to 7.6) per 1 patients (i.e., from 8. events without radiation therapy to 2.3 events with radiation therapy) ( Table 2 ). Hence, women who lived longer were more likely to benefit from radiation therapy. For example, among those aged 7 74 years without comorbid illness, 8-year survival was 84% (95% CI = 83 to 86). This subgroup was most likely to benefit from radiation therapy, with an adjusted NNT of 21 patients (95% CI = 16 to 31) ( Table 5 ). Other patient groups who were more likely to benefit from radiation therapy included those aged years without comorbid illness (adjusted NNT = 22 patients, 95% CI = 17 to 33), those aged 7 74 years with mild comorbidity (adjusted NNT = 24 patients, 95% CI = 18 to 36), those aged years with mild comorbidity (adjusted NNT = 28 patients, 95% CI = 21 to 42), and those aged 8 84 years without comorbid illness (adjusted NNT = 29 patients, 95% CI = 22 to 43). Patients aged 85 years and older with moderate to severe comorbidity were least likely to benefit from radiation therapy, with an 8-year survival of 14% (95% CI = 7.2 to 21) and an adjusted NNT of 125 patients (95% CI = 94 to 185) ( Table 5 ). Comparison Cohort The comparison cohort comprised 5551 women who would have been ineligible for the C9343 trial because of being aged years, having a tumor of cm, or having an ER-negative tumor. When compared with the cohort eligible for trial C9343, the adjusted risk of a second breast cancer event was larger for the cohort ineligible for trial C9343 (HR = 1.5, 95% CI = 1.2 to 1.8; P <.1). However, the interaction term of radiation therapy with eligibility status was not statistically sig - ni ficant ( P =.17), indicating that the relative benefit associated with radiation therapy was similar for the eligible and ineligible cohorts. Although the relative benefit associated with radiation therapy was similar for both cohorts, the absolute benefit associated with radiation therapy was greater for the ineligible cohort. At 8 years, the absolute risk reduction was 7.7 events per 1 patients (95% CI = 3.9 to 11) for the ineligible cohort (i.e., from 11.7 events without radiation therapy to 4. events with radiation therapy) and 5.7 events per 1 patients (95% CI = 3.8 to 7.6) for the eligible cohort (i.e., from 8. events without radiation therapy to 2.3 events with radiation therapy). The benefit of radiation therapy was present for all subgroups of patients ineligible for trial C9343. For example, among patients aged years, the absolute risk reduction was 1 events per 1 patients at 8 years (95% CI = 2.8 to 17) (i.e., from 13.3 events without radiation to 3.3 events with radiation) ( Fig. 2 ). Similarly, among patients with a tumor of cm, the absolute risk reduction was 7.3 events per 1 patients at 8 years (95% CI = 1.2 to 13) (i.e., from 1.9 events without radiation to 3.6 events with radiation) ( Fig. 2 ). Finally, among patients with ER-negative tumors, the absolute risk reduction was 6.7 events per 1 patients at 8 years (95% CI =.2 to 16) (i.e., from 11.3 events without radiation to 4.6 events with radiation) (Fig. 2 ). D ISCUSSION In this community-based cohort of older women with early breast cancer, the 5-year risk of a second breast cancer event, defined as a second ipsilateral breast cancer reported by SEER and/or subsequent mastectomy reported by Medicare, was 5.1% (95% CI = 4.1 to 6.2) in patients treated without radiation therapy and 1.1% (95% CI =.79 to 1.4) in patients treated with radiation therapy. These results are remarkably consistent with those of the Cancer and Leukemia Group B C9343 randomized trial, in which the risk of local regional relapse was 4% (95% CI = 2 to 7) without radiation therapy and 1% (95% CI = to 2) with radiation therapy (Table 2) ( 5 ). However, our results also indicate that Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26 ARTICLES 685

6 Table 3. Predictors of a second breast cancer event * Covariate Hazard ratio (95% confidence interval) P Breast radiation therapy.19 (.14 to.28) <.1 Patient characteristics Age.97 (.95 to 1.).8 Race White 1. (referent) Black 2.22 (1.18 to 4.21).1 Asian/Pacific Islander 1.97 (.78 to 4.99).15 White Hispanic 1.55 (.64 to 3.78).33 Marital status Married 1. (referent) Widowed.69 (.49 to.97).3 Single.93 (.52 to 1.67).81 Separated or divorced.59 (.27 to 1.29).19 Adults in census tract with high school education % 7.5% 1. (referent) 7.6% 12.8% 1.22 (.74 to 2.2) % 2% 1.81 (1.7 to 3.4).3 2.1% 1% 1.3 (.71 to 2.53).36 Charlson comorbidity score 1. (referent) (.72 to 1.49) (.68 to 1.92).61 Tumor characteristics Tumor size (mm).99 (.96 to 1.2).37 Tumor histology Ductal 1. (referent) Lobular.89 (.51 to 1.56).67 Other.87 (.61 to 1.25).45 Estrogen receptor Positive 1. (referent) Borderline 4.6 (.97 to 21.94).6 Unknown 1.67 (.47 to 5.95).42 Progesterone receptor Positive 1. (referent) Borderline.44 (.57 to 3.43).44 Negative 1.49 (1. to 2.22).5 Unknown.6 (.17 to 2.12).43 Treatment characteristics Pathologic axillary staging 1.5 (.74 to 1.49).79 Treatment with chemotherapy.99 (.36 to 2.73).98 Hospital characteristics Surgery performed at teaching hospital.82 (.58 to 1.15).24 * Combined outcome of second ipsilateral breast cancer reported by Surveillance, Epidemiology, and End Results (SEER) data and/or subsequent mastectomy reported by Medicare claims. Patients were at risk for the outcome beginning 9 months after diagnosis. Model is adjusted for all patient, tumor, treatment, and hospital characteristics that were statistically significant in unadjusted analysis at a P.25. The following variables are included in the model but were not reported because they were not statistically significant and had limited clinical relevance: distance to radiation therapy facility, urban/rural status, median income of census tract, and total number of provider visits in the year preceding diagnosis. Hazard ratios for unknown values are not shown in the table and were not statistically significant for any covariate. Model is stratified by SEER geographic site and year of diagnosis. Adjusted P values from the Cox model. All statistical tests were two-sided. Medicare claims reported between 12 months and 1 month before diagnosis were used to identify the presence or absence of 18 noncancer chronic diseases. A weight proportional to the risk of death was assigned to each of these 18 chronic diseases. The Charlson comorbidity score is a sum of these weights (24 26 ). important, readily identifiable characteristics can determine which patients are most likely to benefit from radiation therapy. For example, radiation therapy was associated with a large benefit among women with lobular histology. In addition, healthy women aged 7 79 years were more likely to experience the benefit associated with ra diation therapy, with a NNT of 21 to 22 patients. This value is comparable to those of other accepted interventions, such as antihypertensive therapy, in which 21 women require 1 years of treatment to prevent one coronary heart disease event ( 35 ). In contrast, patients of advanced age or with moderate to severe comorbid illness were less likely to experience the benefit associated with radiation therapy. Identification of patients unlikely to benefit from radiation therapy will serve to minimize the number of older women unnecessarily exposed to the morbidity (5 ), cost (6 ), and inconvenience of breast radiation therapy. Paradoxically, identification of patients unlikely to benefit from radiation therapy may also serve to reduce utilization of mastectomy. In this study, 41% of patients were excluded solely because of treatment with mastectomy. Traditionally, mastectomy has been recommended for older women with early breast cancer because it will obviate the need for breast radiation therapy. However, among women unlikely to benefit from radiation therapy, this indication for mastectomy is no longer relevant; such patients should therefore receive conservative surgery without radiation therapy. Refining the appropriate indications for radiation therapy in older women is a matter of substantial public health importance. In 25, more than 24 women in the United States were diagnosed with a breast cancer that met the inclusion criteria for trial C9343 (36,37 ) ( Appendix B ). By 23, when the elderly population is expected to double (38 ), this number will approach 5. Given the growing number of elderly cancer patients, future studies should consider the benefits of adjuvant therapy in light of the competing risk of death from other causes. Although this approach has not previously been adopted in studying cancer therapy, earlier studies used this approach to estimate the benefits of preventive interventions such as antihyperlipidemic therapy (39 ) or cancer screening (34 ). The current study confirms that being younger than 7 years old, having a tumor of more than 2 cm in diameter, or having an ER-negative tumor (i.e., exclusion criteria outlined in trial C9343) define a patient group at higher risk for a second breast cancer event. In addition, the absolute benefit associated with radiation therapy was greater for women who would have been excluded from trial C9343. Therefore, although the con clusions of clinical trials are often extrapolated to other patient groups (4 ), radiation therapy should remain a standard of care for women who do not meet the inclusion criteria for trial C9343. One important contrast between the current study and trial C9343 concerns the role of radiation therapy in preventing subsequent mastectomy. Although radiation therapy, compared with no radiation therapy, was associated with a lower risk of subsequent mastectomy in this population-based cohort, radiation therapy did not lower the risk of subsequent mastectomy in trial C9343. These results suggest that women who do not receive radiation therapy and who subsequently develop an in-breast recurrence may be more likely to undergo mastectomy in the community setting than in the clinical trial setting. Our study has several limitations. Although treatment with tamoxifen may reduce the risk of local relapse ( 41, 42 ), information regarding tamoxifen prescription and compliance are not available in the SEER Medicare data. During the study period, however, tamoxifen was considered the standard of care ( 43 ) and was prescribed for nearly 9% of older women in the community setting ( 44 ). Further, in prior studies, rates of tamoxifen pres cription and compliance did not appear to differ substantially between patients treated with and without breast 686 ARTICLES Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26

7 Table 4. Unadjusted and adjusted associations between radiation therapy and outcomes Effect size (HR) associated with radiation therapy* Outcome Unadjusted (95% CI) Adjusted for age only (95% CI) Fully adjusted (95% CI) Second ipsilateral invasive breast cancer.19 (.13 to.3).16 (.1 to.26).13 (.7 to.23) Subsequent mastectomy.26 (.19 to.36).22 (.16 to.31).21 (.14 to.31) Any second breast cancer event.23 (.18 to.31).19 (.14 to.26).19 (.14 to.28) * HR = hazard ratio; CI = confidence interval. Defined as a second ipsilateral, invasive, pathologically confirmed breast cancer reported by Surveillance, Epidemiology, and End Results (SEER) data or a billing code for mastectomy reported in Medicare claims. Defined as a billing code for mastectomy appearing in Medicare claims. Defined as a second ipsilateral, invasive, pathologically-confirmed breast cancer reported by SEER data or a billing code for mastectomy reported in Medicare claims or a billing code for mastectomy appearing in Medicare claims. radiation therapy, indicating that the absence of tamoxifen data should not introduce a large directional bias ( 44, 45 ). In addition, the similar event risks in the no-radiation group of the current study and the tamoxifen-only arm of trial C9343 provide reassurance that the absence of tamoxifen data should not introduce a substantial bias. Margin status is also not available in SEER Medicare data. However, these data would probably bias our results toward the null because patients with involved margins may be more likely to receive breast radiation therapy ( 46 ). Finally, the outcome of a second ipsilateral breast cancer reported by SEER does not capture all in-breast recurrences; therefore, subsequent mastectomy as reported by Medicare claims was used as a marker of in-breast recurrence. Although Medicare claims for mastectomy are highly accurate ( 15 ), laterality of mastectomy is infrequently reported. Thus, despite the exclusion of patients with contralateral breast cancers reported by SEER, it remains possible that a fraction of mastectomies identified in Medicare claims were performed on the contralateral breast. Although randomized trials are generally considered to produce the highest level of evidence in clinical research, a recent investigation (47 ) asserts that carefully designed observational cohort studies may provide estimates of treatment effects that are highly reproducible and valid. For example, the effect size of breast radiation therapy in this observational study (HR =.19, Table 5. Number needed to treat (NNT) to prevent one second breast cancer event * Age Comorbidity score No. 8-year survival (95% CI), % Adjusted NNT (95% CI) 7 74 y (83 to 86) 21 (16 to 31) (68 to 76) 24 (18 to 36) (4 to 55) 37 (28 to 55) y (76 to 81) 22 (17 to 33) (58 to 67) 28 (21 to 42) (36 to 51) 41 (31 to 6) 8 84 y (57 to 64) 29 (22 to 43) (4 to 53) 38 (28 to 56) (21 to 36) 61 (46 to 9) 85 y (29 to 38) 53 (4 to 78) (13 to 24) 97 (73 to 143) (7.2 to 21) 125 (94 to 185) * Combined outcome of second ipsilateral breast cancer reported by Surveillance, Epidemiology, and End Results data and/or subsequent mastectomy reported by Medicare claims. CI = confidence interval. A total of 258 patients with unknown Charlson comorbidity score were excluded from this table. Expected survival calculated with Kaplan Meier method. At 8 years of follow up, the unadjusted NNT was 18 (95% CI = 13 to 26). 95% CI =.14 to.28) is consistent in direction and magnitude with those in a recent meta-analysis conducted on randomized clinical trials of breast radiation therapy (HR =.33, 95% CI =.29 to.38) (2 ). This consistency strengthens the validity of our study conclusions and underscores the importance of observational data when evaluating the role of radiation therapy in the general population. In the general population of patients aged 7 years or older with small, ER-positive (or of unknown ER status), lymph node negative breast cancer, radiation therapy after breast- conserving surgery is associated with a lower risk of a second ipsilateral breast cancer and subsequent mastectomy. For those with lobular histology and for those aged 7 79 years with minimal comorbidity, the benefit associated with radiation therapy is similar to that of other accepted medical interventions. For women who do not meet the strict age, size, and hormone receptor criteria outlined for trial C9343, the absolute benefit of radiation therapy is substantial, and radiation therapy should remain the standard of care. Event Risk per 1 Persons No C9343 Eligible Age Size cm ER Negative C9343 Ineligible N=8,724 N=2,245 N=1,489 N=922 Fig. 2. Risk of a second breast cancer event among patients eligible and ineligible for trial C9343. Eight-year event risk without and with radiation therapy is reported for the four mutually exclusive patient groups, including patients eligible for trial C9343, patients ineligible for trial C9343 because of age years, patients ineligible for trial C9343 because of tumor size cm, and patients ineligible for trial C9343 because of estrogen receptor (ER) negative status. Patients in each of the ineligible groups, including age years, tumor size cm, and ER-negative status, meet all the other criteria for trial C9343. Error bars = 95% confidence intervals. All risks are rounded to two significant digits. Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26 ARTICLES 687

8 Appendix A. Variables and Their Definitions Variable Data source Definition Patient characteristics Age Medicare Age at diagnosis in years was entered continuously. Date of birth and date of diagnosis were assumed to occur on the 15th of the month, because only month and year were available for these variables. Race SEER White, black, white Hispanic, Asian/Pacific Islander, and other Year of diagnosis SEER Marital status SEER Widowed, married, single, separated/divorced, or unknown Geographic residence SEER SEER geographic region Distance to nearest radiotherapy facility Medicare and American Board of Radiology Distance from patient s zip code of residence to the zip code of the nearest radiation therapy facility. therapy facility zip codes were compiled from the master list for 199, 1994, and 1998, which is maintained by the American Board of Radiology. Distance was calculated with ZIPFind Deluxe version 5. (Xionetic Technologies, Inc., Bozeman, MT). For the multivariable model, distance was dichotomized at 3 miles because the probability of receiving radiation therapy decreased dramatically at this cutpoint and then remained stable for greater distances. Zip codes were unavailable for 94 patients. Urban vs. rural residence SEER Big metropolitan, metropolitan, urban, less urban, or rural Income Census Median income for the patient s census tract and zip code was available from a linkage with the 199 and 2 census data (23). If census tract data were unavailable, zip code level data were used. Median income was not available for 88 patients, who were treated as a separate stratum. The logarithm of median income was entered into the final model. Education Census Percentage of adults with <12 years of education from the 199 and 2 censuses (23). Data were missing for 185 patients, who were treated as a separate stratum. Quartiles of education were entered into the final model. Comorbidity index Medicare * Using Part A and Part B Medicare claims spanning a prediagnosis interval of 12 months to 1 month, a modified Charlson comorbidity (24) index was calculated by use of the adaptations proposed by Deyo et al. (25) and Romano et al. (26). To enhance specificity, Part B diagnosis codes were included only if they appeared either more than once over a period of more than 3 days or also in Part A claims (27,28). Those without prior claims (labeled as unknown) were treated as a separate stratum in the adjusted analysis. Use of physician services Medicare No. of claims for physician visits on separate days during a prediagnosis interval of 12 months to 1 month. Quartiles were entered into the final model. Tumor characteristics Size SEER Entered continuously into final model Location in breast SEER Inner quadrant, outer quadrant or central, overlapping, and unknown (29) Grade SEER Well differentiated, moderately differentiated, poorly or undifferentiated, or unknown Histology SEER Histology was coded according to the International Classification of Diseases for Oncology, 2nd ed. (3) as follows: ductal (85 or 8521), lobular (852), or other. Estrogen receptor status SEER Positive, borderline, or unknown Progesterone receptor status SEER Positive, borderline, negative, or unknown Laterality SEER Right or left Treatment characteristics Surgery SEER and Medicare From site-specific surgery variable in SEER and from Medicare claims: breast conserving surgery ICD-9 Procedure Codes 85.2, 85.21, 85.22, 85.23, or 85.25; CPT Codes 1911, 1912, 19125, 1916, or 19162) and mastectomy (ICD-9 Procedure Codes 85.41, 85.42, 85.43, 85.44, 85.45, 85.46, 85.47, or 85.48; CPT Codes 1918, 19182, 192, 1922, or 1924) (15). The most extensive surgical procedure reported by SEER or Medicare within 9 months of diagnosis was considered the definitive surgery. therapy SEER and Medicare From radiation therapy variable in SEER and from Medicare claims: ICD-9 Procedure Codes or 92.29; ICD-9 Diagnosis Codes V58., V66.1, or V67.1; CPT Codes or (14,2); and Revenue Center Codes 33 or 333. Patients were considered to have received radiation therapy if either SEER or Medicare reported treatment with radiation therapy within 9 months of diagnosis. Chemotherapy Medicare ICD-9 Procedure Code 99.25; ICD-9 Diagnosis Codes V58.1, V66.2, or V67.2; CPT Codes ; HCPCS Codes J9 J9999 or Q83 Q85; and Revenue Center Codes 331, 332, or 335 (21,22). Patients were considered to have received chemotherapy if any of these codes appeared within 9 months of diagnosis. Method of axillary lymph node assessment SEER Pathologic assessment if at least one lymph node was sampled, otherwise clinical assessment. Hospital characteristics Teaching status Hospital file Teaching status (yes or no) was self-reported in the Hospital Cost Report Information Systems files available for 1996, 1998, and 2 (31). (Appendix continues) 688 ARTICLES Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26

9 Appendix A (continued). Variable Data source Definition Outcomes Second ipsilateral breast cancer SEER Any pathologically confirmed, invasive, ipsilateral breast cancer reported by SEER more than 9 months after diagnosis Subsequent mastectomy Medicare The appearance of any billing claim for mastectomy more than 9 months after diagnosis Second breast cancer event SEER-Medicare The appearance of a second ipsilateral breast cancer reported by SEER and/or a subsequent mastectomy reported by Medicare Repeat breast-conserving surgery Medicare The appearance of any billing claim for breast-conserving surgery more than 9 months after diagnosis * Medicare billing claims included the Medicare Provider and Analysis Review File for inpatient hospitalizations, the Outpatient File (outpatient claims from hospitals), and the Carrier Claims File (physician service claims). To identify billing codes, we used the International Classification of Disease, 9th ed. (ICD-9) (16,17 ) for hospital-related diagnoses and procedures, Current Procedural Terminology (CPT) (18,19 ) for physician services and procedures, and the Health Care Procedure Coding System (HCPCS) J and Q codes for chemotherapy. Medicare claims were available through 22. Surveillance, Epidemiology, and End Results (SEER) data are available in the Population Entitlement and Diagnosis Summary File through To increase follow-up, cases of breast cancer reported by SEER between 2 and 22 were linked to the SEER Medicare data by use of a crosswalk file. Appendix B. Incidence of Breast Cancers Meeting Criteria of Trial C9343 in 25 A. Between 2 and 22, women were diagnosed with breast cancer in a Surveillance, Epidemiology, and End Results (SEER) registry. Of these women, (3.5%) were 7 years or older (37 ). B. For 25, the predicted incidence of breast cancer in the United States is (36 ). C. Given points A and B, the predicted 25 incidence of breast cancer in the United States among women aged 7 years and older is 64 9 ( ). D. In our study, the fraction of women aged 7 years and older who met all criteria for trial C9343 was 37.6% (i.e., /6 717). (The figure includes women treated with conservative surgery, mastectomy, and biopsy. Patients treated with mastectomy or biopsy were included in this calculation, because the goal was to determine the number of future patients to whom the study results may apply.) E. The predicted 25 incidence of breast cancer among individuals meeting entry criteria for trial C9343 is 24 4 ( ). R EFERENCES (1) Favourable and unfavourable effects on long-term survival of radiotherapy for early breast cancer: an overview of the randomised trials: Early Breast Cancer Trialists Collaborative Group. Lancet 2 ;355 : (2) Vinh-Hung V, Verschraegen C. Breast-conserving surgery with or without radiotherapy: pooled-analysis for risks of ipsilateral breast tumor recurrence and mortality. J Natl Cancer Inst 24 ;96 : (3) Fisher B, Redmond C, Poisson R, Margolese R, Wolmark N, Wickerham L, et al. Eight-year results of a randomized clinical trial comparing total mastectomy and lumpectomy with or without irradiation in the treatment of breast cancer. N Engl J Med 1989 ;32 : (4) Fisher B, Bauer M, Margolese R, Poisson R, Pilch Y, Redmond C, et al. Five-year results of a randomized clinical trial comparing total mastectomy and segmental mastectomy with or without radiation in the treatment of breast cancer. N Engl J Med 1985 ;312 : (5) Hughes KS, Schnaper LA, Berry D, Cirrincione C, McCormick B, Shank B, et al. Lumpectomy plus tamoxifen with or without irradiation in women 7 years of age or older with early breast cancer. N Engl J Med 24 ; 351 : (6) Hayman JA, Hillner BE, Harris JR, Weeks JC. Cost-effectiveness of routine radiation therapy following conservative surgery for early-stage breast cancer. J Clin Oncol 1998 ;16 : (7) Carlson RW, Anderson BO, Burstein HJ, Cox CE, Edge SB, Farrar WB, et al. Breast cancer. J Natl Compr Canc Netw 25 ;3 : (8) Bijker N, Peterse JL, Fentiman IS, Julien JP, Hart AA, Avril A, et al. Effects of patient selection on the applicability of results from a randomised clinical trial (EORTC 1853) investigating breast-conserving therapy for DCIS. Br J Cancer 22 ;87 : (9) Murthy VH, Krumholz HM, Gross CP. Participation in cancer clinical trials: race-, sex-, and age-based disparities. JAMA 24 ;291 : (1) Britton A, McKee M, Black N, McPherson K, Sanderson C, Bain C. Threats to applicability of randomised trials: exclusions and selective participation. J Health Serv Res Policy 1999 ;4 : (11) Gross CP, Mallory R, Heiat A, Krumholz HM. Reporting the recruitment process in clinical trials: who are these patients and how did they get there? Ann Intern Med 22 ;137 :1 6. (12) Whelan TJ. Use of conventional radiation therapy as part of breast- conserving treatment. J Clin Oncol 25 ;23 : (13) Warren JL, Klabunde CN, Schrag D, Bach PB, Riley GF. Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 22 ;4 (8 Suppl):IV (14) Virnig BA, Warren JL, Cooper GS, Klabunde CN, Schussler N, Freeman J. Studying radiation therapy using SEER-Medicare-linked data. Med Care 22 ;4 (8 Suppl):IV (15) Du X, Freeman JL, Warren JL, Nattinger AB, Zhang D, Goodwin JS. Accuracy and completeness of Medicare claims data for surgical treatment of breast cancer. Med Care 2 ;38 : (16) ICD-9-CM: International classification of diseases, 9th Rev, 3rd ed. Los Angeles (CA): Practice Management Information Corporation; (17) Buck CJ. 22 ICD-9-CM, Vols 1, 2, and 3; HCPCS Level II. Philadelphia (PA): WB Saunders Company; 22. (18) Kirschner CG, Edwards NK, May DM, McNamara MR, Metge PJ, Pollack A, et al. Physicians current procedural terminology. 4th ed. Chicago (IL): American Medical Association; (19) Anderson CA, Beebe M, Dalton JA, Duffy C, Evans D, Glenn RL, et al. Current procedural terminology CPT 22. Chicago (IL): American Medical Association; 22. (2) 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 1999 ;52 : (21) Warren JL, Harlan LC, Fahey A, Virnig BA, Freeman JL, Klabunde CN, et al. Utility of the SEER-Medicare data to identify chemotherapy use. Med Care 22 ;4 (8 Suppl):IV (22) Du XL, Chan W, Giordano S, Geraci JM, Delclos GL, Burau K, et al. Variation in modes of chemotherapy administration for breast carcinoma and association with hospitalization for chemotherapy-related toxicity. Cancer 25 ;14 : (23) Bach PB, Guadagnoli E, Schrag D, Schussler N, Warren JL. Patient demographic and socioeconomic characteristics in the SEER-Medicare database applications and limitations. Med Care 22 ;4 (8 Suppl):IV (24) Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987 ;4 : (25) Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992 ; 45 : Journal of the National Cancer Institute, Vol. 98, No. 1, May 17, 26 ARTICLES 689