Association Between Treatment With Brachytherapy vs Whole-Breast Irradiation and Subsequent Mastectomy, Complications, and Survival Among Older Women With Invasive Breast Cancer FREE

Brachytherapy is a means of delivering radiation using an implanted radioactive source and has been used to treat various malignancies. Recently, breast brachytherapy has emerged as an alternative to whole-breast irradiation (WBI), the standard radiation treatment after lumpectomy.^1- 2 Compared with WBI, brachytherapy irradiates less breast tissue and requires a much shorter course of treatment.

As many as 10% of older women with breast cancer are now treated with brachytherapy,^3- 5 and at least 50 000 have been treated to date in the United States.⁶ Controversy persists, however, regarding appropriateness of widespread adoption of brachytherapy as the sole radiation treatment modality following lumpectomy.^1,5 Although WBI is proven to decrease risk of local recurrence after lumpectomy and thereby optimize the long-term likelihood of breast preservation,^7- 8 a paucity of mature randomized data directly comparing brachytherapy with standard WBI exists.^1- 2 Given the comparatively smaller volume of breast tissue treated with brachytherapy, concerns persist that brachytherapy may be insufficient for full tumor eradication, potentially leading to excess local recurrences requiring salvage mastectomy.⁹ Additionally, excess risks of postoperative and postradiation complications have not been well quantified.

Long-term randomized trial results comparing brachytherapy with WBI remain years from maturing.¹⁰ In the interim, a comparison of brachytherapy vs WBI in the nonrandomized cohort setting is warranted to help guide ongoing treatment decisions. Accordingly, in a comprehensive cohort of older Medicare patients diagnosed with invasive breast cancer, we sought to compare breast brachytherapy vs WBI with respect to the likelihood of long-term breast preservation, complications, and survival.

The national Medicare data set includes claims data of all health care services for all Medicare beneficiaries in the United States, including institutional (inpatient and outpatient) and provider final action claims.¹¹ This data set includes 100% of Medicare beneficiaries throughout the United States and is thus distinct from the Surveillance, Epidemiology, and End Results (SEER)–Medicare database, which captures only 26% of beneficiaries.¹² This study was granted exempt status by the University of Texas M D Anderson institutional review board.

We identified 119 576 women (excluding entitlement solely due to end-stage renal disease or disability), who were 67 years or older, with a diagnosis of invasive breast cancer from 2003-2007 and who were treated with lumpectomy followed by radiation therapy (eTable 1). Cases were identified using a previously validated algorithm^13- 14 derived from Medicare claims. The algorithm has a published sensitivity of 82% to 87% and specificity exceeding 99% for all early-stage breast cancer inclusive of any treatment or no treatment.^13- 14 We modified this algorithm to further enhance specificity by selecting only cases of breast cancer treated with lumpectomy followed by radiation therapy for a coded diagnosis of breast cancer. Date of surgery was considered the diagnosis date. Our data provided a minimum of a 12-month history of claims before diagnosis to determine patients' clinical history.

We excluded patients with only 1 International Classification of Diseases, Ninth Revision (ICD-9) diagnosis code indicating breast cancer (n<11); metastatic disease at diagnosis (>3990); a history of breast cancer (n = 10 363); and noncontinuous Medicare Part A and B or health maintenance organization (HMO) coverage within 12 months before and after diagnosis (n = 7123), which yielded 98 083 eligible patients. Of these, we excluded patients for whom actual treatment with radiation could not be confirmed using a stringent definition for receipt of radiation (n = 1546), patients treated with both external beam radiation and brachytherapy (n = 775), and patients who did not satisfy our a priori definition of WBI, which required treatment with at least 11 fractions of radiation (n = 3027) to exclude external beam partial breast or palliative WBI, yielding a final analytic cohort of 92 735.

Patients were classified as treated with WBI vs breast brachytherapy if claims indicating treatment were recorded within 12 months of diagnosis. Single-entry catheter-based brachytherapy was also identified (eTable 2).

Subsequent mastectomy was defined as a claim for mastectomy after initiation of radiation therapy. This outcome indicates failure of radiation to achieve its intended clinical purpose, which is to optimize the likelihood of long-term breast preservation. Prior studies validate that the association of radiation with long-term breast preservation can be assessed using this claims-based outcome.^15- 19 Our data set does not provide details on underlying reasons for mastectomy, which therefore could have represented treatment for local recurrence or nonmalignant complication. Time to subsequent mastectomy was calculated from date of first radiation. Overall survival from date of diagnosis was also compared.

Infectious postoperative complications were determined by claims indicating soft tissue infection that may have affected the breast (eg, mastitis and breast abscess). Noninfectious postoperative complications included claims for postoperative shock, hemorrhage, hematoma, seroma, persistent postoperative fistula, or nonhealing surgical wound. Post-radiation–associated complications were determined by claims indicating rib fracture, fat necrosis, breast pain, and pneumonitis (eTable 3). Time to postradiation complications was calculated from the date of the first radiation.

Overall survival was calculated from the date of diagnosis to the date of death. Cause of death is not available in this data set.

Demographic covariates included age, year of diagnosis, and race. Geographic region was based on zip code and US Census Bureau region definitions. Severity of comorbid disease was based on modified Charlson comorbidity score derived from claims during the 12 months preceding diagnosis,^20- 21 and grouped as 0 for no comorbidity, 1 for mild, 2 for moderate, or 3 or higher for severe.

Treatment-related variables included axillary lymph node involvement; axillary lymph node surgery including sentinel node biopsy, lymph node dissection, or both; and receipt of chemotherapy (eTable 2). Socioeconomic covariates derived from the Area Resource File linked by county included rural and urban status, percentage of population by county who were living in poverty, median household income, educational level (percentage with college education or more), density of surgeons, and density of radiation oncologists.^22- 23 For variables with information available in multiple years, information from the year closest to diagnosis year was used.

Analyses were conducted using SAS statistical software version 9.2 (SAS Institute Inc) and assumed a 2-tailed α = .05. We calculated the percentage of women treated with brachytherapy by year and evaluated time trends using the Cochran Armitage test. Univariate predictors of brachytherapy use were tested using the Pearson χ² test for categorical variables and the Wilcoxon rank sum test for continuous variables.

To address our first objective, cumulative incidence of subsequent mastectomy was determined using the Kaplan-Meier method. Patients were censored at the earliest of any of the following: loss of part A or B coverage, conversion to HMO coverage, death, or December 31, 2008, the end of the study period. Sensitivity analysis determined whether event risks remained stable after accounting for competing risks.²⁴ Multivariate proportional hazards models tested the association between radiation treatment (brachytherapy vs WBI) and subsequent mastectomy and adjusted for covariates identified a priori based on clinical significance and univariate analyses (P < .25). For all analyses, missing or unknown covariate values were treated as a separate dummy variable in multivariate models except when the number of missing individuals was too small to permit reliable parameter estimates. In such cases, the number of missing individuals excluded is clearly specified in all relevant tables. Proportional hazards assumption was confirmed by inspection of log (−log [survival]) curves. Interaction terms between radiation treatment and age, race, comorbidity, axillary lymph node involvement, receipt of chemotherapy, geographic region, and year of treatment were tested in preplanned analyses to identify any subset of patients with particularly high or low risk of subsequent mastectomy. Sensitivity analysis tested whether effect sizes remained stable if patients receiving at least 1 fraction of external beam radiation were included in the WBI group. Numbers needed to harm were calculated using the Wald method.²⁵

A subsidiary analysis was conducted using propensity score matching. The probability of receiving brachytherapy was determined using a logistic model adjusted for age, race, comorbidity score, region, urban or rural status, axillary lymph node involvement, axillary lymph node surgery, chemotherapy, year of diagnosis, and the following county-level variables divided into quartiles: median household income, percentage living in poverty, percentage with college or more education, surgeon density, and radiation oncologist density. Patients with missing county-level variables were excluded. Goodness of fit was assessed using the Hosmer and Lemeshow test. Patients treated with brachytherapy vs WBI were matched 1:1 using a greedy 8- to 1- digit matching algorithm.²⁶ Differences in covariate strata by treatment group in the matched cohort were assessed using McNemar χ² test and Wilcoxon signed rank sum test for paired data. A multivariate proportional hazards model, stratified by matched pair,²⁷ tested the association of brachytherapy vs WBI with subsequent mastectomy with and without adjustment for residual imbalanced covariates.

To address our next objective, the univariate association of radiation treatment with frequency of infectious and noninfectious postoperative complications by 30-day increment after diagnosis was compared using a generalized estimating equation. Summary frequencies of these events within 1 year of diagnosis by radiation treatment were compared using the χ² test. Adjusted odds of these events within 1 year of diagnosis were determined using multivariable logistic models with initial candidate covariates selected based on a priori clinical significance or univariate P < .25, and iterative model refinement to minimize collinearity. Goodness of fit was assessed using the method of Hosmer and Lemeshow. Cumulative incidences of postradiation complications (rib fracture, fat necrosis, breast pain, and pneumonitis) were compared using the log-rank test. Sensitivity analysis tested the effect of a more stringent postradiation complication definition as proposed by Klabunde et al²⁰ which requires diagnoses to appear at least once in Part A data, which are physician-coded, or at least twice separated by an interval of at least 30 days in Part B data, which are administratively coded.

To address our final objective, overall survival by radiation treatment was calculated using the Kaplan-Meier method with unadjusted differences evaluated using the log-rank test. Patients were censored if they had not died by the end of the study period. Multivariate proportional hazards regression tested the effect of radiation treatment on survival and adjusted for covariates identified a priori based on clinical significance and univariate analyses (P < .25).

Assuming 6952 women treated with brachytherapy and 85 783 treated with WBI, accrual spanning 5 years with 1 additional year of follow-up, a 5-year risk of subsequent mastectomy of 2% in patients treated with WBI, and a true hazard ratio (HR) of 2.0 for women treated with brachytherapy compared with patients treated with WBI, this study was able to reject the null hypothesis that brachytherapy and WBI are associated with an equal risk of subsequent mastectomy with power approaching 1.000. The Type I error probability associated with this test of the null hypothesis is 0.05.²⁸

Of 92 735 women, median follow-up was 3.03 years (interquartile range, 1.90-4.34) and 13 355 (14.40%; 95% confidence interval [CI], 14.18%-14.63%) were censored during the study interval due to loss of fee-for-service coverage. The women were a mean (SD) age of 74.8 (5.5) years and 82 418 (92.11%; 95% CI, 91.93%-92.28%) were white. A total of 72 251 (77.91%; 95% CI, 77.64%-78.18%) received axillary surgery, 13 504(14.56%; 95% CI, 14.34%-14.79%) received chemotherapy, 11 569 (12.48%; 95% CI, 12.26%-12.69%) had axillary nodal involvement at diagnosis, and 6952 (7.50%; 95% CI, 7.33%-7.67%) received breast brachytherapy, with 5324 (76.58%; 95% CI, 75.57%-77.57%) of those treated with single-entry catheter-based approaches.

Use of brachytherapy increased from 674 of 19 449 (3.47%; 95% CI, 3.21%-3.73%) patients diagnosed in 2003 to 2224 of 17 760 (12.52%; 95% CI, 12.04%-13.02%) patients in 2007 (P < .001 for trend; Figure 1). Patients treated with brachytherapy were less likely to have axillary lymph node involvement or to have received chemotherapy but were more likely to have undergone axillary lymph node surgery (P < .001, all comparisons; Table 1)

Breast brachytherapy was associated with a higher risk of subsequent mastectomy, with 5-year cumulative incidence of 3.95% (95% CI, 3.19%-4.88%) in patients treated with brachytherapy vs 2.18% (95% CI, 2.04%-2.33%) in patients treated with WBI (P < .001; Figure 2 and Table 2) On multivariate analysis, brachytherapy remained associated with an increased risk of subsequent mastectomy (HR, 2.19; 95% CI, 1.84-2.61, P < .001; Table 3).

A borderline significant interaction term between radiation treatment type (brachytherapy vs WBI) and axillary lymph node involvement (P = .05) indicated that patients with involved nodes were at particularly high risk of subsequent mastectomy when treated with brachytherapy vs WBI (HR, 5.08; 95% CI, 2.94-8.80; P < .001). Nevertheless, even patients without involved axillary nodes experienced an increased risk of subsequent mastectomy when treated with brachytherapy (HR, 2.09; 95% CI, 1.74-2.51; P < .001). The interaction term between radiation treatment type and single-entry vs non-single–entry catheter was not statistically significant (P = .99). Subsequent mastectomy risk in patients who received single-entry catheter placement vs WBI (HR, 2.19; 95% CI, 1.76-2.73) was similar to the risk in patients who did not receive single entry catheter placement vs WBI (HR, 2.19; 95% CI, 1.66-2.90). Interaction terms for radiation treatment by age, race, comorbidity, receipt of chemotherapy, and geographic region were also not significant (P > .05). The interaction term for radiation therapy treatment by year of diagnosis was not significant (P = .44), indicating that outcomes of brachytherapy failed to improve over time.

Propensity score-based 1:1 matching yielded a cohort of 6939 matched pairs, 99.81% all patients treated with brachytherapy and 8.09% of all patients treated with WBI (Hosmer and Lemeshow P value for logistic model to calculate propensity score, 0.88). Covariates were more evenly balanced in this cohort, but residual imbalance existed for age (P = .02), race (P = .03), axillary lymph node surgery (P = .01), and chemotherapy (P = .02; Table 1). Among the matched pairs, treatment with brachytherapy compared with WBI was associated with an increased risk of subsequent mastectomy (HR, 1.86; 95% CI, 1.37-2.52; P < .001) that persisted after adjustment for imbalanced covariates (HR, 1.87; 95% CI, 1.36-2.58; P < .001; eTable 4).

Breast brachytherapy was associated with a higher risk of infectious and noninfectious postoperative complications, with 1916 of 6952 (27.56%; 95% CI, 26.51%-28.63%) of patients treated with brachytherapy vs 14 518 of 85 783 (16.92%; 95% CI, 16.67%-17.18%) treated with WBI experiencing any complication within 1 year of lumpectomy (P < .001). Specifically, by 1 year, 1126 patients (16.20%; 95% CI, 15.34%-17.08%) treated with brachytherapy experienced skin or soft tissue infection compared with 8860 (10.33% ; 95% CI, 10.13%-10.53%) treated with WBI (adjusted odds ratio [OR], 1.76; 95% CI, 1.64-1.88; P < .001; goodness of fit P = .70; eTable 5). Similarly, by 1 year, 1132 patients (16.25%; 95% CI, 15.39%-17.14%) treated with brachytherapy experienced noninfectious postoperative complications compared with 7721 (9.00%; 95% CI, 8.81%-9.19%) treated with WBI (adjusted OR, 2.03; 95% CI, 1.89-2.17; P < .001; goodness of fit P = .83; eTable 6). When analyzed in 30-day intervals through the first 540 days after diagnosis, patients treated with brachytherapy compared with WBI experienced an increased risk of infectious (OR, 1.82; 95% CI, 1.70-1.95; P < .001) and noninfectious (OR, 1.90; 95% CI, 1.77-2.04; P < .001) postoperative complications (Figure 3).

Brachytherapy was generally associated with higher risk of postradiation complications, with 24.96% (95% CI, 23.29%-26.72%) of patients treated with brachytherapy vs 18.80% (95% CI, 18.44%-19.17%) treated with WBI experiencing any complication within 5 years of radiation (P < .001). Specifically, 5-year cumulative incidence of breast pain was 14.55% (95% CI, 13.39%-15.80%) in patients treated with brachytherapy vs 11.92% (95% CI, 11.63%-12.21%) in patients treated with WBI (P < .001); fat necrosis was 8.26% (95% CI, 7.27%-9.38%) vs 4.05% (95% CI, 3.87%-4.24%; P < .001); and rib fracture was 4.53% (95% CI, 3.63%-5.64%) vs 3.62% (95% CI, 3.44%-3.82%; P ≤ .01). Only 5-year incidence of pneumonitis was higher in patients treated with WBI (0.72%; 95% CI, 0.66%-0.80%) than for those treated with brachytherapy (0.12%; 95% CI, 0.05%-0.26%; P < .001; Table 2).

At 5 years, an absolute 1.77% (95% CI, 1.30%-2.24%) excess mastectomy risk in patients treated with brachytherapy compared with WBI meant that, for every 56 women treated with breast brachytherapy, 1 woman was harmed with unnecessary mastectomy (number needed to harm [NNH], 56; 95% CI, 45-77). At 1 year, an absolute 10.64% (95% CI, 9.57%-11.72%) excess postoperative complication risk in women treated with brachytherapy meant that, for every 9 women treated with brachytherapy, 1 was harmed with an unnecessary postoperative complication (NNH, 9; 95% CI, 8-10). At 5 years, an absolute 6.16% (95% CI; 5.11%-7.21%) excess radiation-associated complication risk in patients treated with brachytherapy meant that, for every 16 women treated with brachytherapy, 1 was harmed with an unnecessary postradiation complication (NNH, 16; 95% CI, 14-20).

At 5 years, overall survival was 87.66% (95% CI, 85.94%-89.18%) among patients treated with brachytherapy compared with 87.04% (95% CI, 86.69%-87.39%) among patients treated with WBI (unadjusted HR, 0.87; [95% CI, 0.78-0.98; P = .02; referent is WBI). This difference did not persist with multivariable adjustment (HR, 0.94; 95% CI, 0.84-1.05; P = .26).

In our comprehensive sample of older women with breast cancer, we found significantly increased risk of subsequent mastectomy associated with breast brachytherapy vs standard WBI. Furthermore, we found excess risks of postoperative and postradiation complications associated with breast brachytherapy. Exploratory subgroup analyses failed to identify any subset of patients demonstrating significant clinical benefit from brachytherapy compared with standard treatment. These results represent, to our knowledge, the first comprehensive, population-based study to directly compare the clinical outcomes associated with use of breast brachytherapy vs standard WBI in older patients. Additional study is required to confirm the validity and generalizability of these findings.

Results additionally suggest that the cumulative absolute number of women experiencing excess harm associated with brachytherapy may grow, proportional to the increasing number of women treated with brachytherapy. Results from our analysis are consistent with published reports^3- 4 confirming rapid and increasing adoption of brachytherapy, with as many as 12.5% of patients in our cohort treated with brachytherapy in 2007.

Our results underscore existing controversy over appropriateness of widespread adoption of breast brachytherapy as the sole radiation treatment modality following lumpectomy because few data are available to quantify benefits and harms of brachytherapy in direct comparison with standard WBI.¹ A nonrandomized brachytherapy registry trial of 1440 women reported 5-year local recurrence risks comparable with subsequent mastectomy risks reported in our analysis, but importantly, the prior study lacked a control group.²⁹ Other conflicting data have been reported, with randomized trials evaluating interstitial brachytherapy³⁰ and intraoperative brachytherapy³¹ reporting similar tumor control rates in comparison with WBI. These trials may have limitations in applicability, however, due to potential important differences in technique and target volume, especially in comparison with brachytherapy practices in the United States during the era captured by our analysis. The Targeted Intraoperative Radiotherapy vs Whole Breast Radiotherapy for Breast Cancer (TARGIT-A) randomized trial reported favorable local control rates among patients receiving a single radiation treatment given intraoperatively followed by selective use of WBI.³¹ Although results of this trial are intriguing, relatively few local recurrence events have been reported to date and follow-up remains relatively short. Our analysis does not specifically distinguish intraoperative radiotherapy, given its rarity during the study era and, thus, is insufficient to evaluate this particular technique. To address persistent questions, the ongoing RTOG 0413/NSABP B-39 trial randomizes patients to WBI vs partial breast irradiation (delivered with brachytherapy or external beam radiation) with accrual likely completing in 2012. Results are highly anticipated, but notably, the cumulative results for 3- to 5-year local recurrence risks as well as long-term complications will still be unavailable for some time.

Given that breast brachytherapy is, at present, offered routinely off-protocol throughout the United States, our study provides critical interim companion data to awaited randomized results³² and may help clinicians and patients quantify the risk-benefit ratio of brachytherapy compared with standard treatment alternatives. Accurate risk-benefit assessment must also consider, on the one hand, the established effectiveness of WBI for achieving local control, with associated 5-year recurrence as low as 0.6% to 0.9% in older patients treated in the modern era³³; on the other hand, the possibility that brachytherapy outcomes might become increasingly favorable, for example, in high-volume centers as physician experience matures or in the context of development of and adherence to patient selection criteria. Evidence suggests that approximately 20% of patients treated with breast brachytherapy during the era spanned by our analysis fell into an “unsuitable” treatment category as defined by American Society of Therapeutic Radiology and Oncology Consensus Panel guidelines,^34- 35 although optimally discriminating selection criteria that identify patients with the lowest local recurrence risk after brachytherapy have not yet been clearly defined.^29,34

Our sample focused on older patients with fee-for-service Medicare coverage. Future studies are required to validate findings in younger patients with other insurance status. Median follow-up was relatively short, with longer follow-up in future studies required to determine whether the relative risk of outcomes is modified by time. Second, our definitions of invasive cancer and radiation were claims based and may be subject to misclassification bias. However, definitions were based on a previously validated algorithm¹³ and sought to maximize specificity. Persistent nondifferential misclassification could have biased results toward the null. Third, our administrative data set was unable to adjust for covariates such as cancer stage, histology, surgical margins, hormone receptor status, endocrine therapy, and radiation dose, with potential residual confounding of our model. Our results require future validation with prospective and randomized studies, with our retrospective cohort data considered complementary to such analyses. However, if more favorable cases were differentially selected to receive brachytherapy,³ such confounding would likely have caused an underestimate of mastectomy risk for brachytherapy patients. We sought to account for residual confounding using propensity score analysis, although ultimately our results will require validation in the prospective setting. Fourth, detection bias may have affected complication outcomes, if, theoretically, assessment and billing for lower-grade complications was more frequent in patients treated with brachytherapy because it is a newer treatment. Nevertheless, higher-grade, severe complications requiring direct medical intervention would likely not have been differentially coded between the 2 groups. Longer-term follow-up is required to fully characterize toxicity profiles.

Finally, the outcome of subsequent mastectomy could have been a marker for local tumor recurrence or treatment-related complications and additional studies with detailed pathologic information regarding patterns of failure are needed. Thus, our claims data are limited in their ability to confirm actual cancer recurrence, even though the outcome of subsequent mastectomy still reflects the clinically relevant goal of breast preservation. Our study alone does not prove causality between brachytherapy treatment and subsequent mastectomy, local tumor recurrence, or complications. Nevertheless, prior studies suggest a high prevalence of clinically occult spread of breast cancer to other parts of the breast.^36- 37 For example, a pathologic analysis of 130 mastectomy specimens with primary tumors 2 cm or smaller found that only 41% had pathologically confirmed unicentric disease. A total of 42% of patients had multifocal or multicentric disease extending more than 2 cm from the primary tumor and an additional 17% had multifocal disease within 2 cm of the primary tumor.³⁷ Because brachytherapy typically irradiates a 1-cm margin of breast tissue surrounding the lumpectomy cavity, these pathologic studies suggest a biologically plausible mechanism for an increased risk of local recurrence and subsequent mastectomy in patients treated with brachytherapy compared with WBI.

In older Medicare beneficiaries diagnosed with invasive breast cancer and treated with lumpectomy, brachytherapy compared with WBI was associated with a decreased likelihood of long-term breast preservation and an increased likelihood of complications, but no difference in overall survival. Potential public health implications of these findings are substantial, given the high incidence of breast cancer, along with the recent rapid increase in breast brachytherapy use. Although these results await validation in the prospective setting, they also prompt caution over widespread application of breast brachytherapy outside the study setting.