- © 2006 by American Society of Clinical Oncology
Radiation Therapy After Mastectomy Between 1991 and 1999 in Elderly Women: Response to Clinical Trial Information
- From the Department of Radiation Oncology and Center for Outcomes and Policy Research, Division of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA
- Address reprint requests to Rinaa S. Punglia, MD, MPH, Dana-Farber Cancer Institute, Center for Outcomes and Policy Research, 454B S21-24, 44 Binney St, Boston, MA 02115; e-mail: rpunglia{at}lroc.harvard.edu
Abstract
Purpose No systematic study has analyzed the changing patterns of use of postmastectomy radiation therapy (PMRT) during the period of information dissemination regarding the benefit of radiation therapy in this setting. We sought to study the receipt of PMRT in elderly women in this period, using a population-based cohort of women with breast cancer.
Patients and Methods Using data from the linked Surveillance, Epidemiology, and End Results–Medicare database, we analyzed the use of radiation therapy between 1991 and 1999 in 19,699 women with stage I to III breast cancer who received mastectomy as definitive surgery. Multivariable logistic regression was used to investigate the association of diagnosis year and the use of PMRT, after controlling for clinical and sociodemographic factors.
Results During the entire period studied, 2,160 (11.0%) patients treated with mastectomy received PMRT. The use of PMRT significantly increased in women diagnosed in 1996 (odds ratio [OR], 1.26; P = .04), 1997 (OR, 1.70; P < .0001), 1998 (OR, 1.57; P < .0001), and 1999 (OR, 1.77; P < .0001), relative to those diagnosed in 1991 after controlling for cancer, and sociodemographic and treatment characteristics. There were significant differences in the temporal trends of radiation use among different regions of the country (P < .0001), and between teaching versus nonteaching institutions (P = .04).
Conclusion The use of PMRT in elderly women increased significantly during the period of data dissemination about its efficacy; however, the trends in adoption varied among different practice settings.
INTRODUCTION
Postmastectomy radiation therapy (PMRT) reduces the risk of local-regional (chest wall and adjacent nodal groups) recurrence in women who receive mastectomy for treatment of invasive breast cancer. In addition to preventing local-regional recurrence, PMRT has been hypothesized to prevent seeding of tumor cells into the circulation from persistent local disease or to prevent reseeding in those whose distant disease has been eradicated by systemic therapy. This hypothesis led to the belief that improving local-regional control may translate to an improvement in overall survival. With more effective systemic therapy to treat distant disease, the influence of local-regional control on overall survival may be even more pronounced.
To realize a survival benefit from radiation after mastectomy, however, there needs to be a significant risk of local-regional recurrence without radiation. Which women constitute such a high-risk group has been the subject of considerable controversy. Large randomized trials initiated in the late 1970s and early 1980s to assess the benefit of PMRT in the setting of systemic therapy began to report their findings in the 1990s. Although prior retrospective studies had been used to assess the benefit of radiation therapy after mastectomy in selected patient groups, the randomized trials, which each included more than 1,000 patients treated with systemic therapy, represented more applicable data to inform the question regarding who should receive radiation in the postmastectomy setting. During 1994, findings from the two Danish studies, one conducted in premenopausal and the other in postmenopausal women, were reported in abstract form in Europe.1,2 The Danish trial conducted in premenopausal women at the time of abstract publication not only revealed an improvement of 24% in rates of local-regional recurrence for the radiation arm, but also a statistically significant improvement of 7% in the 7-year survival rate.1 The postmenopausal trial, however, revealed only a 10% improvement in relapse-free survival that did not translate into any survival difference at time of abstract publication.2
In 1997, two articles, one from the Danish premenopausal trial and another trial studying premenopausal women in British Columbia, confirmed the survival benefit of radiation therapy after mastectomy for women at high risk of recurrence in this age group.3,4 However, it was not until 1999 that a full report of the Danish study conducted in postmenopausal women was published. Despite the initial report of no survival difference in 1994, this article demonstrated a survival benefit for the first time even in this older population.5
Due to differences in surgical technique, the translation of these trials into treatment guidelines for patients treated in the United States has been the subject of much controversy. A patterns-of-care study evaluated clinical factors associated with the use of PMRT in the United States during the years of 1998 and 1999.6 Nevertheless, no systematic study has analyzed the changing patterns of PMRT use during the period of information dissemination regarding the benefit of radiation therapy in this setting. Therefore, we sought to study the variation in the receipt of PMRT in the years before and during the introduction of data regarding its efficacy.
PATIENTS AND METHODS
Data Sources
Patients were taken from the linked Surveillance, Epidemiology, and End Results–Medicare (SEER-Medicare) database. Eleven tumor registries participating in the SEER program during the 1990s captured approximately 97% of incident cases,7 covering a representative sample of approximately 14% of the United States population.8 Registries collect tumor and demographic data on each patient. Inpatient and outpatient Medicare claims, physician, laboratory, durable medical equipment, home health, and hospice billings have been linked to SEER for patients 65 years and older.9 Sociodemographic information at the census tract level for each patient is also included. The current SEER-Medicare database contains SEER information through 1999 and Medicare claims through 2001.
Cohort Selection
The cohort consisted of American Joint Committee on Cancer stage I to III female unilateral breast cancer patients diagnosed in a SEER region between January 1, 1991, and December 31, 1999. Histologies included were ductal, lobular, mixed ductal and lobular, tubular, medullary, mucinous, papillary, and other or unspecified adenocarcinomas. We excluded patients enrolled in Medicare for end-stage renal disease or disability, patients with previous cancers, those whose death date differed by more than 3 months between SEER and Medicare, those with diagnoses made from autopsy or death certificates, or those with unknown dates of diagnosis. We also excluded patients who did not have continuous Medicare enrollment (Part A and Part B) or if they were enrolled in a health maintenance organization (HMO) any time from 13 months before diagnosis (for use in comorbidity assessment) through 12 months after mastectomy.
All patients underwent mastectomy as identified in SEER (site-specific surgery codes 30 to 70 and 90) or Medicare: American Medical Association Current Procedural Terminology (CPT) codes 19180, 19182, 19184 to 19187, 19200, 19211 to 19216, 19220, 19224 to 19229, 19240, or 19250 to 19255; International Classification of Diseases (9th revision, clinical modification (ICD9-CM) procedural codes 85.41, 85.43, 85.45, or 85.47; and hospital Diagnosis-Related Group (DRG) codes 257 or 258. For mastectomies identified solely in SEER (1.1% of total), 4 months after diagnosis was used to calculate a proxy mastectomy date. For inclusion, women could not receive chemotherapy or radiation therapy before mastectomy, given that patients may be subject to different criteria regarding the use of PMRT. Patients were allowed additional cancer diagnoses after this first breast cancer, but were excluded if the subsequent cancer diagnosis was within 1 year of first mastectomy. Accordingly, patients were excluded if they did not survive at least 1 year after mastectomy. Five patients with incomplete information about nodal evaluation were excluded. The Institutional Review Board of Dana-Farber/Partners Cancer Care approved this study.
Patient Characteristics
Explanatory variables used in this study included diagnosis year (to study trends in the receipt of radiation therapy related to the release of clinical trial results), tumor characteristics, other clinical characteristics (age at diagnosis, comorbidities), treatments received (nodal examination, breast reconstructive surgery), as well as sociodemographic factors, distance from nearest radiation treatment facility, and type of treatment facility (teaching institution v nonteaching hospital).
Year of diagnosis (1991 through 1999) was studied as a categoric variable (1991 was the referent) to identify any nonlinear trends. Variables for tumor characteristics such as size and the number of positive nodes were categorized according to a priori clinical cutoffs. Comorbidities were identified by looking for diagnostic billing codes for specific health conditions during the year before diagnosis of breast cancer using the Deyo implementation10 of the Charlson score,11 and applied to both inpatient and outpatient claims, as suggested by Klabunde et al.12 The Charlson score was then categorized as 0, 1, 2, or 3 or more. Treatments received were identified from Medicare billing claims. Early reconstruction surgery was defined as that which occurred within 4 weeks of mastectomy and was identified using ICD9-CM, CPT, and DRG codes.
Socioeconomic status quintiles were developed on the basis of information availability, prioritizing census tract median income and per capita income according to zip code. If census tract and zip code information were missing (< 1% of cohort), the patient was classified in the lowest socioeconomic quintile.13 Education was evaluated at the census tract level using quintiles. A patient was classified as having a history of Medicaid enrollment if she was part of the Medicaid program at any time between 1986 and 2002, which was used to define a personal history of low income. The cancer registries were categorized into regions: West (San Francisco, CA; Hawaii; New Mexico; Seattle, WA; Utah; San Jose, CA; and Los Angeles, CA), Midwest (Detroit, MI; Iowa), Northeast (Connecticut), and South (Atlanta, GA; rural Georgia). Race was evaluated as white, black, or other; ethnicity was Hispanic or non-Hispanic. Marital status at diagnosis was married versus other.
The distance to the nearest radiation facility for each patient was determined by using an algorithm based on latitude and longitude that calculates the distance from the zip code of the patient's residence to that of the closest radiation therapy facility included in the 2000 American Hospital Association Annual Survey of Hospitals (which did not have information about radiation facilities not associated with a hospital).14 A patient was identified as having received mastectomy in a teaching hospital if there was an institutional payment for indirect medical education during their hospitalization.
Outcome Studied
Because of potentially long chemotherapy regimens that may be administered before PMRT, we defined our outcome as radiation initiated within a year of mastectomy. Radiation therapy administration was identified in Medicare using ICD9-CM code V58.0; ICD9-CM procedure codes 92.20 to 92.29; CPT codes 77000 to 77999; Revenue Center codes 0330, 0333, 0339; DRG code 409; inpatient (Medicare Provider Analysis and Review [MEDPAR]) radiology oncology indicator and inpatient radiology therapeutic indicator; and Berenson-Eggers Type of Service code P7A. SEER identifies radiation administration within 4 months of diagnosis. Patients who had SEER-identified external-beam radiation, not administered before or during surgery, were also included as having radiation therapy.
Statistical Analyses
The following statistical analyses were conducted using SAS for Windows (Version 8.2; SAS Institute, Cary, NC). First, univariate analyses were conducted with PMRT receipt as the outcome, using χ2 and Fisher's exact tests for categoric variables, and t tests or Wilcoxon rank sum tests for continuous variables. A forward and backward elimination algorithm was used with logistic regression to determine the final multivariable model. Results are presented as odds ratios (ORs) with 95% CIs and P values.
Multivariable analyses were not adjusted for use of chemotherapy, given that receipt of chemotherapy was believed to be highly collinear with receipt of radiation, and the use of chemotherapy is not a factor determining whether or not PMRT should be administered in treatment guidelines.15,16
Interaction terms to test a priori hypotheses (eg, diagnosis year with type of treatment facility, region of country, patient age, nodal status, comorbidities, and patient income) were then studied using the final model. Those variables that revealed significant effect modification (ie, the interaction term with year was significant) were studied in separate analyses. Adjusted odds ratios for these variables were studied per year and CIs were derived by the method described by Figueiras et al.17
RESULTS
Among 19,699 patients with stage I to III breast cancer (47.1% stage I, 45.4% stage II, 7.4% stage III) treated with mastectomy, 2,160 (11.0%) patients received PMRT during the entire period. In addition to tumor characteristics such as larger size (P < .0001), higher grade (P < .0001), lobular histology (P < .0001), lack of hormone receptor expression (P < .0001), and increasing number of positive nodes (P < .0001), diagnosis year (P < .0001) was associated with PMRT on univariate analysis (Table 1).
Univariate Predictors of RT Use After Mastectomy
After we controlled for significant cancer, and sociodemographic and treatment characteristics, multivariable analysis revealed increased use of PMRT during the later years of the period studied (Table 2; Fig 1). The odds for receipt of radiation therapy relative to women diagnosed in 1991 increased to 1.26 for women diagnosed in 1996 (95% CI, 1.01 to 1.56), in 1997 to 1.70 (95% CI, 1.38 to 2.10), in 1998 to 1.57 (95% CI, 1.26 to 1.95), and in 1999 to 1.77 (95% CI, 1.42 to 2.20). A sensitivity analysis restricting the cohort to women with stage I disease revealed a corresponding increase in PMRT use during the later years (OR, 1.33; 95% CI, 1.07 to 1.65, for diagnoses after 1995 or v 1995 or earlier).
Odds ratios for receiving postmastectomy radiation therapy and 95% CIs separated by year of diagnosis (1991 to 1999) after controlling for age, comorbidity, distance to the nearest radiation treatment facility, region of country, history of low income, number of removed nodes, mastectomy in a teaching institution, number of involved nodes, tumor size, grade, histology, and estrogen receptor status.
Factors Significantly Associated With RT Receipt After Mastectomy in Multivariable Analysis
Secondary analyses revealed significant interaction between year of diagnosis and whether or not mastectomy was performed in a teaching institution (P = .04), and year of diagnosis and region of country (P < .0001). The ORs for each year of diagnosis relative to 1991 separated by region of country are shown in Figure 2. Separating the odds of receiving radiation relative to diagnosis year by region of the country reveals two disparate temporal patterns of radiation adoption. In the South and the West, radiation use steadily increased after 1994. But in the Midwest and Northeast, after an initial increase, odds of radiation use returned to 1991 levels by the end of the period. Compared with 1991, the likelihood of receiving radiation therapy during 1999 in the West had increased to 2.29 times (95% CI, 1.69 to 3.11) and in the South to 2.90 times (95% CI, 1.29 to 6.52). However, by 1999, the odds for receiving PMRT in the Northeast had returned to 1.15 (95% CI, 0.62 to 2.16) and remained almost constant in the Midwest at 1.11 (95% CI, 0.74 to 1.67) relative to use in each region, respectively, during 1991.
Adjusted odds ratios for receiving radiation therapy after mastectomy and 95% CIs in each of the four regions studied for each year, 1991 to 1999, relative to the Midwest in 1991.
The ORs for each diagnosis year relative to 1991 separated by whether or not mastectomy was performed in a teaching institution are displayed in Figure 3. By 1999, the likelihood of receiving PMRT if mastectomy was performed in a teaching hospital returned to 1.28 (95% CI, 0.86 to 1.91) times the likelihood of that in 1991, but remained elevated if mastectomy was not performed in a teaching hospital to 2.04 (95% CI, 1.58 to 2.65) times the likelihood of that in 1991.
Adjusted odds ratios for receiving radiation therapy after mastectomy and 95% CIs based on type of institution where mastectomy was performed for each year, 1991 to 1999, relative to teaching hospitals in 1991.
DISCUSSION
We found significant variation in the use of radiation therapy after mastectomy in older women during the years of 1991 to 1999 within the SEER-Medicare data set, after controlling for tumor, patient, and treatment characteristics. Relative to women who underwent mastectomy in 1991, patients who were diagnosed during or after 1996 had a statistically significantly increased likelihood of receiving radiation therapy. Exploratory analyses revealed significant differences in the time trends of radiation use by region of the country and by type of institution where mastectomy was performed.
What could have caused this variation? One possibility is the differential interpretation of the Danish abstracts presented in 1994, in which a survival benefit in younger women treated with PMRT was revealed, but in which there seemed to be no difference in survival between the arms in the postmenopausal trial with 7 years of follow-up. Given that our study was limited to women receiving Medicare benefits, it seems that some physicians extrapolated the benefit seen in premenopausal women to this older age group. Indeed, in the South and the West, the odds of receiving radiation therapy in our data set increased to even greater levels after publication of the premenopausal randomized trials that confirmed the overall survival benefit with radiation therapy in 1997.3,4
For others, initial optimism regarding the translation of the overall survival benefit to older women may have been tempered in later years by the lack of difference in survival seen in the postmenopausal trial during the interim analysis presented in the 1994 abstract. This postmenopausal trial included patients younger than age 70 years (median age, 62 years).2 The absence of survival benefit for radiation use in the older postmenopausal group relative to the premenopausal group1 may have been used to hypothesize an even smaller benefit for the elderly population that we studied, in which the median age was 76 years.
In the setting of accruing trial information regarding efficacy of PMRT, both patterns of use (that of benefit extrapolation or that of tempered enthusiasm based on initial negative findings of trial results) seem reasonable. This phenomenon of marked variation in medical practice has been described in a number of different settings.18-20 Early adoption of radiation therapy in nonacademic hospitals may be due to greater financial incentives related to the addition of radiation in this setting.
Clinical practice guidelines15,16 about the use of radiation therapy after mastectomy were not published until after our period of analysis. Indeed, considerable controversy regarding the translation of the trials into treatment recommendations for women undergoing surgery in the United States still exists, especially for patients with one to three positive nodes. Given the lack of treatment recommendations during this era, we included all women with stage I to III breast cancer who received mastectomy in our analysis, regardless of nodal status, and instead adjusted our analyses for clinical factors such as tumor size and number of positive nodes, given that we were interested in both overuse and underuse of radiation therapy. Moreover, sensitivity analyses with stage I patients omitted revealed that our results were essentially unchanged (data not shown) and there was no significant interaction between nodal status and year of diagnosis (P = .13).
Limitations of our study include those common to observational studies using administrative data. This data source only captures Medicare patients, has incomplete data on the roughly 15% of patients in managed care, and does not have information regarding prescription drugs. There may also be differences in the patient population and practice patterns in a fee-for-service setting versus those in HMOs.21-23 For example, HMO patients may have fewer comorbidities than patients in the Medicare population.24 In addition, the SEER sample is not a national probability sample: SEER data come from cancer registries from selected regions that are representative of the national population in terms of education and income, but the sample is more heavily urban and foreign-born than the rest of the population. Regional variation noted in the data may be specific to the SEER registries of the region, and not the entire region. Methods for comorbidity adjustment continue to undergo development and revision.12 In addition, data on margin status after mastectomy are not available. However, only a small minority of mastectomies performed have positive margins.25 We controlled for all variables available to us, but could not account for any variable not provided that may have been changing during the period of analysis. In addition, administrative data were not collected for research purposes; therefore, there is uncertainty about their accuracy and completeness. For example, treatment identification relied largely on Medicare procedure codes, which may be underreported. Finally, the diagnoses dates available from SEER-Medicare precluded us from assessing the full impact of the postmenopausal Danish trial published in manuscript form in 1999 during the end of the era that we analyzed.5 However, presentations at national meetings have been shown to influence community treatment patterns for primary breast cancer.26
In conclusion, we have demonstrated a significant increase in the use of radiation therapy after mastectomy in older women during the period of 1991 to 1999, corresponding to the accumulation of favorable trial results. However, the adoption of radiation therapy during this period varied by region and by type of institution. Such variation may be due to differences in dissemination or interpretation of data. Although different interpretation of inconclusive data may be unavoidable, the implications of our findings emphasize the need for efficient and thorough dissemination of research data to practicing physicians. Moreover, in settings of limited or inconclusive data, optimal care would include treatment decisions based on patient preferences rather than geographic and institutional variation.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
Author Contributions
Conception and design: Rinaa S. Punglia, Jane C. Weeks, Craig C. Earle
Administrative support: Jane C. Weeks, Bridget A. Neville
Provision of study materials or patients: Craig C. Earle
Collection and assembly of data: Bridget A. Neville, Craig C. Earle
Data analysis and interpretation: Rinaa S. Punglia, Jane C. Weeks, Craig C. Earle
Manuscript writing: Rinaa S. Punglia, Jane C. Weeks, Craig C. Earle
Final approval of manuscript: Rinaa S. Punglia, Jane C. Weeks, Bridget A. Neville, Craig C. Earle
Footnotes
-
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
- Received January 18, 2006.
- Accepted May 1, 2006.
