Breast conserving surgery (BCS) with radiation has been considered a viable alternative treatment to mastectomy for most women with early stage breast cancer since the 1990 NIH Consensus Conference (1, 2) because of evidence that women who undergo BCS with radiation experience equivalent overall survival compared to mastectomy (3–8). BCS with radiation may offer advantages for cosmetic outcomes, potentially mitigating psychosocial sequelae (9), but generally involves 5 to 7 weeks of postoperative radiation, which, among other factors, may influence a patient to choose mastectomy over BCS with radiation (10). BCS without radiation increases risk of local recurrence and mortality (8, 11–13) and is considered non-guideline treatment (2); the exception is women ≥70 years of age diagnosed with stage I, hormone receptor (HR) positive breast cancer, and on hormonal therapy, where BCS without radiation may be considered accepted treatment (14). Despite the demonstrated benefits of BCS with radiation, prior research based in representative cancer registry data has shown that Asian American (AA) women are considerably more likely than other racial/ethnic groups to have mastectomy (15–18) as well non-receipt of radiation after BCS (17).

Factors that have been associated with choice or use of BCS with radiation over mastectomy include: younger age (19, 20), smaller tumor size and lack of nodal involvement (21, 22), shorter travel distance to radiation facility (23, 24), higher socioeconomic status (SES) (25, 26), physician preference (27–29), and later years of diagnosis (25, 30). Patient involvement with the decision-making process has been associated with both BCS with radiation and mastectomy (10, 31, 32). It is unclear the extent to which these or other factors influence treatment for early-stage breast cancer among AA women. Compared to non-Hispanic White women, the odds of receiving BCS have been shown to be lower among foreign-born AA women than among US-born AA women (33). Another US study based on 82 Chinese breast cancer patients showed that lower SES, foreign birthplace or recent immigration, and speaking no or limited English were associated with mastectomy (34), and a population-based study showed that adjustment for hospital and provider characteristics explained a substantial proportion of the higher odds of mastectomy among Vietnamese women compared to White women (16). A recent study of women in Tianjin, China, showed that patient SES and insurance status, rather than tumor characteristics, were primary factors driving mastectomy over BCS with radiation (35). A survey of providers in the San Francisco Bay Area who treat AA breast cancer patients showed that tumor-to-breast ratios, patients’ attitudes toward preserving the breast, cultural factors, and transportation difficulties were perceived to be important factors in the higher rates of mastectomy in this population (36).

Therefore, in six AA ethnic groups, we set out to identify factors associated with receipt of mastectomy or BCS omitting radiation in the large, population-based Surveillance, Epidemiology, and End Results (SEER) California Cancer Registry (CCR) dataset enhanced with the ability to assess immigrant status, neighborhood factors including SES and ethnic enclave, and hospital characteristics. In addition to polytomous regression in our observational study, we also apply recursive partitioning (RP) to identify and clustered subgroups with the highest receipt of mastectomy or omission of radiation following BCS.

Among 20,987 AA women diagnosed with stage 0-II breast cancer between 1990–2007 and eligible for BCS with radiation, 37.2% received BCS with radiation; 49.7% received mastectomy; and 13.2% received BCS with omission of radiation (Table 1).

In multivariate logistic regression analyses, women age 60 and older were more likely than younger age groups to receive mastectomy (Table 2). Compared to US-born Japanese, US-born Korean, and foreign-born Korean, Chinese, Filipina and Vietnamese were more likely to receive mastectomy. Modeled as a linear variable (as opposed to categorical variable as shown in Table 2), successive increases in single year of diagnosis were associated with lower odds of receiving mastectomy (OR=0.94, 95% CI 0.94–0.95). Compared to women diagnosed in 1990, women diagnosed after 2001 were approximately half as likely to receive a mastectomy. Tumor size was the most important clinical predictor of mastectomy; compared to women with <1 cm tumors, women with tumor size 4–5 cm were over 6 times more likely to receive a mastectomy. Histology, grade, HR status, nodal involvement, and anatomic location of the tumor were also associated with mastectomy, as were hospital characteristics including smaller hospital size, non-NCI cancer center, low hospital SES patient composition, and high hospital AA patient composition. Women residing in low SES ethnic enclaves were more likely to undergo mastectomy than those in high SES, less Asian ethnic neighborhoods.

Successive increases in single year of diagnosis were associated with 5% increases in BCS omitting radiation (Table 2). Tumor characteristics including larger tumor size, DCIS or unknown histology, unknown grade, HR status negative, and multifocal or NOSanatomic site were associated with BCS omitting radiation; however, as discussed below, these associations with clinical characteristics need to be interpreted with caution as they may reflect underascertainment of radiation therapy. BCS omitting radiation was also more likely among AA women living in low SES, ethnic enclave neighborhoods, and women diagnosed in hospitals that were smaller, and with relatively lower patient SES and higher composition of AAs.

The RP tree modeling mastectomy versus BCS with radiation (Figure 1 and Table 3) revealed 11 mutually-exclusive subgroups with tumor size being the most important predictor, as indicated by the first tree split. Year of diagnosis and anatomic location were also featured prominently on the RP tree. The subgroups with the highest proportion of mastectomy (71.8%) were AA women in nodes 1 with ≥2 cm tumors, a subgroup that represented 53% of the total sample. Nodes 2–3 were women with tumors <2 cm, but with multifocal or NOS location of the tumor (node 2, 64.9% mastectomy), or foreign-born Chinese, Filipina, Vietnamese, or US-born Japanese, Korean, Vietnamese diagnosed before 1996 (node 3, 64.7% mastectomy). In contrast, the node with the lowest percent mastectomy (node 5, 35% mastectomy) included AA women with tumors <2 cm, tumor mass confined within one quadrant or were overlapping lesions, and diagnosed in the latter part of the study period, between 2003–2007. There were no splits by hospital nor by neighborhood characteristics.

The RP tree modeling BCS omitting radiation versus BCS with radiation revealed that known versus unknown HR status (primarily reflective of calendar year) was the most important predictor (data not shown). Among women with HR positive and negative tumors, 21% omitted radiation, and there were no additional splits after this node. For women with unknown HR status, three subgroups had high proportions (>60%) of omission of radiation: 1) those with grade I-II tumors in public hospitals with relatively higher proportions of Asians; 2) those aged 75 or older with DCIS in hospitals with relatively fewer Asians; and 3) Koreans, foreign-born Vietnamese, and US-born South Asians aged less than 75 with DCIS in hospitals with relatively fewer Asians.

AAs are among the most rapidly growing racial/ethnic populations in the US and particularly in California, where more than one-third of all US AAs reside (55–59). Breast cancer is the most commonly diagnosed cancer among AA women, and incidence rates are higher in the US compared to most cancer registries in Asia (60). Rates have been found to be increasing over time among US AA women - rates of invasive breast cancer were 104 per 100,000 in 1988–1994 among Californian US-born AA women, and 136 per 100,000 in 2000–2004; among foreign-born, rates for these same time periods were 71 and 79, respectively; annual percentage changes were as high as 4% per year among US-born Filipinas and foreign-born Koreans (45). Since the 1990 Consensus Conference (2), BCS with radiation has steadily increased in general, but AA women are considerably less likely to have BCS with radiation and more likely to omit radiation after BCS (18, 61). Taking advantage of a large, population-based cancer registry enhanced with patient-level immigrant status, neighborhood characteristics including SES and ethnic enclave (62), as well as hospital characteristics, this report contributes new insights into patterns of treatment for early-stage breast cancer among six large ethnic populations of AA women. Our study found that although tumor characteristics are the most important predictors of treatment, patient race/ethnicity and nativity, and hospital and neighborhood SES and racial/ethnic composition, are also important determinants.

In addition to traditional logistic regression, we used RP to identify interactions among patient, hospital and neighborhood factors, and discrete subgroups of AA women with varying proportions of mastectomy or BCS without radiation. Our results indicate that some factors that were independently significant in logistic regression analyses did not feature in the RP trees, suggesting that their independent effects in the total sample were not seen in more discrete patient subgroups. Our finding that clinical factors, such as tumor size and anatomic location, are most important in the mastectomy decision tree is consistent with clinical practice guidelines (37); yet, the high rates of mastectomy among AA women with even moderate-sized tumors (2–5 cm) suggests that high tumor-to-breast ratio is likely an important consideration for these women. The appearance of year of diagnosis as an important factor in the decision tree likely reflects temporal changes in the adoption of treatment recommendations since the original NIH Consensus Conference (2), as examined previously (25).

AA ethnicity and nativity were also predictors of mastectomy receipt among women diagnosed over the same time period with tumors less than 2 cm. Given the same tumor characteristics, specific AA ethnicities differed by nativity in the extent to which they underwent mastectomy. In decision tree analyses, foreign-born Chinese, Filipina, and Vietnamese had high rates of mastectomy, despite small tumor sizes, while foreign-born Japanese and South Asians, who tend to be English-speaking and more acculturated, even when foreign-born, had low mastectomy rates. These findings are reflective of cultural and immigrant/language factors playing a strong role, and consistent with previous findings (31, 34, 36, 63). Thus, besides being a powerful tool for informing further research to identify underlying reasons for these treatment differences, the RP results can be used by providers and hospitals to develop programs and initiatives, such as patient navigators, cultural competency provider training, and availability of translators, that may be implemented within their institutions to ensure that all women are informed of their treatment options. Results from our qualitative study (unpublished) indicated that low SES AA women who were linguistically isolated were often not informed about their treatment choices; however, within one community (San Francisco Chinatown), community and hospital patient navigator programs were extremely successful in helping the Chinese residents receive BCS with radiation, if desired, by providing transportation, translation, appointment scheduling, and overall navigation services.

To our knowledge, no previous study has evaluated breast cancer surgical treatment type among AA subgroups over an 18-year time interval. We found a trend of decreasing rates of mastectomy, consistent with prior reports (25). In an additional analysis (data not shown), we found increasing rates of BCS with omission of radiation since 1990 in women with stage I-II tumors, but not DCIS tumors, and these trends were more pronounced for cases who also received chemotherapy, for reasons that are unclear, but warrant further investigation. This may be because the standard order of therapies (BCS, chemotherapy for 4-6 months, then radiation) pushes radiation outside of the timeframe in which treatment data are collected by the registry.

Our finding of increased rates of BCS omitting radiation supports a growing body of work (64). Previously implicated predictors for radiation omission following BCS include older age, negative nodes, larger hospital size, tumors greater than 2 cm, geographic location, and race (19, 23, 65–70). In our study, although variations by AA ethnicity and nativity were not statistically significant, neighborhood and hospital race/ethnicity composition and SES were significantly associated with omission of radiation after BCS.

However, our results may be impacted by the under-ascertainment of radiation in the cancer registry, particularly if there are delays in the administration of radiation (71). While a recent report found a 32% under-ascertainment of radiation compared to self-report in the Los Angeles SEER registry, one of the registries included in the current analysis, a comparison of SEER and Medicare claims data showed that agreement on radiation was 94% (kappa = 87%) and under-ascertainment by SEER was 7% (72). In the report based on Los Angeles SEER registry data, radiation under-ascertainment was significantly higher among patients with the following characteristics: White race, more advanced stage, younger age, lower income, Medicare or no insurance, had mastectomy, received chemotherapy, had (self-reported) delays in initiation of radiation, and diagnosed in non-ACoS (American College of Surgeons) hospitals (~40% of California hospitals) (73). As under-ascertainment was more likely among patients of lower SES, this may have over-estimated our findings for the associations we found with neighborhood and hospital characteristics. In particular, the association of clinical characteristics with omission of radiation following BCS should be interpreted with caution; they may be reflective of this under-ascertainment of radiation considering that receipt of chemotherapy is a factor associated with missing data on radiation, and patients with worse prognosis/higher stage are more likely to have received a recommendation for chemotherapy. Our study also has several other potential limitations. We opted to include all stage 0-II tumors, rather than invasive tumors only. Although there has been debate about whether BCS with radiation is the preferred treatment for DCIS, a recent study showed excellent long-term prognosis for both invasive and DCIS tumors (12). Our time trend results were unchanged when we excluded DCIS from our analysis. Our RP analysis was not a concern in this regard, given that RP will naturally discriminate on DCIS/ invasive classification. However, one limitation involved in analyses utilizing RP is potential instability in resultant trees that occurs due to the method RP uses of within-population sampling. In addition, it is possible that a proportion of women with breast site NOS or multifocal (n=2,125) may have had multifocal disease and thus have contraindications for BCS with radiation (2). Our polytomous logistic regression results were not substantively altered when excluding these women, and RP naturally discriminates on breast site to identify the most homogeneous subgroups for this variable. Our results may also be affected by the observational nature and inherent data limitations in cancer registry data. Since registry data on surgery captures the most extensive surgical resection, we are unable to consider women who may have had repeated lumpectomies and unclear margins, and who may opt for mastectomy to avoid further repeated surgeries. In addition, we are unable to consider contraindications to BCS with radiation or radiation due to the lack of comorbidity and other patient-level data (19). The lack of data on language and insurance status is an additional limitation. Finally, our study lacks reliable data on breast reconstruction, and the ability to differentiate those who had mastectomy with reconstruction versus mastectomy only. Given that reconstruction can address many of the psychological impacts of mastectomy alone (74), and, especially when available concurrently with mastectomy, may influence a woman’s decision to choose mastectomy over BCS. However, rates of (self-reported) reconstruction from a previous study were very low (only 1 of 21 Asian women who received mastectomy also received reconstruction), and thus unlikely to greatly impact the findings.

This study has a number of strengths, including the large population-based sample collected over a protracted time period, which allowed sufficient statistical power to detect moderate associations and to apply RP. We used RP, a powerful statistical technique to identify meaningful subgroups of AA most likely to receive mastectomy. Our study is the first, to our knowledge, to apply RP to a large population-based sample of AA women and identify ethnicity and nativity as important predictors.

In a population-based sample of AA breast cancer patients, a group with traditionally high rates of mastectomy, our study shows that immigration, neighborhood and hospital factors are associated with mastectomy and omission of radiation. Additional research is needed to understand cultural factors that underlie the decision-making process among AA women, as well as reasons behind the associations with neighborhood and hospital factors. By focusing on the interactions of patient, hospital, and neighborhood factors in the differential receipt of breast cancer treatment, our study identifies subgroups of women for further study, and, most importantly, results that can be immediately translated into public health and patient-focused initiatives to ensure that all women, regardless of race/ethnicity, immigrant status, SES, and language, are fully informed about their treatment options (9).