Surgery of the Primary Tumor Offers Survival Benefits of Breast Cancer with Synchronous Ipsilateral Supraclavicular Lymph Node Metastasis

Controversy exists around the locoregional management of the primary tumor for breast cancer associated with synchronous ipsilateral supraclavicular lymph node metastasis (sISLM) due to the rarity of the disease and limited available data. This study aimed to compare outcomes of patients in the Surveillance, Epidemiology, and End Results (SEER) database with sISLM who underwent surgical resection and radiation of the primary tumor with those who did not. This population-based retrospective study included breast cancer patients with sISLM without distant metastases from 2004 to 2016 in the SEER database. In this study, patients had been stratified by operative management, and propensity score matching (PSM) had been successfully applied. A total of 1172 breast cancer patients with sISLM were included in the study: 863 (73.6%) of patients underwent the primary tumor resection, and 309 (26.4%) patients did not undergo surgery. The median survival time in the surgery group was longer compared to the nonsurgery group in the overall cohort and the PSM cohort. We concluded that the primary tumor resection was associated with improved survival. Subgroup analysis further demonstrated that local surgery was not inferior to radical surgery. For selected breast cancer patients with sISLM, surgery is a promising local intervention which may improve the survival.


Introduction
The incidence of breast cancer with ipsilateral supraclavicular lymph node metastasis (ISLM) without distant metastasis is as low as 1-4% [1][2][3]. Two types of ISLM can be identified in breast cancer, synchronous ISLM (sISLM) and metachronous ISLM (mISLM) [4]. sISLM is defined as ISLM at the primary diagnosis of breast cancer (T1-4 N3c M0), and mISLM as the occurrence of supraclavicular lymph node metastasis after the initial diagnosis and treatment of breast cancer.
In the 5th edition of the AJCC TNM staging system, breast cancer with the classification of sISLM was modified from N3 to M1 due to its poor outcome and developing distant metastasis within 1 year [5,6]. Soon after that, in the 6th edition of the AJCC TNM staging system, breast cancer with sISLM, which was no longer regarded as distant metastasis, was reclassified as stage IIIC (N3c) instead of stage IV and it has maintained the IIIC (N3c) classification to date [7][8][9].
Despite the low incidence, there has always been significant controversy about the local treatment, especially surgery, of the primary tumor site of breast cancer with sISLM. Fan et al. [4] demonstrated that the primary tumor size and radiotherapy after surgery, not including surgery itself, were independent prognostic factors for sISLM. Chen et al. [10] found that young age and the surgical removal of the primary tumor and supraclavicular nodes after sISLM were associated with a significantly better prognosis and survival. Besides, a retrospective study in Japan showed that local-regional therapy, including surgery and radiation, was incapable of reversing the poor prognosis of sISLM breast cancer patients [11]. Breast surgeons tend to perform the primary tumor and lymph node dissection in patients with newly diagnosed sISLM [12]. Currently, whether breast cancer patients with sISLM can benefit from the primary tumor resection remains debatable.
Hence, our study aimed to evaluate the survival benefit of surgery of the primary lesion in breast cancer patients with sISLM based on the data from the Surveillance, Epidemiology, and End Results (SEER) database. Therefore, we retrospectively analyzed the data of 1172 patients from the SEER database to evaluate the main demographic and clinicopathological characteristics affecting prognosis.
Our study provides a more in-depth and comprehensive understanding of the clinical features of breast cancer with sISLM and attempts to obtain more evidence to form a theoretical basis for surgery treatment.

Patients and methods
Data source and cohort selection Data were acquired from the SEER database between January 1, 2004, and December 31, 2016, because the 6th AJCC TNM staging system was adopted by SEER starting in 2003. We extracted the data for all cases that were initially diagnosed as malignant primary breast cancer, and multiple primary malignant tumors were excluded. A total of 638,902 patients who were diagnosed with the primary breast cancer between January 1, 2004, and December 31, 2016, were used for analysis in this study. Breast cancer patients who did not have N3c stage disease or who had distant metastasis were excluded. After these steps, we excluded patients who were male or for whom data were unavailable for some critical variables. The follow-up time of the patients was from breast cancer diagnosis until death or the end of the follow-up period. Finally, 1172 female patients who were diagnosed with sISLM (stage N3c but not M1) were included in the analysis (S. Fig. 1).
The SEER is a freely available database [13], and the data released by the SEER database do not require informed patient consent, as cancer is a reportable disease in the USA. The original data in this study were downloaded from the SEER Web site server (https://seer.cancer. gov/data/) via the SEER*Stat software version 8.3.5 in the client-server model.

Variables
The following demographic and clinicopathological characteristics of N3c stage patients before and after PSM were included in the analysis process ( Table 1): age at diagnosis, race, marital status, laterality, histologic type, tumor differentiation grade, T stage based on the Derived AJCC Stage Group (6th) [14], estrogen receptor (ER) status, progesterone receptor (PR) status, HER2 status, molecular subtype, surgery, radiotherapy, and chemotherapy. Patients were divided into two major subgroups: no surgery and surgery, which could be divided again into the local surgery and radical surgery subgroups ( Table 2). The SEER registry provided hormone receptor (HR) information for each patient. Specifically, ER, PR, and HER2 status were included as positive, negative, and undetermined status. A major subset of the patients' HER2 status in this cohort was unavailable because the HER2 status data were collected by SEER starting in 2010. For radiation, due to the data availability of the SEER database, the extent of the radiation fields is not known. Specifically, it is unclear whether the radiation included the sISLM as well.

Propensity score matching (PSM)
This study was a retrospective and observational study, so the surgery assignment was not random. Some significant covariates of the patients in the active treatment and control groups were heterogeneous and possibly affected the outcomes. Therefore, we further compared the survival rate between the surgery and nonsurgery cohorts by using 1:1 nearest-neighbor matching, setting the caliper as 0.02. The PSM process was applied to minimize the selection bias and to approximately balance the baseline covariates with the analytic settings between groups [15].

Statistical analysis
Using the Chi-square test, we compared patient characteristics between the surgery and nonsurgery patients. Overall survival (OS) was defined as the time from diagnosis to death from any cause, and disease-specific survival (DSS) was determined based on the date of initial diagnosis to the date of disease-related death. OS and DSS were the primary endpoints of this study. Kaplan-Meier analysis and the log-rank test were used to estimate OS and DSS before and after PSM. Cox proportional hazard regression was used to identify a hazard ratio (HR), and a corresponding 95% confidence interval (CI) was calculated. A forest plot was utilized to conduct a subgroup analysis comparing the survival rate. All data were examined using SPSS statistical software (version 22.0; IBM Corporation). All statistical tests were two-sided, and the statistical significance level was set at P \ 0.05.  Removal of all breast tissue, the nipple, the areolar complex, and variable amounts of breast skin in continuity with the axilla. The specimen may or may not include a portion of the pectoralis major muscle e Removal of breast tissue, nipple, areolar complex, variable amount of skin, pectoralis minor, and/or pectoralis major, as well as en bloc axillary dissection f Removal of breast tissue, nipple, areolar complex, variable amounts of skin, pectoralis minor, and/or pectoralis major, as well as removal of internal mammary nodes and en bloc axillary dissection

Baseline characteristics of the patients before and after PSM
In the present study, the patient characteristics are presented in Table 1 Based on the eligibility criteria, a cohort of 1172 patients was enrolled in the present study. The median age of the patients was 55 (range 22-98) years. The surgery and nonsurgery cohorts included 863 (73.6%) and 309 (26.4%) patients, respectively. After propensity score matching, a total of 464 patients (surgery 232 (50.0%) versus nonsurgery 232 (50.0%)) were matched. All variables were properly balanced between these two groups. The baseline characteristics of the patients before and after propensity score matching are summarized in Table 1.
Patients undergoing surgery could be subdivided into local surgery (n = 335, 38.8%) and radical surgery (n = 528, 61.2%) subgroups, as defined in Table 2.  Fig. 1. The survival analyses indicated that the surgery patients had a significantly better OS (P \ 0.0001) and DSS (P \ 0.0001) compared with those of nonsurgery patients (Fig. 1a, b). The OS and DSS differences persisted in the PSM cohort (Fig. 1c, d OS P \ 0.0001, DSS P \ 0.0001).
Additionally, OS and DSS of the local surgery group were 63.1 and 65.8% at 3 years, 54.8 and 58.3% at 5 years, and 42.1 and 48.7% at 10 years. Comparatively, OS and DSS of the radical surgery group were 61.1 and 64.0% at 3 years, 49.2 and 52.9% at 5 years, and 37.4 and 43.1% at 10 years (Table 3). The median OS for the local surgery group and the radical surgery group was 87.29 months and 69.85 months, respectively. Also, the median DSS for the local surgery group and the radical surgery group was 114.60 months and 79.54 months, respectively. There was no significant difference for the OS (P = 0.1994) and DSS (P = 0.1738) between the patients who underwent local surgery and those who underwent radical surgery (Fig. 2a,  b). PSM between the local surgery group and the radical surgery group was performed (specific data are not shown.) Also, no significant differences were found between these two groups for OS (P = 0.5187) and DSS (P = 0.4668) after PSM (Fig. 2c, d).
Moreover, Kaplan-Meier curves for OS and DSS were generated based on the surgery and radiation variables, which are presented in Fig. 3. The Kaplan-Meier analyses Univariate log-rank test P values are reported indicated that surgery-treated patients had a remarkably better OS and DSS than nonsurgery group (P \ 0.0001) (Fig. 3a, b). There was no significant difference in nonsurgery patients for the OS and DSS between the radiation and nonradiation groups (OS P = 0.5593, DSS P = 0.6967, S. Fig. 2a, b). There was a significant difference for the OS and DSS in surgery patients between the radiation and nonradiation (OS P = 0.0078, DSS P = 0.0101 S. Fig. 2C  & 2D).

Subgroup analyses
To assess whether the distribution of the subject characteristics was consistent, two subgroup analyses of overall survival and disease-specific survival were performed in prespecified subgroups using a forest plot (Figs. 4, 5). The two prespecified stratification factors were the presence or absence of surgical treatment (subgroup analyses A, Fig. 4), and between local surgery and radical surgery (subgroup analyses B, Fig. 5).
Among the 1172 patients included in subgroup analysis A, for OS, the forest plot indicated that there was a significant difference when the efficacy of surgery to no surgery was compared. Most variables showed that surgery benefited the patients with sISLM compared to no surgery (Fig. 4). Explicitly speaking, no statistical difference was observed in unknown marital status and nonchemotherapy subgroup for OS, unknown marital status, lobular carcinoma, unknown HER-2 status, and luminal B molecular subtype for DSS (Fig. 4).
In the subgroup analysis B involving 863 patients, the forest plot showed that the efficacy of local surgery and radical surgery was similar (Fig. 5). There were statistical differences in the 36-45 age subgroup and ER-positive status subgroup for OS, indicating that local surgery offers less hazard ratio than radical surgery. Similarly, in the 36-45 age subgroup, the unknown marital status subgroup, and ER-positive status subgroup, patients who underwent local surgery have a better prognosis for DSS. Nevertheless, most of the other subgroups did not show statistical differences. It may indicate that, for sISLM patients, compared with radical surgery, local surgery has an equivalent prognostic value (Fig. 5).

Discussion
In recent decades, the classification of patients with sISLM in the AJCC TNM staging system has been revised several times due to the different prognostic conclusions drawn from various studies [5-7, 16, 17]. These findings also raise questions about the most appropriate local treatment for breast cancer patients with sISLM. The most heated debate is whether the implementation of the surgery on the primary lesion affects the patient's prognosis. A meta-analysis published in 2015 suggests that the prognosis for breast cancer patients with ipsilateral ISLM is similar to patients with stages IIIb/c disease and different from patients with stage IV disease. The authors also proposed that radical, instead of palliative therapy, seems more plausible for these patients [18]. Other studies suggested that breast cancer women with sISLM should be considered as a local-regional disease, and aggressive local therapy may indeed improve outcomes [19,20]. However, studies found that the locoregional surgical therapy after the diagnosis sISLM did not correlate with survival [4,21].
In our present study, we evaluated 1172 primary breast cancer patients with sISLM from the SEER database between 2004 and 2016. In Kaplan-Meier curve analysis, forest plots of subgroup analyses for OS and DSS indicated that surgery was a significant prognostic factor for sISLM patients (Figs. 1, 4). Further, the subgroup analysis showed that both local surgery and radical surgery might have an association with improved prognosis (Fig. 5). Moreover, about enhancing N3c patients' survival, the efficacy of local surgery is not inferior to that of radical surgery (Fig. 5). For radiotherapy, one of the most important means of local treatment, there was a significant difference between the radiation and nonradiation groups in surgery patients but not in nonsurgery patients (Fig. 3). These results suggested to our clinicians that, in the face of this specific type of breast cancer, the surgical approach might significantly improve the patient's quality of life and prognosis. Interestingly, local surgery had a similar prognostic effect as that of radical surgery, suggesting that in the clinical setting, a smaller, local treatment would be a better choice.
Retrospective studies including our study are associated with several limitations, such as inability to control for selection bias and lack of information about the administration of systemic therapy response to therapy and sequencing of treatment in patients who get multimodal therapy. Nevertheless, we tried to perform PSM to reduce the adverse effects of the natural bias and increase the objectivity and scientificity of the research. There was an association with surgery and improved prognosis and that a randomized control trial would be required to demonstrate or explore this further.

Conclusion
In conclusion, for patients with breast cancer with sISLM, surgery is an effective local intervention and the efficacy of local surgery is not inferior to that of radical surgery. The