The clinical and pathological features of the study population (n = 10,441) are shown in Table 1. The median OS was 29.0 months. Both primary site and distant metastatic site resections were performed in 272 (2.61%) patients (R0 group), 4,025 patients (38.55%) underwent primary site resection only, 409 patients (3.92%) underwent metastases resection only, and 5,735 patients (54.93%) had no surgery. The characteristics of the HR+ and HR- populations are shown in Supplementary Table S1.

According to the follow-up results, 7,125 events (breast cancer-specific deaths) were observed. From the Kaplan-Meier survival curves for these four groups, which are shown in Figure 1, the R0 group showed the best overall survival outcome with a median survival time (MST) of 51 months, followed by the primary resection group (MST = 43 months) and metastases resection group (MST = 31 months). The no resection group achieved an MST of 21 months. The difference among the four groups was significant (P < 0.001).

The stratified analyses based on age at diagnosis, race, grade, tumor stage, radiation, HR status and metastatic site were conducted to explore the differences among the four groups and to determine which subgroup would benefit most from surgery. The survival results of each subgroup are shown in Table 2. It is obvious that the three groups with surgery achieved better MST than the no resection group in each subgroup.

The results of the multivariate analysis using the Cox proportional hazards model are shown in Table 3. Age over 45 years, black race, poor grade, T3/T4 stage, metastases in organs and HR- status were independently associated with decreased OS, while other race and the three surgery groups were independently associated with increased OS. Compared with the no resection group, the R0 group showed the best hazard ratio of 0.558 (95% CI, 0.471-0.661), the primary resection group and metastases resection group also had a good survival benefit with a hazard ratio of 0.590 (95% CI, 0.557-0.625) and 0.782 (95% CI, 0.693-0.883).

The MST and 5-year OS of the four groups based on different HR status are shown in Table 4. The corresponding Kaplan-Meier survival curves are shown in Figure 2. In the HR+ population, the survival of each group with surgery, including the R0 resection group (MST = 66 m; 5-year OS = 54.1%), primary resection group (MST = 52 m; 5-year OS = 44.9%) and metastases resection group (MST = 38 m; 5-year OS = 31.7%) were all significantly longer than the no resection group (MST = 28 m; 5-year OS = 22.0%) (P values were all < 0.001). Furthermore, the R0 group gained an additional survival benefit compared with the primary resection group (P value <0.001). On the other hand, in the HR- population, the survivals of the R0 group (MST = 18 m; 5-year OS = 26.7%) and primary resection group (MST = 24 m; 5-year OS = 25.0%) were both significantly longer than the no resection group (MST = 12 m; 5-year OS = 11.8%) but not the metastases resection group (MST = 12 m, 5-year OS = 6.8%) (P = 0.526). Furthermore, the R0 group did not achieve a better survival benefit compared with the primary resection group (P = 0.691).

To assess the interaction of HR positivity and the role of surgery, three surgery groups were merged into one group, the surgery group. Univariate logistic regression analysis (results are shown in Supplementary Table 2) by HR positivity revealed that HR-negative patients were significantly more likely to undergo surgery (OR = 1.201; P < 0.001); however, after adjusting for other factors, multivariate analysis (results are shown in Supplement Table 3) showed that HR positivity was not independently associated with greater or less likelihood of surgery (P = 0.814).

The SEER program included 17 population-based cancer registries, together comprising approximately 28% of the total population of the United States.[21–24] Women initially diagnosed with stage IV primary invasive breast cancer from the 2004 to 2008 year were included in the analysis. Patients after year 2008 were not included to ensure a sufficiently long follow-up time.

Breast cancer patient data were obtained from the SEER Program using Case Listing. Site record ICD-O-3 was limited to the breast. Case patients with sarcomas of the breast (based on the histology codes 8800, 8801, 8805, 8815, 8830, 8850, 8858, 8890, 8935, 8980, 8982, 8983, 9120, 9180, 9181, and 9260) were excluded. Multiple primary cancers were also excluded to make accessible analyses of cancer-specific survival. The patients with a survival of less than one month were excluded because such patients may die of surgical complications.

All of the study data—including demographic characteristics, ER, PR status, tumor stage, grade, classification of metastatic site, surgery of primary site and distant site, radiotherapy, cause of death and survival months—were all collected. HER2 status was not collected because it was not available for the cases before year 2010.

Cancer-specific survival was calculated from the date of diagnosis to the date of death related to breast cancer. Death attributed to other causes was considered as censored observation. Cases of grade 4 in histology were combined with cases of grade 3 because the outcome for cases assigned grade 3 or grade 4 was not significantly different.[25] As recorded in the SEER database, the metastatic sites of breast cancer were classified into several groups as follows: 1. distant lymph node(s), 2. metastases in designated organs: adrenal (suprarenal) gland, bone, other than adjacent rib, contralateral (opposite) breast, lung, ovary, satellite nodule(s) in skin other than primary breast, 3. metastases in other organs, 4. metastases in at least two of the above sites, 5. unknown.

The ER and PR results were combined and analyzed jointly as hormone receptor (HR) status. HR+ was defined as ER+ or PR+. HR- was defined as both ER- and PR-. ER/PR borderline case patients (borderline ER: n = 22, 0.2%; borderline PR: n = 78, 0.7%) were defined as unknown. Age at diagnosis was categorized as ≤ 45 years old versus >45 years old. All patients were divided into the following four groups: R0 group (patients who underwent primary site resection and distant metastatic site resection); primary resection group (patients who underwent only primary site resection); metastases resection group (patients who underwent only distant metastatic site resection); and no resection group (no resection was performed on any patient).

The observed differences between the different surgery groups were analyzed statistically by chi-squared test. Univariate analysis of survival was performed using Kaplan-Meier methods, while group comparisons were performed using the log-rank test. Adjusted hazard ratios along with 95% intervals were calculated using the Cox regression model. Additionally, univariate and multivariate logistic regression analyses were applied to assess factors, especially hormone receptor status, associated with undergoing surgery. Differences were considered to be statistically significant if the P value was less than 0.05. SPSS 19.0 software (IBM Corp., Armonk, USA) was used for data analysis.