According to the above inclusion and exclusion criteria, a total of 18,322 patients were enrolled in this study. Table 1 shows baseline demographic and clinicopathological characteristics of the included patients according to metastatic sites. Median survival after diagnosis of metastatic disease was 26.0 months, with 8,442 patients (46.1 %) being alive at the end of the follow-up period.
The largest subgroup was the bone metastasis, comprising 39.8% of all patients (7,292), followed by multiple metastasis (33.07%, 6,059), lung metastasis (10.94%, 2,005), liver metastasis (7.34%, 1,346), other metastasis (7.34%, 1,344), and brain metastasis (1.51%, 276). The median patient age at initial diagnosis was 62 years (range 20–100 years). The mean patient age for liver metastasis was the lowest (59.0 years), while lung metastasis was the highest (66.2 years) among all metastasis groups (range 61.2-64.3 years). Poorly or undifferentiated tumors were most common in liver metastasis (48.66%), followed by lung metastasis (47.16%) and brain metastasis (42.39%). Patients with lung metastasis (50.69%) and multiple metastases (48.07%) tend to have larger tumors of T3-T4 at initial diagnosis. Furthermore, these two groups of patients also have a higher proportion of later N stage (N3), making up 29.48% and 33.94% respectively. The distribution of subtypes is significantly different between initial metastatic sites (Table 1). Bone was the predominant initial site of metastasis for the HR+/HER2- (64.4%) group and the least common site in the HR-/HER2+ group (3.24%). Of patients with brain, lung and liver metastasis, 26.9%, 21.22%, and 14.42%, respectively, had the triple negative subtype. Detailed patient characteristics are presented in Table 1.
Kaplan-Meier Survival analysis
Of all the 18,322 patients finally recruited, 9,880 patients had died by the end of the last follow‐up, 7,239 of whom died of breast cancer specifically. The median survival of all patients since diagnosis of distant metastases was 26.0 months. The median survival ranged from 8.0 months (95% CI 5.65-10.35 months) for patients with brain metastasis to 36.0 months (95% CI 34.74-37.27 months) for patients with bone metastasis.
The Kaplan‐Meier plots were displayed in Figure 2 to show the survival of all populations. Fig. 2A shows the overall survival (OS) of stage IV patients who were enrolled into the study according to metastatic sites. Patients with bone metastasis had the best survival, with 3-year OS rate of 50.5%, followed by patients with other metastasis, liver metastasis, and lung metastasis, (with OS rate of 41.9%, 38.2%, and 37.5% respectively). Patients with brain metastasis and multiple metastasis had worse OS than other subgroups: the 3-year OS rate was 19.9%, and 27.4%, respectively (p < 0.001). Fig. 2B shows the breast cancer specific survival (BCSS) according to metastatic sites of stage IV patients. Similar to OS, brain metastasis had the worst survival, with a 3-year BCSS rate of 50.6%. The 3-year BCSS rate with liver, lung, and multiple metastases were, 67.7%, 66.1%, and 62.6% respectively. Bone and other metastasis had better BCSS, for which the 3-year BCSS rate was 76.6% and 65.3%, respectively (p < 0.001).
Breast cancer subtype is an important prognostic factor since it affects survival significantly. The triple negative pattern tends to decrease survival dramatically compared to the other groups (Figure S1). Survival estimates overall (Figure, 3) and breast cancer specific (Figure S2) as stratified by subtype are graphically displayed in the Figure; HR+/HER2- (A), HR+/HER2+(B), HR-/HER2+ (C), and TN (D). In the first three subtypes, worst survival was seen in the brain metastasis and best survival in the bone metastasis, while in triple negative subtype, patients with multiple metastases had the worst prognoses.
Cox regression analysis of survival
In order to further figure out the effect of multiple factors on BCSS and OS, the Cox proportional hazard model was applied to the analysis. Univariable analysis of BCSS and OS proved unmarried marital status, African descent, high tumor grade, large tumor size, later N stage, and the Her2 positive and triple negative cancer subtypes to be distinct risk factors for poor survival (hazard ratio [HR] > 1, p < 0.001). By contrast, married status, other race, apply of chemotherapy, radiotherapy, and surgery were found to be protective factors for better survival (hazard ratio [HR] < 1, p < 0.001) (Table 2). As for metastatic sites, the results were consistent with those from the Kaplan-Meier analysis. Patients with bone metastasis had the best BCSS and OS, followed by patients with other, liver, lung, and multiple metastases. Specifically, patients with brain metastasis exhibited the worst BCSS, with a hazard ratio of 1.708 (95% confidence interval [CI] = 1.442-2.023, p < 0.001), and OS (HR= 2.492, 95% CI =2.161-2.874, p < 0.001), when compared to bone metastasis.
The variables age, race, marital status, grade, subtype, T stage, N stage, surgery, radiotherapy, chemotherapy and metastatic sites were subsequently analyzed with the multivariable Cox analysis. After adjusting for age, race/ethnicity, marital status, tumor grade, breast cancer subtype, tumor size, nodal status, surgery, radiotherapy, and chemotherapy in the analysis, metastatic site remained an independent prognostic factor of BCSS (p < 0.001) and OS (p < 0.001). The detailed results of Cox regression analysis of BCSS and OS can be found in Table 3. Compared to patients with bone metastasis, the BCSS (HR 0.994, 95% CI 0.881–1.122, p = 0.921) and OS (HR 0.994, 95% CI 0.897–1.100, p = 0.902) of patients with lung metastasis were not significantly different. Patients with other metastasis had similar BCSS (HR 0.955, 95% CI 0.827-1.103, p = 0.532) as patients with bone metastasis, but worse OS (HR 1.127, 95% CI 1.001-1.269, p = 0.048). Patients with liver (HR 1.384, 95% CI 1.208–1.586, p < 0.001; OS, HR 1.428, 95% CI 1.272–1.602, p < 0.001) and multiple metastases (BCSS, HR 1.475, 95% CI 1.361–1.599, p< 0.001; OS, HR 1.806, 95% CI 1.684–1.937, p< 0.001) had worse BCSS and OS than bone metastasis. Similarly, multivariable analysis also indicated that those patients with brain metastasis had significant inferior BCSS (HR 1.975, 95% CI 1.551-2.514, p < 0.001) and OS (HR 2.307, 95% CI 1.862-2.859, p < 0.001) over other metastatic sites. Age (p < 0.001), marital status (p < 0.001), tumor grade (p < 0.001), triple negative subtype (p < 0.001), tumor size (p < 0.001), surgery (p < 0.001), chemotherapy (p < 0.001) were also statistically significant in a multivariable analysis with forward selection.
We also did a competing risk analysis to estimate the marginal probability for each cause of death. The competing risk analysis on cause of death for the entire cohort and according to metastatic sites is reported in supplementary data (Fig. S3). Of the cohort, the respective 5-year estimates of breast cancer-specific mortality and other cause of mortality were 56.2% and 20.7%.