A retrospective metastatic TC cohort including patients with primary mTC (M1 stage) from 2010 to 2016 was selected from the Surveillance, Epidemiology and End Results (SEER) program dataset. Additional inclusion criteria were as follows: patients must be no <14 years old of age; TC was the primary diagnosis of cancer; distant metastatic sites must be documented, including distant lymph nodes, lung, liver, bone, and brain; patient received active follow-up; and patients had a survival length ≥ 1 month. We excluded patients with insufficient data of pathological characteristics, distant metastases, M1 stage, extragonadal germ cell tumor, and data of survival outcomes. Patients with controversial data (e.g., patients at M1a stage without metastases of lung or distant lymph nodes) were also excluded from the cohort (Figure S1). Covariates such as metastatic sites, surgery of primary tumor, TNM stage, serum tumor markers after orchiectomy, and pathological characteristics were included in the analysis. Survival time (months), the status of survival, and cause of death were also included in the analysis. S stage was obtained according to different level of patient's postoperative tumor markers including alpha fetoprotein (AFP), human chorionic gonadotropin (hCG), and lactate dehydrogenase (LDH) based on the staging system of the 2016 Tumor, Node, Metastasis (TNM) of the International Union Against Cancer (UICC). Data such as systemic therapy, radiation therapy, and surgery of metastatic site were not available in the database. Unless specifically emphasized, distant lymph nodes generally refer to following three conditions: (1) retroperitoneal lymph nodes specified as above the diaphragm; (2) external iliac nodes, pelvic nodes and inguinal nodes without or unknown previous scrotal or inguinal surgery prior to presentation of the testis tumor; (3) other kinds of metastatic distant lymph nodes. It is worth noting that external iliac nodes, pelvic nodes as well as inguinal nodes with previous scrotal or inguinal surgery prior to presentation of the testis tumor were classified as regional nodes. And retroperitoneal nodes below the diaphragm, nodes around spermatic vein, periaortic nodes, and pericaval nodes were also classified as regional lymph nodes.

Overall survival (OS) and cancer-specific survival (CSS) were the two major outcomes of this study. OS was defined as the interval of time from TC diagnosis to death. CSS was defined as the interval from the date of TC diagnosis to death due to TC or other causes.

Demographic factors were reported as mean, median, and interquartile range for continuous variables, and frequency and proportion for categorical variables. Comparisons of means and proportions were analyzed by the Student's t-test and Chi-squared test, respectively. Survival was analyzed by the Kaplan-Meier survival analysis. Breslow test as well as log-rank test were performed to analyze the discrepancies between OS and CSS. A Cox proportional model with hazard ratios and 95% CI was then implemented for multivariate analyses of the cohort. All statistical analyses were performed using SPSS Statistics software version 20.0 (IBM, NY, US).

Overall, 1,661 patients with mTC were enrolled in the current study, and were identified according to our defined inclusion and exclusion criteria (Figure S1). Patients' demographical characteristics are shown in Table 1. The average age at diagnosis was 33.2 years, with a median of 31 years. There was no significant difference in the distribution of the number of patients in each year (from 2010 to 2016). Most of the population (90.4%) was ethnically white. More than two-thirds of the patients were unmarried. Primary TC on the right side was more common than on the left side. Approximately 84.3% of patients underwent radical orchiectomy. Patients' clinicopathological information are also summarized in Table 1 according to histological type. Since the number of patients (n = 80) with non-germ cell testicular cancer (NGCTC) was too small, these individuals were not listed separately in Table 1. It is notable that patients with seminoma were older than those with non-seminomatous germ cell tumor (NSGCT) (P < 0.001).

The distributions and discrepancies in the sites of metastases of TC patients are illustrated in the Venn diagram of Figure 1. The most common sites of metastases were lung (1,029 cases, 61.9%), followed by distant lymph nodes (601 cases, 36.2%), liver (252 cases, 15.2%), bone (112 cases, 6.7%), and brain (106 cases, 6.4%). Most patients (953, 57.4%) had a single site of distant metastases, followed by two sites (444, 26.7%), three sites (73, 4.4%), and four sites (10, 0.6%). No more than four metastatic sites were found in any one case of the cohort. When classified according to histological type, the distribution was different than that of metastases in TC patients (Figures S2A, S2B). Patients with seminoma usually had distant lymph node metastases (199 cases, 49.4%), followed by that to lung (108 cases, 26.8%) (Figure S2A). Furthermore, patients with NSGCT were more likely to have metastases to lung (855 cases, 72.5%), followed by distant lymph nodes (385 cases, 32.6%) (Figure S2B).

Since it is commonly acknowledged that patients with distant lymph node metastases have favorable prognoses, we took distant lymph node metastases as a reference and patients with a single metastatic site were included in the Kaplan-Meier analysis (n = 947). Patients with brain metastases were excluded in the analysis of patients with only one metastasis due to insufficient samples (only six patients). In Figures 2A–F, it was consistent with clinical experience that patients diagnosed with liver, lung, and bone metastases had worse outcomes in the measurement of both OS and CSS, compared with those with distant lymph node metastases alone. Using log-rank test to compare the number of metastatic sites in all patients (n = 1,661) with TC, significantly longer OS and CSS was observed in patients with one metastatic site than those with more metastases (OS: HR = 2.196, 95% CI = 1.760–2.739; CSS: HR = 2.492, 95% CI = 1.964–3.162; both P < 0.001) (Figures 3A,B) In the subgroup of patients with only one metastatic site (n = 947), patients with lymph node and lung metastases showed significantly longer OS and CSS compared with bone and liver metastases by log-rank comparison (P < 0.001) (Figures 4A,B). Since curves of liver metastases and bone metastases crossed with each other in the Kaplan-Meier analysis, comparisons between liver and bone metastases required further multivariable Cox regression analysis.

Parameters besides histology, T stage, N stage, S stage, and site of metastases were included in the multivariable Cox regression model of single-site metastasis patients. As shown in Table 2, histology, T stage, and metastatic site were independent prognostic factors for mTC patients both in OS and CSS (all P < 0.05). S stage was a prognostic factor only for CSS (P = 0.016). However, it still showed a trend in predicting OS (P = 0.052). N stage showed no statistical difference in predicting OS and CSS of mTC patients. In the analysis of metastatic sites, bone and liver represented the two groups with the worst prognosis of OS and CSS (all P < 0.05) (Table 2). Lung and brain metastases did not achieve statistical difference.

Among mTC patients with single metastases, a total of 251 patients with seminoma and 655 patients with NSGCT were included in the subgroup analysis, while patients with brain metastases were excluded (n = 6).

For patients with NSGCT, bone and liver metastases showed worse outcome than lung and distant lymph node metastases both in OS and CSS (both P < 0.001) (Figures 5C,D). The same results could be validated in multivariable Cox models (Tables 3, 4). Especially, bone metastases showed a worst outcome compared with the other three types of metastases (OS: HR = 5.116, 95% CI = 2.031–13.884; CSS: HR = 5.243, 95% CI = 1.900–14.466; both P = 0.001).

For patients with seminoma, although there was no statistical difference in the Kaplan-Meier analysis, it still showed a certain trend that liver metastases indicated the worst outcome than the other three types of metastases both in OS and CSS (Figures 5A,B). Subsequent Cox models showed the same trend, although they also did not meet statistical significance (Table 4). Interestingly, patients with bone metastases show similar survival trend as lung and lymph node metastases.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.