The data of the present research were obtained from the latest Surveillance, Epidemiology, and End Results (SEER) Plus database (SEER 9 Registries data, with additional treatment information, Nov 2021 sub). The SEER database is an authoritative source of information on cancer, covering approximately 34.6% of the population in the USA. The records of male patients diagnosed with breast carcinoma between 1990 and 2016 were extracted using the SEER*Stat software (version 8.4.1), ensuring long-term follow-up of at least 5 years to estimate the risk of developing a second primary cancer. The International Classification of Diseases for Oncology third edition (ICD-O-3) was used to identify breast malignancy by site code C50 (including C50.1 to C50.9). The three key variables “year of diagnosis”, “sequence number”, and “total number of in situ/malignant tumors for patient” of the SEER Plus database were used to determine the status of SPM. Cases that were diagnosed as synchronous cancers occurring as SPM within 2 months after initial diagnosis or those in which the breast malignancy was not the patients’ first primary malignancy were excluded. The inclusion criteria were as follows: (1) male breast malignancy was the only or the first primary malignancy; (2) histological diagnosis confirming the existence of breast malignancy; and (3) under treatment and the follow-up data were available. The exclusion criteria were as follows: (1) incomplete cases with missing information on important variables; (2) the SPM (if any) data were incomplete; (3) initially diagnosed with distant metastasis; and (4) synchronous cancers. The flowchart of case selection is shown in Supplementary Figure 1 .

A total of 1,843 MBC patients were involved in this study. Variables such as age, race, marital status, year of diagnosis, sequence number, total number of in situ/malignant tumors for patient, histological type, tumor grade, TMN stage, surgery performance, radiotherapy performance, chemotherapy performance, months from diagnosis to treatment, the hormone receptor (HR) status, HER2 status, survival time, and cause of death were extracted. Age was regrouped into six groups (<45, 45–55, 55–65, 65–75, 75–85, and 85+). Race was regrouped into white, black, and other. Marital status included married, single, and divorced. Histological type was divided into infiltrating duct, adenocarcinoma, and other by the SEER Plus database. The HR status was classified as HR positive [estrogen receptor (ER) and/or progesterone receptor (PR) was positive] and HR negative (both ER and PR were negative). TMN stage was adjusted to the 6th AJCC staging edition by the SEER Plus database in the additional analysis. The site and the diagnosis time of the SPM were recorded. Overall survival (OS) refers to the time from the initial cancer diagnosis to cancer-specific death.

The cumulative incidence of cancer-specific death and the occurrence of SPM were calculated based on the Fine and Gray competing risk model. The Kaplan–Meier method was constructed to estimate the difference in OS between MBC survivors with and without an SPM. The entire cohort was randomly divided into a development cohort (70%) and a validation cohort (30%) for the development and validation for the competing risk nomogram. Standardized mean differences (SMDs) were used to assess distributional differences in the baseline variables between the development and validation cohorts. As HER2 status is known to be tested after 2010, and HER2 status should be routinely diagnosed clearly in breast cancer patients nowadays, sensitivity analyses were carried out excluding those MBC patients whose HER2 was unknown or whose diagnosis was made prior to 2010.

The univariate and multivariate Cox regression analyses were firstly performed to identify variables that significantly affected the breast cancer-specific survival and occurrence of SPM. However, applying only univariate and multivariate Cox regression analyses was inadequate, because aside from the primary tumor, there were other factors that might threaten the patients’ lives, such as accidents and infectious or other serious diseases. As a result, death due to other causes acted as a competing risk event to death due to a specific cancer. Hence, the Cox proportional hazards model might overestimate the incidence rate of the outcome with the passage of time. Similarly, death due to primary breast cancer or other causes also acted as a competing event for the MBC patients to develop an SPM—only those cured from MBC could have the probability of developing an SPM during their long survival time. In this study, the additional Fine and Gray competing risk analysis was applied to compare the association among different causes of death with a competing risk framework: death due to breast cancer or death due to other causes. Then, as for the occurrence of SPM in MBC survivors, the Fine and Gray method was also applied: death due to primary breast cancer or other causes was the competing event in the development of an SPM.

In order to help clinicians predict the survival probability of MBC patients and their individual probability to develop an SPM, nomograms were established based on the multivariate competing risk models. Next, we identified low-risk and high-risk survivors by calculating the 50th quantiles of total points of the nomograms and compared the difference of their survival time. Validation of these nomograms was performed by calculating the concordance index (C-index) and plotting calibration curves by a bootstrapping method with 1,000 resamples. Furthermore, the receiver operating characteristic (ROC) curves were drawn to estimate the predictive value by calculating the area under the ROC curves (AUCs). Meanwhile, decision curve analyses (DCAs) were conducted to show the clinical effectiveness of the nomogram models.

All analyses were performed using R software (version 4.21, https://www.r-project.org/). Significance level was set as p < 0.05.

A total of 1,843 MBC patients, who were initially diagnosed between 1990 and 2016, were finally enrolled in the present study. Among these MBC patients, 60 (3.26%) developed at least one SPM. A total of 339 (18.39%) patients died from MBC, and 707 (38.4%) patients died from other causes. Among those survivors who suffered from an SPM, prostate cancer represented 24 (40%) of all SPMs, followed by lung and bronchus cancer at 6 (10.0%), melanoma of the skin at 5 (8.3%), secondary breast cancer at 4 (6.7%), liver cancer at 3 (5.0%), urinary bladder cancer at 3 (5.0%), kidney and renal pelvis cancer at 2 (3.3%), NHL at 2 (3.3%), pancreas cancer at 2 (3.3%), rectal cancer at 2 (3.3%), and stomach cancer at 2 (3.3%). The SPM details of these MBC survivors are shown in Figure 1 . The median OS of all the enrolled patients was 102.41 months. The median latency from diagnosis of initial breast primary cancer to subsequent diagnosis of the SPM was 67.2 months. The detailed information of these MBC patients is summarized in Tables 1 , 2 .

As is shown in Figure 2A , there was no significant difference in OS between the development and validation cohorts (p = 0.83). The OS of MBC patients who did not suffer from an SPM was 101.87 ± 68.17 months, while the OS of those who suffered from an SPM was 118.63 ± 75.76 months. Those who developed an SPM have a significantly longer OS ( Figure 2B , p = 0.027).

Univariate and multivariate Cox regression were applied in the development cohort to select the predictive variables for the prediction models of cancer-specific death. MBC patients whose HER2 was unknown or diagnosed prior to 2010 were excluded in the following sensitivity analyses as mentioned above. As is shown in Table 3 , tumor grade, TMN stage, surgery, and chemotherapy were related to OS in the univariate analysis, while in the multivariate Cox regression, chemotherapy failed to show a significant relation with OS.

Univariate and multivariate Cox regression were also applied to select the predictive variables for the occurrence of SPM. As is shown in Table 4 , marital status showed a significant relation with the occurrence of SPM in the univariate analysis. Moreover, in the multivariate Cox regression, age, race, tumor differentiated grade, histological type, TMN stage, chemotherapy, and the waiting time from diagnosis to begin treatment were significant.

The Fine and Gray method was used to estimate the risk predictors for cancer-specific death and the occurrence of SPM. The results of the characteristics are provided in Table 5 . Age, race, marital status, histological type, TMN stage, therapy, the waiting time from diagnosis to begin treatment, HR status, and HER2 status were the significant risk factors for both cancer-specific death and the development of an SPM.

The first two nomograms were established based on the previously mentioned risk factors to predict the survival probability of MBC patients. Age, race, marital status, tumor differentiated grade, histology, TMN stage, surgery, radiotherapy, chemotherapy, duration to begin treatment, HR status, and HER2 status, which were selected by the Fine and Gray method, were enrolled in nomogram model 1 to predict the 5-year, 8-year, and 10-year survival probability of MBC patients ( Figure 3A ). Meanwhile, age, tumor differentiated grade, TMN stage, and surgery, which were selected by the multivariate Cox regression, were included in nomogram model 2 ( Figure 3B ) to predict the same survival probability above. The C-index of model 1 was 0.710 in the development cohort and 0.703 in the validation cohort, while model 2 had a C-index at 0.728 in the development cohort and 0.718 at the validation cohort. Both model 1 (AUC = 0.713) and model 2 (AUC = 0.757) achieved a better predictive value than the AJCC TMN staging system (AUC = 0.689) in the ROC analysis shown in Figure 4A . The integrated discrimination improvement (IDI) and net reclassification improvement (NRI) between model 1 and TMN stage were 0.610 (95% CI 0.490–0.258) and 0.333 (95% CI 0.182–0.508), respectively. Meanwhile, The IDI and NRI between model 2 and TMN stage were 0.059 (95% CI 0.036–0.193) and 0.290 (95% CI 0.154–0.513), respectively. The calibration curves show that both model 1 and model 2 had good agreement between predicted probability and the observed outcome ( Figures 4B, C ). The DCA also showed that model 1 ( Figures 4D, E ) and model 2 ( Figures 4F, G ) had a good discrimination in both the development and validation cohorts. We divided the patients into a low-risk group and a high-risk group at the 50th percentile of nomogram total points and compared the difference of the survival time among these subgroups. Figures 5A, B show that there were significant differences in survival time between different risk groups in both the development cohort (p = 0.0022) and the validation cohort (p = 0.002), based on nomogram model 1 ( Figure 3A ). Meanwhile, Figures 5C, D also show the survival time difference between different risk groups in the development cohort (p = 0.001) and the validation cohort (p = 2e-04), based on nomogram model 2 ( Figure 3B ), which indicated that both of these nomogram models had a good discrimination capability for the survival probability of the MBC patients. The details of these two nomograms are shown in Supplementary Tables 1 , 2 .

An additional nomogram model 3 was established to predict the probability of MBC survivors developing an SPM within 10 years after the initial diagnosis. All of the risk factors selected by the Fine and Gray method were included in model 3 ( Figure 6A ). The C-index of model 3 was 0.909 in the development cohort and 0.494 in the validation cohort. The AUC of the ROC curve in model 3 is 0.934 ( Figure 6B ). The calibration curve is shown in Figure 6C . The DCA curve is shown in Figure 6D in the development cohort and in Figure 6E in the validation cohort. The details of these risk factors are shown in Supplementary Table 3 .