Study data were extracted using SEER*Stat software (version 8.3.9), utilising the “Incidence-SEER Research Plus Data, 18 Registries, Nov 2020 Sub (2000–2018)” dataset. All cases were initially identified using International Classification of Diseases for Oncology, Third Edition (ICD-O-3) histology and site codes relevant to the study population.

Due to the absence of the “SEER Combined Mets at DX-brain” variable for patients diagnosed before 2010, it was impossible to determine the presence of brain metastases in these cases. Therefore, the study included only patients diagnosed with brain metastases between 2010 and 2018. Those with missing data on age at diagnosis, race, sex, marital status, or incomplete follow-up were excluded. The primary endpoint of this study was OS, defined as the time from diagnosis to death from any cause or last follow-up.

The overall study design is illustrated in a flowchart ( Figure 1 ). A total of 55,094 patients diagnosed with brain metastases between 2010 and 2018 were retrospectively identified from the SEER database. After data cleaning and selection based on predefined inclusion and exclusion criteria, eligible patients were randomly assigned to training and testing sets in a 7:3 ratio. A multivariable Cox regression model was applied to identify independent prognostic factors, and a nomogram was constructed based on variables including age, sex, race, income, marital status, primary site, surgery, radiotherapy, and chemotherapy to predict 1-year OS. The performance of the nomogram was assessed using the concordance index (C-index), time-dependent Receiver Operating Characteristic (ROC) curves, and calibration plots at 3, 6, and 12 months in both training and testing sets. The total nomogram score for each patient was calculated, and the corresponding AUC was used to compare the predictive accuracy of the nomogram with the TNM clinical staging system.

The primary endpoint of this study was OS. Categorical variables included sex, race (White, Black, Other), and marital status (married, single, divorced/separated, widowed). Age was treated both as a categorical variable (using clinically meaningful age groups) and as a continuous variable in selected models.

Kaplan-Meier curves and the log-rank test were used to assess the survival rate. To evaluate prognostic factors associated with OS, univariate Cox proportional hazards regression was first performed. Variables with p < 0.05 in univariate analysis were included in a multivariate Cox regression model to identify independent predictors of survival. HR and 95% confidence interval (CI) were reported.

The effect of age on the likelihood of developing brain metastases and related survival outcomes was analysed continuously through restricted cubic splines RCS within Cox regression models, with three to five knots placed at percentiles of the age distribution. Model performance was evaluated using the concordance index (C-index) to assess discriminative ability, and time-dependent ROC curves were utilized to evaluate predictive accuracy at different time points (e.g., 3, 6, 12 months). Calibration plots comparing predicted versus observed survival probabilities were used to assess model calibration.

All statistical analyses were carried out using R software (version 4.3.1) with relevant packages including “survival”, “rms”, “timeROC”, and “ggplot2”.

55,094 patients diagnosed with brain metastases between 2010 and 2018 were enrolled in this study ( Table 1 ). The survival distribution of the patients followed a U-shaped pattern, with a higher proportion of patients having survival times of less than six months or more significant than twelve months. Patients having survival times of between six and 12 months were less common. Specifically, 59.1% of patients with brain metastases had an OS of less than six months, including 4.2% who did not reach the endpoint. In contrast, 22.8% of the population had survival times exceeding 12 months. Among patients with survival times less than 12 months, the proportion of males was higher than females. However, the trend reversed in those with survival times more significant than 12 months, with a significantly higher proportion of females (p < 0.001).

Among all primary sites of brain metastases, lung cancer had the highest incidence (79.2%), followed by skin cancer (3.8%), breast cancer (3.5%), and kidney parenchyma cancer (3.1%). At the initial diagnosis of brain metastases, a younger age was associated with a better subsequent prognosis. The mean age at initial diagnosis for patients with survival times of 0–6 months was 66.41 years. The mean ages for the better survival groups were significantly lower, at 62.79 and 60.41 years, respectively.

In addition to the factors above of age, sex, and primary site, other variables such as race, income, marital status, and treatment modalities also significantly influenced the prognosis of brain metastases. All observed differences were statistically significant (p < 0.001). However, due to the retrospective nature of the SEER database, certain data points may be incomplete or missing, potentially impacting the accuracy and generalizability of these findings.

The p-values in Table 1 were calculated based on grouping the population into three categories according to survival time. For categorical variables, we applied the chi-square test; for continuous variables, one-way ANOVA was used. When the assumptions of the chi-square test or ANOVA were not met, non-parametric tests were employed to assess differences among the survival groups.

As shown in Figure 1 , most patients with brain metastases from various primary sites had a median survival time of less than six months. The poorest prognosis was observed in patients with brain metastases from liver cancer, where the median survival time was two months (95% CI: 2–3 months). Brain metastases from stomach cancer had a median survival time of three months (95% CI: 3–4 months). Lung cancer, which represented the most significant proportion of brain metastases, also had a poor prognosis with a median survival time of five months (95% CI: 4–5 months). Conversely, patients with brain metastases from breast cancer, lymphoma, prostate cancer, and testicular cancer had a median survival time exceeding ten months. Among these, testicular cancer had the most favourable prognosis, with a median survival time surpassing 17 months (95% CI: 14–45 months).

In addition to the primary site of origin, age is a critical factor influencing the prognosis of patients with brain metastases. Cancer is often considered a disease of aging, with its occurrence and outcomes closely linked to the aging process (23). We employed RCS to model these effects and explore further the relationship between the primary site of brain metastases and the age-related risk variation in survival.RCS is a flexible statistical method for modelling non-linear relationships between variable (24). By applying RCS in our analysis, we were able to examine how the risk of mortality changes with age among patients with brain metastases from different primary cancers. This approach provides a nuanced understanding of the impact of age on survival, taking into account the varying risks associated with other primary sites.

As illustrated in Figure 2 , the findings of our study indicate that the risk of survival following the development of brain metastases from the majority of primary sites increases with age. However, the pattern of this age-related risk varies among different primary cancers. For patients with brain metastases originating from lung cancer, breast cancer, and lymphoma, the mortality hazard exhibits a rapid increase with age across all age groups (P < 0.001). In contrast, for patients with brain metastases from testicular cancer, colorectal cancer, and EndometrialCarcinoma, the risk of mortality due to brain metastases remains relatively stable until the age of 65, after which it increases sharply. It is noteworthy that patients with brain metastases from liver cancer and rectal cancer also exhibit an age-related increase in mortality risk. However, this trend does not reach statistical significance (P>0.05). These findings emphasise the necessity of considering both the primary site of cancer and the patient’s age when evaluating prognosis and developing treatment strategies for brain metastases.

The decision between radiotherapy and surgical resection (”surgical resection” in this manuscript specifically refers to neurosurgical resection of metastatic brain lesions, not resection of the primary tumours.)can be influenced by several factors, including the size and diversity of the tumour (25–27). While the majority of contemporary studies concentrate on tumour size to ascertain the necessity of surgery for brain metastases, there has been a lack of research investigating whether the primary location of the tumour influences the decision to operate. As illustrated in Figure 3 , the results of our study indicate that surgical intervention for brain metastases originating from various primary sites generally reduces the risk of death. For instance, the mortality risk for lung cancer metastases following surgery is reduced by nearly half, with an HR of 0.49 [95% CI 0.46-0.53]. After adjusting for age and gender, the HR is 0.52 [95% CI 0.48-0.56]. Similarly, the HR for kidney parenchyma metastases is 0.43 [95% CI 0.37-0.49] and 0.44 [95% CI 0.38-0.50] after adjustment. The HR for breast cancer metastases is 0.63 [95% CI 0.54-0.74] and 0.66 [95% CI 0.56-0.77] after adjustment. Nevertheless, it should be noted that not all tumours with brain metastases are suitable for surgical treatment. For instance, patients with brain metastases from liver cancer, bladder cancer, or esophageal cancer do not exhibit a significant survival benefit post-surgery (P>0.05).

Seventy percent of patients with brain metastases were randomly assigned to the training set, while 30% were assigned to the test set. The prediction model incorporated variables related to survival, such as sex, age, race, income, marital status, surgery, radiotherapy, and chemotherapy. To enhance usability, we created a nomogram (see Figure 4 ) that assesses the importance of these variables in predicting patient survival. Higher scores indicate a more significant impact on survival. The model’s robustness was evaluated using the C-index, ROC, and calibration curves. It showed excellent performance in both sets, with calibration curves in Figures 5A, B indicating that predicted survival values at 3, 6, and 12 months closely matched observed values, reflecting high stability. The C-index was 0.723 for the training set and 0.722 for the test set. The time-dependent AUC for the training set was 0.83, 0.80, and 0.77 for the respective time points, while the test set yielded values of 0.83, 0.80, and 0.76, as shown in Figures 5C, D .

This study has several limitations that should be acknowledged. First, its retrospective design based on data from the SEER database may introduce selection biases and preclude establishing definitive causal relationships. Second, the quality and completeness of data within the SEER database are inherently limited, with certain variables potentially incomplete or missing, thus potentially impacting the accuracy and generalizability of our findings. Additionally, detailed treatment information, including specifics regarding chemotherapy, radiotherapy, and immunotherapy, was not available, which could significantly influence patient survival outcomes. Furthermore, since the SEER database predominantly includes data from specific geographic regions, the generalizability of our results may be restricted by variations in race, socioeconomic status, and healthcare access. Moreover, the study did not incorporate potentially important prognostic factors such as genetic biomarkers, molecular characteristics, treatment responses, nor did it account for recurrence or progression of brain metastases, which might limit the precision of the predictive model. The primary focus on short-term survival analysis, with limited medium- and long-term follow-up data, further restricts the assessment of long-term survival outcomes and quality of life. Finally, the absence of detailed analysis regarding variations in surgical techniques, timing of surgical intervention, and institutional expertise may impact the robustness of our conclusions. Therefore, future research through large-scale prospective cohort studies is essential to validate and enhance the predictive accuracy of our findings.