We collected data from a unicentral retrospective cohort of 461 patients with NSCLC-associated BM, who visited the Cancer Hospital of China Medical University between January 2005 and December 2017. The inclusion criteria encompassed (1) age ≥18 years; (2) a confirmed pathological diagnosis of NSCLC through tracheoscopy, lung puncture biopsy, metastasis biopsy, or surgical biopsy; (3) a pathological diagnosis of LUSC or LUAD; (4) NSCLC-induced BM confirmed by imaging and/or pathology, such as head-enhanced magnetic resonance imaging (MRI); (5) access to complete clinical data and follow-up information; (6) the absence of other malignancies. All enrolled patients who did not have an endpoint event (death) were followed up through outpatient visits, inpatient care, or telephone follow-up for at least 1 year.

Various pathological variables in patients at the time of their BM diagnosis were documented, including age (< 65 vs. ≥ 65), gender (male vs. female), pathological type (LUSC vs. LUAD), general health status score assessed by the Karnofsky Performance Status (KPS < 80 vs. ≥ 80), synchronous BM occurrence (yes/no), number of intracranial metastases (single vs. multiple), number of extra-cranial organs affected (1–2 vs. > 2), radiotherapy (yes/no), surgery (yes/no), chemotherapy (yes/no), EGFR gene mutation(yes/no), and PET-CT (yes/no). Overall survival is defined as the date from the date of BM diagnosis to the date of patient death or last follow-up visit.

IBM SPSS 26.0 software was used to create patient groups with synchronous versus metachronous BM originating from lung cancer through a 1:2 propensity score matching (PSM) approach. This was done to reduce selection bias and account for potential confounding variables. Univariate and multivariate Cox proportional hazards regression analyses, for both the raw and the propensity score-matched dataset, were used to assess survival risk factors. Variables with p-values < 0.1 in the univariate analysis were included in the multivariate analysis. Results were expressed as hazard ratio (HR) and 95% confidence interval (CI). Additionally, a univariate Cox proportional risk regression analysis was used to assess the effect of gene mutation and pathological type on survival stratification in patients with synchronous versus metachronous BM. Kaplan–Meier analysis and a two-sided log-rank test were used to assess the effect of the type of lung cancer and the presence of driver mutations on survival outcomes. A chi-square test was performed using R 4.2.1 to analyze the baseline characteristics before and after PSM. A Cox proportional risk regression model was used for assessing prognostic correlations, survival curve plotting, and survival data analysis. All hypothesis tests were two-sided, and results were considered significant when p-values were less than 0.05.

A total of 461 patients who were initially diagnosed with BM from NSCLC and admitted to the Cancer Hospital of China Medical University from March 2005 to December 2017 were included in this study. The selection process resulted in the exclusion of 61 patients for the following reasons: 19 patients due to an unclear diagnosis of pathological types, 11 patients with pathological types other than lung adenocarcinoma and squamous lung cancer, 23 patients due to missed visits, and 8 patients with incomplete clinical information. Ultimately, 400 patients were included in this study. Among them, 130 patients had synchronous BM and 270 patients had metachronous BM. On this basis, a 1:2 PSM was performed, resulting in 130 patients with synchronous metastases and 260 patients with metachronous metastases, achieving a balanced distribution between the groups ( Figure 1 ).

Clinical features of the 400 patients are presented in Table 1 . There was no significant stats discrepancy between the two groups of patients with synchronous versus metachronous brain metastases from lung cancer before PSM in terms of age at diagnosis of brain metastases [years], gender, KPS score, type of pathology, number of metastases, number of affected extracranial organs, brain surgery, brain radiotherapy, EGFR genetic mutations, and PET-CT (p > 0.05). There were between-group differences in whether chemotherapy was administered (p < 0.05), and further equalization of baseline characteristics was required. the distribution of baseline characteristics was well balanced between the two groups of patients with synchronous and metachronous brain metastases after PSM, and no statistically significant difference was observed between the two groups (p > 0.05).

Univariate analyses were used to analyze the influence of different covariates on OS. In the univariate analysis conducted before PSM, synchronous BM (HR: 1.242 [95% CI: 1.003–1.539], p = 0.047), squamous carcinoma (HR: 1.522 [95% CI: 1.145–2.024], p = 0.004), and KPS score < 80 (HR: 1.264 [95% CI: 1.024–1.558, p = 0.029) were identified as risk factors affecting OS of BM in patients with NSCLC. On the other hand, female sex (HR: 0.725 [95% CI: 0.592– 0.889], p = 0.002) and EGFR gene positivity (HR: 0.656 [95% CI: 0.513–0.839], p < 0.001) were identified as significant protective factors associated with longer OS. Multifactor analysis was performed for elements with p-values less than 0.1 in the univariate analysis ( Table 2 ), depicting synchronous BM (HR: 1.335 [95% CI: 1.076–1.657], p = 0.009), pathological type of squamous carcinoma (HR: 1.361 [95% CI: 1.018–1.820], p = 0.037), KPS score < 80 (HR: 1.392 [95% CI: 1.124–1.724], p = 0.002) as risk factors affecting OS of BM in patients with NSCLC.

The univariate and multivariate analysis results for 390 patients with NSCLC-associated BM after PSM were generally consistent with those before PSM. In the univariate analysis following PSM: a male sex, a pathological type of squamous carcinoma, KPS score < 80, and EGFR gene negativity were identified as risk factors for OS in patients with NSCLC exhibiting BM. On the other hand, metachronous BM (HR: 0.771 [95% CI: 0.622–0.957], p = 0.018) emerged as a protective factor for OS. The multifactor analysis was performed on factors with p-values less than 0.1 in the results after univariate analysis ( Table 3 ), identified metachronous BM (HR: 0.715 [95% CI: 0.575–0.889], p = 0.002) as an independent protective factor for OS; whereas the pathological type of squamous carcinoma and KPS score < 80 were identified as independent risk factors for OS in NSCLC-associated BM. Moreover, subgroup analyses according to patients’ clinical characteristics confirmed that metachronous BM was associated with a significantly longer OS in all subgroups ( Figure 2 ).

The Kaplan–Meier curves of OS in patients with BM from lung cancer according to the results of multivariate analysis conducted before and after PSM, under each independent factor, showed that metachronous BM is associated with a significantly longer OS ( Figure 3 ).

To further evaluate the prognostic impact of synchronous versus metachronous BM on patients with NSCLC, we analyzed the relationship between the presence or absence of synchronous BM and the pathological type and EGFR driver genes. The impact on OS in patients with synchronous and metachronous BM showed no significant difference when considering the pathological type of adenocarcinoma (p = 0.075). However, patients with metachronous BM of squamous carcinoma had longer OS than those with synchronous BM (p = 0.01) ( Table 4 and Figure 4A ). In terms of EGFR genes, there was no difference in the effect of synchronous BM with positive EGFR genes versus metachronous BM on OS (p = 0.270). However, metachronous BM with negative EGFR genes had longer OS compared to those with synchronous BM (p = 0.044) ( Table 4 and Figure 4B ).

The distribution characteristics of patients with BM are illustrated in Sankey plots. These plots visualize the relationship between BM status, pathological type, and EGFR gene mutation. The analysis indicates that the pathological type of synchronous BM and metachronous BM is mostly LUSC and most of the LUSC do not exhibit gene mutations ( Figure 5 ).

According to the results of univariate analysis and multivariate analysis using the COX proportional hazards regression model, we incorporated pathological types and EGFR genes into the model. Subsequently, Nomogram plots of the 1-year, 2-year, and 3-year survival probabilities of lung cancer with BM were generated ( Figure 6 ). The Nomogram plots enable the assignment of scores for each variable by drawing an upward vertical line along the score axis for each assigned variable. These scores are then accumulated to determine the prognostic score for each variable. According to the prognosis score of each patient, a vertical line along the probability axis is drawn downward, to obtain the corresponding 1-year, 2-year, and 3-year survival probability values.

The calibration of the nomogram was assessed through calibration plots. Nomogram predicted and actual 1-year, 2-year, and 3-year OS rates were plotted and compared to further validate the predictive performance of the nomogram ( Figure 6 ).

The discriminative ability of the nomogram to predict 1-year, 2-year, and 3-year OS was assessed using the area under the curve (AUC) of the receiver operating characteristic curve (ROC). The C indices of 1-year, 2-year, and 3-year predicted survival were 0.660, 0.631, and 0.685, respectively. The AUC was between 0.5 and 1.0, indicating a decent level of discriminative ability for the nomogram, with 0.5 representing a random outcome ( Figure 6 ).