The data about cancer in the SEER database is continually reported in every state of the United States and retrieved with no need for informed patient consent. The present study complied with the Declaration of Helsinki (as revised in 2013).

Data in this population-based study were abstracted from the SEER. SEER*Stat Software version 8.3.4 (https://seer.cancer.gov/seerstat/; Information Management Service, Inc., Calverton, MD, USA) was used to generate the case listing.

Our data came from the National Cancer Institute’s (NCI) SEER database, andbecause SEER did not record BM information until 2010.Patients diagnosed before that year were excluded. Therefore, a retrospective cohort study was conducted on patients diagnosed from 2010 to 2016. SEER is an open access U.S. cancer database from 18 population-based cancer registries. SEER currently collects and publishes cancer incidence and survival data covering approximately 28% of the U.S. population, which is representative of the population.

We screened patients from the SEER database with pathologically diagnosed NSCLC, including adenocarcinoma, squamous cell carcinoma, and adenosquamous cell carcinoma, according to the International Classification of Diseases of Cancer Version 3 (ICD‐O‐3) histological codes 8140, 8070, 8560, 8046. Ethical approval and informed consent were waived because the SEER data were freely available and our investigation was retrospective. The exclusion criteria for NSCLC patients in this study were as follows: (I) no BM; (II) patients with multiple primary malignant tumors; (III) patients with NSCLC whose survival was less than one month or whose survival data were not available were excluded. We excluded patients who did not undergo primary pneumonectomy and mediastinal lymphadenectomy, such as local lobectomy, lymphadenectomy, laser ablation, or cryotherapy. Follow-up was from diagnosis of NSCLC to death or the end of the follow-up period.

The purpose of this study was to compare the benefits of primary lung resection with mediastinal lymph node dissection in patients with NSCLC with BM. This was a retrospective study, so the surgery assignment was not random. Some of the key covariates of patients in the active and control groups were heterogeneous and could have influenced the results. Therefore, we further compared the difference in survival between the surgical and untreated groups by univariate analysis using 1:2 nearest neighbor matching and setting the caliper value to 0.02. The PSM process has been applied to minimize selection bias and roughly balance baseline covariates in an intergroup set of analyses (12).

The primary endpoint of the study was overall survival (OS). Chi-square test was used to compare the characteristics of surgical patients and control patients. Covariates in this study included multilevel factors (such as age, sex, race, marital status, insurance status, tumor tissue type, tumor size, lymph nodes, degree of differentiation, chemotherapy and radiotherapy). SEER data recorded a small number of tumor lesions larger than 20cm. We thought these might be incredible, so we included them in part to understand the participation statistics. Propensity scores are used to reduce selection bias. Kaplan‐Meier analysis was used to estimate OS before and after PSM. Log-rank tests were performed to compare survival differences in patients, lesions, and treatment-related characteristics. To perform a multivariate analysis in a matched population, we constructed a Cox proportional risk model to identify predictors ofsurvival. P value<0.05 considered that the difference was statistically significant. All statistical analyses were performed using IBM SPSS Statistics 22.0 (IBM, Armonk, NY, USA).

From 2010 to 2016, 188,840 patients with newly diagnosed NSCLC were identified in the SEER data set. A total of 21,811 patients diagnosed with advanced NSCLC with BMs were selected based on inclusion criteria described in the study population. Of these, 15,703 met the inclusion criteria for this study. The surgical and untreated groups included 203 (1.31%) and 15,500 (98.69%) patients, respectively ( Figure 1 ).

Kaplan‐Meier analysis showed that overall survival was significantly improved in patients who underwent surgery compared to the control group (P<0.001, Figure 2A ). Median survival was 27 months for patients who underwent primary lung resection with mediastinal lymph node dissection at the start of the NSCLC diagnosis, compared with the control group. After matching patients undergoing surgical treatment with the propensity score, we balanced nearly all available covariates between groups, while a few covariates such as age, degree of tumor differentiation, tumor size, and tumor N-stage showed differences. After excluding the mismatched population, 203 surgical patients and 406 untreated patients were matched at 1:2 PSM ( Table 1 ). To balance covariates, significant differences in survival time were also observed between patients treated surgically after PSM matching and those who did not receive treatment (P<0.001, Figure 2B ).

Table 2 lists the median survival results of univariate Kaplan-Meier analysis in the matched population. During surgery (P < 0.001), age (P < 0.001), sex (P = 0.010), marital status (P = 0.014), pathological type (P < 0.001), degree of differentiation (P= 0.001), tumor size (P < 0.001), lymph node (P < 0.001), chemotherapy (P < 0.001). There was significant difference in survival rate among covariables. Patients who did not undergo primary resection with mediastinal lymph node dissection, age >65 years, male, divorced, poorly differentiated or undifferentiated, tumor size >5cm, mediastinal lymph node metastasis, and did not undergo chemotherapy had poor survival. Table 3 shows the multivariable predictors of mortality for the propensity matched sample. Patients in the propensity matching sample who underwent surgery had a significant reduction in mortality compared with those who did not (HR: 0.243, 95%CI: 0.162-0.365, P < 0.001). Age >65 years, squamous cell carcinoma, mediastinal lymph node metastasis, and no chemotherapy were independentlyassociated with higher mortality (P < 0.001, < 0.001 and < 0.001, Table 3 ). Mortality was higher in male, divorced, poorly differentiated or undifferentiated, and tumor size >5cm than in female, married, well-differentiated or moderate differentiated, and tumor size ≤5cm.

Multivariate analysis assessed the predictors of survival in surgical patients, and the results were shown in Table 4 . Patients >65 years of age who received surgical treatment had an increased risk of death compared with patients ≤65 years of age (HR= 1.587, 95%CI: 1.027-2.453, P= 0.034, Table 4 ). Patients who underwent surgery for squamous cell carcinoma had an increased mortality rate compared with those who underwent surgery for adenocarcinoma (HR= 2.009, 95%CI: 1.180-3.420, P = 0.010). Survival was significantly lower in patients undergoing chemotherapy than in patients not undergoing chemotherapy (HR = 2.555, 95%CI: 1.640-3.979, P < 0.001).

After PSM matching, differences in age, tumor differentiation, tumor size, and lymph nodes still existed between groups, so subgroup analysis was further performed. By age subgroup analysis, overall survival after PSM was longer in patients who underwent surgery than in patients who did not receive treatment (P < 0.001 and P < 0.001, Figure 3 ). In the subgroup of tumor differentiation degree, the survival rate of surgery group was significantly higher than that of untreated group (P < 0.001 and P < 0.001, Figure 4 ). In the tumor size subgroup, survival was significantly higher in the surgically treated group than in untreated patients (P < 0.001 and P < 0.001, Figure 5 ). Based on subgroup analysis of lymph nodes, patients who underwent surgery had significantly higher survival rates than untreated patients(P < 0.001 and P < 0.001, Figure 6 ).

There are some limitations in this study, which may affect our research results to some extent. The SEER database did not record data on patient complications, smoking history, and physical conditions. It is well known that some factors such as the number of BM, genetic changes and the treatment of BM affect the prognosis and clinical efficacy of patients (30, 35, 36), but the SEER database does not provide relevant information. These limitations may have some effect on overall prognosis, and we hope to refine this section in future prospective studies. Although the database contains specific radiotherapy and surgical methods, the lack of information on specific chemotherapy schemes has a certain impact on surgical evaluation, which is also the limitation of this paper. Therefore, we need more detailed data to assess the efficacy and adverse effects of surgery. Finally, although PSM was used to reduce selection bias in the surgical group, the retrospective nature of this study makes it difficult to avoid bias in other confounding factors.