Baseline study characteristics and quality assessment
We identified 14289 studies in our efficient literature search as shown in Fig. 1, and 36 studies were excluded because they were reduplicative studies. After checking the title and abstract, we retrieved 138 potential studies to review the full-text manuscript. At that point, 95 studies were avoided because they lacked an outcome of interest or since they had no compelling information and control groups. Ultimately, 43 studies met our selection criteria for the final analysis. The characteristics and demographic data of all studies included are presented in Table 1.14-56 The retrieved studies were published from 2004 to 2019, and a total of 11415 patients were recruited. Of all studies, 26 were cohort studies and other 17 were case-control studies.
For quality assessment, the NOS was used for evaluation because all of the included studies were case-control studies or cohort studies. The results showed that all included studies were of fair quality (Table 1).
Meta-analysis of clinical characteristics of patients
A meta-analysis of the relevant studies suggested that the prevalence of BM was significantly higher among female patients (OR=1.18, 95%CI: 1.11-1.25, P<0.00001) (Fig. 2B). Nevertheless, patients with younger than 60 years old (OR=1.11, 95%CI: 0.78-1.59, P=0.55) (Fig. 2A), a smoking history (OR=1.06, 95%CI: 0.78-1.46, P=0.070) (Fig. 2C) and a treatment history (OR=0.84, 95%CI: 0.62-1.14, P=0.27) (Fig. 2D) did not show significant differences between BM with NSCLC group and sample NSCLC group. Obvious heterogeneity was observed among age (I²=99%, P<0.00001) and a smoking history (I²=69%, P<0.00001), thus a random-effects model was utilized. In addition, a fixed-effects model was used for the other indexes as there was no obvious heterogeneity of the above studies.
Meta-analysis of tumor-related indexes
A meta-analysis of the relevant studies suggested that NSCLC patients with adenocarcinoma (OR=2.43, 95%CI: 1.92-3.08, P<0.00001) (Fig. 3A) was a risk factor for BM. Conversely, squamous carcinoma was a protective factor (OR=0.39, 95%CI: 0.27-0.54, P<0.00001) (Fig. 3B). Meanwhile, the prevalence of BM was significantly higher among patients with higher T stage (OR=1.37, 95%CI: 1.04-1.82, P=0.03) (Fig. 3C), higher N stage (OR=2.01, 95%CI: 1.25-3.22, P=0.004) (Fig. 3D), with the number of 1ymphatic metastasis larger than 6 (OR=2.86, 95%CI: 1.89-4.33, P<0.00001) (Fig. 3E), EGFR gene mutation (OR=2.45, 95%CI: 1.65-3.65, P<0.00001) (Fig. 3G) and KRAS gene mutation (OR=2.88, 95%CI: 1.76-4.72, P<0.00001) (Fig. 3H). In contrast, patients with other distant metastasis (OR=1.34, 95%CI: 0.47-3.79, P=0.58) (Fig. 3F) and ECOG scale (OR=1.30, 95%CI: 0.89-1.91, P=0.17) (Fig. 3I) did not show significant differences between two groups. Furthermore, a fixed-effects model was utilized considering no obvious sample heterogeneity in the above studies while a random-effexts model was used for studies with obvious heterogeneity.
Meta-analysis of clinical laboratory parameters
The results showed NSCLC patients had higher levels of CEA (WMD=10.94, 95%CI: 7.47-14.40, P<0.00001) (Fig. 4A), CA199 (WMD=20.23, 95%CI: 12.20-28.26, P<0.0001) (Fig. 4B), CYFRA211 (WMD=1.78, 95%CI: 0.04-3.51, P=0.04) (Fig. 4C), NSE (WMD=9.66, 95%CI: 6.18-13.14, P<0.00001) (Fig. 4D) and CA125 (WMD=22.39, 95%CI: 9.79-34.98, P=0.0005) (Fig. 4E). Obvious heterogeneity was observed among these five indexes (I²＞50%; P<0.05), thus, a random-effects model was utilized.
Meta-analysis of survival rates
Eight studies assessed the survival rates. The resulted indicated that, as time goes on, the survival rates at 1-, 2-, 3-, 5-year of BM with NSCLC patients were significantly decreased compared to that of NSCLC patients (P<0.05) (Fig. 5).
Sensitivity analysis and risk of bias
The Newcastle-Ottawa Quality Assessment scales (Table 1) and GRADE evaluation (Fig. 6, Fig. 7) indicated that the included studies were of acceptable quality. Risk of bias was summarized and evaluated in Fig. 8 and Fig.9 . A sensitivity analysis was conducted to evaluate the influence of each included study, the results showed that heterogeneity and the pooled ORs or WMDs of BM were not significantly altered by any single study, indicating that our conclusions are relatively reliable. Funnel plot were generated for the indexes and further shown in Supplementary Fig. S1.