Characteristics and eligible studies
A total of 182 studies were initially identified as potential articles, and 103 studies were excluded as duplicates. After reviewing titles and abstracts, 44 studies were excluded since they were non-comparative studies or irrelevant topics. Then, 35 potentially eligible articles were selected for full-text assessed, and 17 studies were excluded due to the lack of survival data. Thus, 18 studies compromising 1290 patients were considered eligible in the light of the inclusion and exclusion criteria. The screen procedure was thoroughly implicated via a flow diagram in Figure 1.
The characteristics of the eligible studies are presented in Table 1. These studies were published between 2017 and 2020, and their sample size ranged from 20 to 199. A total of ten different cancer types were included in our meta-analysis, including prostate cancer, gastric cancer, colorectal cancer, renal cell carcinoma, osteosarcoma, nasopharyngeal carcinoma, glioblastoma, cervical cancer, breast cancer, and hepatocellular carcinoma. Among these 18 studies, quantitative real-time polymerase chain reaction (qRT-PCR) was used as detection assay in 17 studies, and fluorescence in situ hybridization (FISH) analysis was performed in one study. As for survival outcomes, association between SNHG12 expression level and OS were reported in all studies except for three studies only reporting RFS and clinicopathological outcomes respectively. In all included studies, patients were divided into high or low SNHG12 expression groups according to the cut-off value. Moreover, the follow-up months ranged from 45 to 160 months, and univariate or multivariate analysis were used in each survival analysis. As for clinical stage, there were four kinds of clinical stage classification system, including tumor node metastasis (TNM) classification system, the International Federation of Gynecology and Obstetrics (FIGO) stage, Enneking stage, and The World Health Organization (WHO) grade. Additionally, all eligible studies were considered as high methodological quality with their NOS scores ≥7.
Association between lncRNA SNHG12 and OS/ RFS
A total of 15 studies were included for OS analysis. Since no obvious heterogeneity was observed among these studies (I2=0.0%, p=0.967), fixed-effects model was employed to synthesize pooled HR and corresponding 95% CI. The aggregated data suggested that high expression level of SNHG12 was significantly correlated to poor OS (HR=1.97, 95%CI 1.56-2.48, p<0.001) (Figure 2A), indicating that lower SNHG12 expression in cancer patients may predict a better survival outcome.
Two studies regarding prostate cancer and hepatocellular carcinoma provided related data for RFS analysis. In the absence of apparent heterogeneity among these studies (I2=0.0%, p=0.380), fixed-effects model was applied to calculate the HR and its 95%CI. As demonstrated in Figure 2B, the pooled results indicated that high SNHG12 expression level predicted unfavorable RFS in prostate cancer and hepatocellular carcinoma (HR=1.71, 95%CI 1.05-2.78, p<0.05).
Sensitivity analysis was performed in order to assess whether any individual study would affect the result of pooled OS. By removing each included study, we found that the pooled result had a slight fluctuation when “Zhang, R 2019” was removed (Figure 2C). Thus, the pooled HR was analyzed again after omitting “Zhang, R 2019”, and the result demonstrated that high expression of SNHG12 was still correlated to worse OS in different kinds of cancers (HR=2.02, 95%CI 1.57-2.59, p<0.00001, and I2=0.0%, p=0.957, fixed model), indicating the stability and reliability of this meta-analysis.
Begg`s funnel plot and Egger`s regression test were employed to evaluate potential publication bias. As shown in Figure 2D, no apparent asymmetry was observed in the Begg`s funnel plot and the result of Egger`s regression further proved it (p>|t|=0.160). Therefore, no significant publication bias existed in this meta-analysis.
Subgroup analysis of association between SNHG12 and OS
Even though the study heterogeneity was low in OS analysis (I2=0.0%, p=0.967), several stratified analyses were performed based on tumor type (digestive system tumor or others), sample size (more or less than 60), survival analysis method (univariate or multivariate analysis), and cut-off value (mean or median). As shown in Figure 3 and Table 2, all subgroup analyses based on different stratified factors did not alter the association between SNHG12 and OS in multiple kinds of cancers.
Association between SNHG12 and clinicopathologic characteristics
ORs and corresponding 95%CI were applied to investigate the association between SNHG12 and clinicopathologic features including age, gender, tumor size, Gleason score, TNM stage, WHO grade, LNM and DM. Fixed-effect model was applied in all analyses and no statistically significant correlation was detected between SNHG12 and age, gender (Supplementary Figure 1, Table 3). Recognizing, as demonstrated in Figure 4 and Table 3, high SNHG12 expression had significant association with larger tumor size (OR=5.05, 95%CI 2.67-9.55, p<0.00001), LNM (OR=3.32, 95%CI=2.32-4.75, p<0.00001), DM (OR=2.35, 95%CI 1.46-3.78, p=0.005), poorer TNM stage (OR=3.61, 95%CI 2.51-5.17, p<0.00001), higher WHO grade (OR=11.34, 95%CI 4.60-27.95, p<0.00001) and higher Gleason score in prostate cancer (OR=2.69, 95%CI 1.59-4.53, p=0.0002). We could not assess the association between SNHG12 expression and other clinicopathological parameters owing to insufficient data.
Online cross-validation in TCGA dataset
We used TCGA dataset to evaluate SNHG12 expression levels in multiple kinds of cancers in order to further validate the pooled results. As depicted in Figure 5, SNHG12 showed aberrant expression levels in breast invasion carcinoma (BRCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), liver hepatocellular carcinoma (LIHC), colon adenocarcinoma (COAD), rectum adenocarcinoma (READ), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), sarcoma (SARC), and stomach adenocarcinoma (STAD) when compared with normal control. Moreover, the violin plot implicated that SNHG12 expression level was significantly correlated with pathological stage in human pan-cancers. Additionally, the survival plots in GEPIA indicated that high expression of SNHG12 predicted worse OS (HR=1.1, p<0.05) and DFS (HR=1.1, p<0.05), which verified our results in this meta-analysis.