Clinicopathological significance and prognosis of long noncoding RNA SNHG16 expression in human cancers: a meta-analysis

Electronic databases including PubMed, EMBASE, Cochrane Library, and Web of Science were searched. The search time was from the establishment of each database to June 20, 2019. The literature search terms included “Small nucleolar RNA host gene 16” or “SNHG16” or “Long non coding RNA SNHG16,” and “cancer” or “carcinoma” or “tumor” or “neoplasm.” The references of relevant literature were tracked for additional relevant studies.

After the literature search, two researchers independently assessed the literature. The inclusion and exclusion criteria are displayed in Table 1.

We recorded the following information: first author, publication date, country, cancer type, number of patients, sample type, sample detection method, cut-off value of SNHG16 expression level, clinical features mentioned above, HR and 95% CI of OS. If HR and 95% CI were provided in the study, we extracted them directly. If the relevant data were not reported, we extracted and analyzed data from Kaplan-Meier curves for OS according to the method described by Tierney [30]. Two investigators independently assessed the data, and when there were differences, a third researcher decided whether or not to include the study. Two researchers independently used the Newcastle-Ottawa Scale (NOS) to evaluate the quality of the included studies. Literature with a score ≥ 6 were defined as high quality.

Meta-analysis was performed with StataSE15.0 (Stata Corporation). Heterogeneity tests were performed based on Cochran’s Q and Chi-square-based I² tests. If P > 0.10, I² < 50% indicates that there is no significant heterogeneity in each study, and statistical analysis was performed using a fixed effects model; otherwise there was significant heterogeneity between the studies and a random effects model was used for the analysis. Subgroup analysis was used to explore sources of heterogeneity. The odds ratio (OR) and 95% CIs were combined to assess the association of SNHG16 expression with clinicopathological parameters, and the HR and 95% CI included in each study were combined to map the forest to evaluate the effect of SNHG16 expression on OS in human cancers. Publication bias was quantified using Begg’s funnel plot and Egger’s test. The reliability of the meta-analysis was tested by a sensitivity analysis. P < 0.05 was considered statistically significant.

A total of 145 articles were retrieved (PubMed (n = 40), EMBASE (n = 52), Cochrane Library (n = 0), and Web of Science (n = 53)). According to the above-mentioned literature inclusion and exclusion criteria, 16 articles [10‐25], consisting of 1299 patients, were finally included. The number of patients in the included studies ranged from 32 to 275 patients. All the research studies were from China. Twelve types of human cancers were included in the meta-analysis, including bladder cancer, cervical cancer, colorectal cancer, esophageal squamous cell carcinoma, gastric cancer, glioma, hepatocellular carcinoma, non-small cell lung cancer, osteosarcoma, ovarian cancer, pancreatic cancer, and papillary thyroid cancer. The expression level of SNHG16 was detected by using qRT-PCR in fifteen studies, and only one study used ISH. OS was reported in fourteen studies, and disease free survival (DFS) and progression free survival (PFS) were reported in only one study. Thus, OS was selected as the major survival outcome for our meta-analysis. HR was extracted directly in five studies and estimated from survival curves indirectly in the other 9 studies. The cut-off values for the expression level of SNGH16 were different in these studies, including the mean, median, and fold change compared with non-tumor tissues, and in the study using ISH, strongly positive samples were defined as having high expression of SNGH16. The summary of screening results of the literature is shown in Table 2, and a flow chart describing the literature search and selection process is provided in Fig. 1.

To demonstrate the clinical features of SNHG16 expression level in human cancers, we analyzed and summarized all the clinicopathological data from the included studies. As shown in Table 3, five studies composed of 373 patients revealed a significant association between SNGH16 overexpression and larger tumor size (OR: 3.357; 95% CI: 2.173–5.185; P < 0.001) using a fixed effects model, and no heterogeneity was found (I² = 0%; P = 0.813). In eight studies including 591 patients, we found that overexpression of SNGH16 had a significant correlation with advanced TNM stage (OR: 2.930; 95% CI: 1.522–5.640; P = 0.001). A random effects model was performed for the analysis of TNM stage because of the heterogeneity (I² = 64.200%; P = 0.007). A total of three studies including 187 patients reporting the relationship of SNGH16 expression with histological grade were analyzed. Our data demonstrated that elevated SNGH16 expression was associated with poor histological grade (OR: 3.943; 95% CI: 1.955–7.952; P < 0.001). Due to no heterogeneity (I² = 13.800%; P = 0.313), a fixed effects model was used. However, no significant relationship between SNHG16 expression and smoking status, sex, distant metastasis and lymph node metastasis was found in the meta-analysis.

As presented in Table 4 and Fig. 2, in total, 14 articles reporting the association between SNHG16 expression level and OS, including 1148 patients, were included in the meta-analysis. The results showed that high SNHG16 expression was significantly correlated with poor OS (HR: 1.866; 95% CI: 1.571–2.216; P <0.001). There was no heterogeneity (I² = 25.800%; P = 0.176) in the data, so a fixed effects model was used. In addition, subgroup analysis for extract method and detection method was performed. The subgroup analysis revealed that the extract method of HR, either the data in paper or survival curves, had a significant influence on OS (data in paper: HR: 2.912; 95% CI: 1.729–4.906; P < 0.001; survival curves: HR: 1.571; 95% CI: 1.155–2.135; P = 0.004), and the heterogeneity results were I² = 13.500%, P = 0.009, I² = 2.260%, P = 0.972, respectively. For the detection method of SNHG16 expression, the overall HR for the qRT-PCR group for OS was 1.830 (95% CI:1.538–2.177, P < 0.001), with no heterogeneity (I² = 20.200%, P = 0.239). Compared with the group with a low expression level of SNHG16, upregulated SNHG16 showed a statistically significant decrease in OS. For the set of cancer types, high expression of SNHG16 was significantly associated with shorter OS in patients with cancers of the urinary system (HR: 2.523, 95% CI:1.540–4.133; P <0.001), digestive system (HR: 2.406, 95% CI:1.556–3.721; P <0.001), and other cancers (including glioma and non-small cell lung cancer) (HR: 1.786, 95% CI:1.406–2.267; P <0.001). However, in terms of the reproductive system and musculoskeletal system, elevated SNHG16 expression was not predictive of unfavorable OS (HR = 1.592, 95% CI: 0.948–2.674, P = 0.079; HR:1.274, 95% CI: 0.727–2.233, P = 0.398, respectively). Besides other cancers, all of the above cancers showed little heterogeneity between them (I² < 50%, P > 0.1). For other cancers, significant heterogeneity was found (I² = 87.9%, P = 0.004) (Table 3), which may be due to the differences between cancers of different systems.

As presented in Table 5, there is just one study providing data on DFS or PFS respectively. We couldn’t make meta-analysis to pool the results. As shown in Table 5, Han et al. reported that high SNHG16 expression was significantly correlated with poor DFS (HR: 4.505; 95% CI: 1.980–10.309; P <0.001). Lu et al. demonstrated that high SNHG16 expression was related with shorter PFS (HR:3.167; 95% CI:1.552–6.231; P<0.021).

To identify whether individual studies had an impact on OS, sensitivity analysis was performed. The results suggested that no single study affected the stability of the HR values, indicating that the results of this meta-analysis data are stable and reliable (Fig. 3a).

The potential publication bias of the meta-analysis was assessed by Begg’s funnel plot and Egger’s test. We observed that the shape of the funnel diagram was almost symmetrical and did not show any signs of significant asymmetry (Fig. 3b). As shown in Fig. 3b, there was no obvious publication bias for OS, as a result of the Begg’ test (P = 0.584) and Egger’s test (P = 0.234) (Table 6). Likewise, there was no obvious evidence for significant publication bias in terms of sex, lymph node metastasis, or TNM stage (Table 6). We did not evaluate the publication bias for smoking, distant metastasis, tumor number, tumor size, and histological grade because the number of included studies was small.