3.1 Screening of studies
Totally, 479 articles were identified through the original search strategy (Figure 1). Of them, 218 duplicate articles were ruled out, and an additional 61 were eliminated due to the lack of relevance to this study after their abstracts were read. Later, the full-texts of the remaining 59 studies were carefully read; 44 satisfied our study inclusion criteria and were enrolled for analyses. In detail, 12 studies presented data on the RASSF1A promoter methylation rate within HCC and assessed the association of this methylation with clinicopathological characteristics. Besides, 29 studies only assessed the frequency of RASSF1A promoter methylation, while 3 only evaluated the clinicopathological characteristics.
3.2 Characteristics of the enrolled articles
The features of the enrolled articles were shown in Table 1. Altogether, 44 case-control studies involving 4,777 individuals published from 2002 to 2019 were enrolled in the analyses [10, 17-59]. Twenty-eight articles originated in Asia, consistent with the epidemiology of HCC. America produced the second highest number of enrolled papers (n = 8), followed by Africa (n = 5), while Italy and Germany published one article each. Three types of sample sources were predominantly investigated, including tissues (n=31), peripheral blood (n=11), and both tissues and peripheral blood (n=2). In all our enrolled studies, HCC patients were regarded as ‘cases’; those without the disease were considered ‘controls’, and were assigned to the non-tumor group and normal group. Of those articles examining HCC risk, 11 used blood, 31 adopted tissues, and two examined both blood and tissues. Heterogeneous methods were adopted for the detection of the RASSF1A methylation status among the enrolled studies. The Newcastle-Ottawa scale (NOS) was adopted to assess the quality of the 41 articles that reported the RASSF1A methylation rates in the cases and controls, with scores ranging from 5 to 8, indicating a relatively high methodological quality (Table S1). Another three studies that only reported on the disease’s clinicopathological characteristics were not eligible for NOS assessment, and were thus not evaluated.
3.3 Effect of RASSF1A promoter hypermethylation on HCC in the pooled analyses
3.3.1 Comparison of RASSF1A promoter hypermethylation between HCC and non-tumor groups
Data from 34 studies including 2,075 HCC patients and 2,276 non-tumor controls underwent meta-analyses for the evaluation of the effect of RASSF1A promoter hypermethylation on HCC risk (Figure 2). We found that the frequency of RASSF1A gene promoter hypermethylation was remarkably related to a high HCC risk in the overall comparison (odds ratio [OR] = 6.87, 95% confidence interval [CI] = 4.98-9.50, P < 0.001), and moderate heterogeneity was present (I2 = 64.1%, P = 0.000).
Further subgroup analyses stratified by sample type (blood and tissue), detection method (methylation-specific polymerase chain reaction [MSP] and others) and sample size (≥100 and <100) were also performed to explore the possible heterogeneity sources across the various articles enrolled. Subgroup analyses stratified by sample type showed that RASSF1A gene promoter hypermethylation was significantly associated with HCC risk (blood: OR = 6.93, 95% CI = 4.12-11.65, P<0.001; tissue: OR = 7.12, 95% CI = 4.78-10.59, P<0.001). In addition, in the subgroup analysis stratified by the detection method, RASSF1A gene promoter hypermethylation was evidently related to HCC risk (MSP: OR = 7.30, 95% CI = 5.17-10.29, P<0.001; others: OR = 6.20, 95% CI = 3.13-12.30, P<0.001). Similarly, the pooled results were consistent between the subgroups stratified by sample size (≥100: OR = 6.74, 95% CI = 4.28-10.61, P<0.001; <100: OR = 6.67, 95% CI = 4.46-10.00, P<0.001) (Figure 3).
3.3.2 Comparison of RASSF1A promoter hypermethylation between HCC and normal groups
Totally, 26 studies enrolling 1,898 HCC patients and 1,002 normal controls were pooled for the assessment of how RASSF1A promoter hypermethylation affects HCC risk (Figure 4). In the meta-analysis, the promoter methylation of RASSF1A was related to HCC risk in the cancer samples relative to the controls (OR = 31.05, 95% CI = 13.73–70.20, P < 0.001); in addition, a high heterogeneity level was detected across the various articles (I2 = 79.6%, P = 0.000).
Subgroup analyses revealed that the promoter methylation of RASSF1A was significantly correlated with the risk of HCC in all the subgroups stratified by sample type, detection method and sample size (Figure 3).
3.4 Relationship of the promoter hypermethylation of RASSF1A with the clinicopathological features
This study investigated a total of 11 characteristics from 15 studies that investigated the correlation of RASSF1A gene promoter hypermethylation with the clinicopathological features of HCC. The comprehensive data on the numerous clinicopathological features associated with HCC, and the association with the RASSF1A gene was presented in Table 2. As shown in the pooled analyses, RASSF1A promoter hypermethylation was remarkably related to tumor size (≥5 cm vs. < 5 cm, OR = 1.92, 95% CI = 1.07-3.42, P = 0.028) and hepatitis B virus (HBV) infection (positive vs. negative, OR = 1.50, 95% CI = 1.05-2.14, P = 0.026), but was not significantly associated with sex (male vs. female, OR = 1.36, 95% CI = 0.95-1.96, P = 0.094), age (≥50 vs. < 50, OR = 1.74, 95% CI = 0.82-3.69, P = 0.152), hepatitis C virus (HCV) infection (positive vs. negative, OR = 0.93, 95% CI = 0.20-4.26, P = 0.928), level of alpha fetoprotein (AFP) (≥20 μg/L vs. < 20 μg/L, OR = 1.25, 95% CI = 0.47-3.27, P = 0.657), tumor number (multiple vs. single, OR = 0.80, 95% CI = 0.47-1.36, P = 0.410), liver cirrhosis (presence vs. absence, OR = 1.06, 95% CI = 0.60-1.87, P = 0.834), histopathological stage (I+II vs. III+IV, OR = 1.84, 95% CI = 0.53-6.36, P = 0.338), tumor differentiation (poor vs. moderate or well, OR = 0.91, 95% CI = 0.41-2.02, P = 0.820) or portal venous invasion (presence vs. absence, OR = 0.61, 95% CI = 0.16-2.40, P = 0.481).
3.5 Meta-regression and sensitivity analyses
As for the results of the pooled meta-regression analysis on the correlation between the promoter hypermethylation of RASSF1A and HCC risk in both groups, a trend for sample type, detection method and sample size was demonstrated (Table S2). Heterogeneity was detected in the pooled results; as a result, this study evaluated the contributions of diverse investigated features to heterogeneity. Nonetheless, there was no statistical significance (all P values > 0.05, Table S2). The heterogeneity proportions in both groups ranged from -9.70% to 8.14% (all P values > 0.05), with a high level of residual heterogeneity (τ2 range, 0.506-3.226). Owing to a lack of sufficient data in the enrolled articles, this study did not incorporate other factors that possibly contribute to heterogeneity into the meta-regression analyses.
To further investigate the robustness of the pooled results in both groups by sensitivity analyses, a random-effects model was adopted to eliminate one study at a time. None of the studies had a significant influence on the pooled results, indicating that our estimates were robust and reliable (Figure S1).
3.6 Publication bias
With regards to the non-tumor group, the funnel plot appeared to be asymmetric (Figure S2A), and statistical significance was observed in Begg’s test (P = 0.021), which raised the possibility of publication bias, although no significant publication bias was discovered in Egger’s test (P = 0.208). Subsequently, the “trim and fill” method was adopted for the evaluation of the possible impact of publication bias on the pooled effect. In consequence, the symmetric funnel plot was generated through the filling of 10 hypothetical negative articles (Figure S2B). Typically, the adjusted OR obtained from the pooled analysis incorporating these hypothetical studies was still significant (OR = 5.14, 95% CI = 3.69-7.16, P < 0.001). Similarly, for the normal group, both Egger’s test (P < 0.001) and the funnel plot revealed the presence of potential publication bias (Figure S2C), regardless of the absence of statistical significance in Begg’s test (P = 0.332). Thereafter, seven hypothetical negative studies were filled through the “trim and fill” approach, but RASSF1A promoter methylation was found to be significantly associated with HCC risk in the pooled analyses (OR = 15.71, 95% CI = 7.40-33.36, P < 0.001) (Figure S2D).
3.7 Association of the promoter hypermethylation of RASSF1A with HCC-related prognoses
3.7.1 Baseline patient characteristics
Data on the promoter methylation of RASSF1A were identified within DNA methylation profiles from 380 The Cancer Genome Atlas (TCGA)-derived HCC as well as 50 non-carcinoma samples. Based on UCSC assembly-Dec.2013 (GRCh38/hg38), 11 probes situated at the promoter region of RASSF1A were selected (including cg13872831, cg24859722, cg04743654, cg00777121, cg08047457, cg12966367, cg21554552, cg25747192, cg06172942, cg25486143, cg27569446), and they contained the RASSF1A gene CpG island A (chr3: 50340373–50341109). In the TCGA cohort, the RASSF1A promoter methylation levels within the HCC samples significantly increased compared to those in the adjacent non-carcinoma liver tissues (Figure S3). According to the probe methylated levels, all samples were classified into the hypomethylated (n=196) and hypermethylated (n=184) groups. Among the 380 TCGA-derived HCC samples, 349 had information available on overall survival (OS) and survival status, while 342 had data on disease-free survival (DFS) and recurrence status.
3.7.2. RASSF1A promoter hypermethylation in the prediction of HCC-related prognoses
In the Kaplan–Meier survival analysis, HCC cases with RASSF1A gene promoter hypermethylation were found to have poorer OS (median OS: 3.90 years vs 6.73 years; P = 0.0206) and DFS (median DFS: 1.38 years vs 3.01 years; P = 0.0003) values than the hypomethylated cases (Figures 5A and 5C). Additionally, receiver operating characteristic (ROC) curve analysis was conducted for the determination of the sensitivity and specificity of RASSF1A gene promoter hypermethylation in prognosis prediction. The areas under the curve (AUCs) pertaining to RASSF1A gene promoter hypermethylation in the prediction of the OS of HCC patients at 1, 2, 3 and 5 years were 0.51, 0.60, 0.60 and 0.58, respectively (Figure 5B). Meanwhile, the time-dependent AUC values concerning RASSF1A gene promoter hypermethylation in the prediction of the OS of HCC patients at 1, 2, 3 and 5 years were 0.61, 0.69, 0.63 and 0.73, separately (Figure 5D). Accordingly, we inferred that RASSF1A gene promoter methylation status exhibited high sensitivity and specificity.