Study Selection
According to the PRISMA flow diagram (figure 1), we initially obtained 29,182 studies through primary database searching and 12 through manual searching. After screening the titles/abstracts we selected 374 potentially relevant articles. Among them, 260 were excluded due to non-English language (19 studies) and unsuitable study designs (review papers/case reports/ cross sectional/meta-analysis/experimental; 241 studies). Then, 114 full text studies were checked for their eligibility. Some studies were excluded due to duplicates or irrelevant study design/insufficient information/unqualified articles until we finally obtained 23 included studies.
Eventually, as much as 23 potential articles were included which recruited 9,792 participants and consisted of 3,237 HCC cases and 4,843 controls (table 1). The papers were published 2004 to 2018. Each study has sample size ranged from 45 to 1,624 participants. Of all included studies in this meta-analysis, most studies came from Asia, in which 7 studies were from Mainland China [7,25-30], 4 from Taiwan [16,31-33], 2 from India [19,34], 2 from Korea [8,35], 1 from Japan [36], 1 from Thai [37], and 1 from Turkey [38]. There were also some studies with non-Asian countries, including 2 from Egypt [39,40], 1 from Brazil [41], 1 from Italy [42], and 1 from Tunisia [43]. From the 23 included studies, the aetiology of HCC was mostly caused by HBV (12 studies), followed by mixed cause (8 studies), HCV (2 studies), and alcohol/smoking (1 study). Some studies observed 1 locus, and others observed more than 1 loci. Studies observing polymorphism at -308 was the most frequent (19 studies), while SNP -1031 was the less frequently observed (5 studies). Allele frequencies of SNPs in each population is showed on table 3.
Association between SNP TNF-α -1031 and HCC risk
Only five studies regarding the association between SNP TNF-α -1031 and HCC risk with 825 cases and 1518 controls were available. The number of cases for CC and TC genotypes was reported together in the studies by Niro et al [42] and Jin et al [35] which could only be used for dominant-model analysis (CC/TC vs TT; Table 2). As the heterogeneity among studies for all models (I2) was less than 60% and p-value for the heterogeneity was more than 0.05, fixed-effects models were applied. However, we did not obtain any significant association between SNP TNF-α -1031 and HCC risk in all model analyses. For the estimation of publication bias, we found no visual asymmetry in Funnel Plot analysis.
Association between SNP TNF-α -863 and HCC risk
We included eight studies with 1642 cases and 2746 controls to determine the association between SNP TNF-α -863 and HCC risk. The numbers of cases for CA and AA genotypes were reported together in the study by Niro et al [42], thus it could only be used for dominant-model analysis (AA/CA vs CC; table 2). In studies with heterogeneity among studies (I2) more than 60% and p-value for the heterogeneity was less than 0.05, we applied random-effects models. We found significant association between the allele model of A versus C of TNF-α C/A SNP with the risk of HCC (OR=1.31, 95% CI=1.03-1.67, p=0.03). The dominant-model analysis (CA/AA vs CC) also showed significant association between SNP TNF-α 863 C/A and HCC risk (OR=1.19, 95% CI=1.03-1.36, p=0.02; figure 2). As heterogeneity was found in the statistical analyses, we did sensitivity analyses to evaluate the sources of heterogeneity. We found that heterogeneity between studies was mainly caused by the study by Kummee et al [37], as after this study was excluded, no significant heterogeneity was found.
Association between SNP TNF-α -857 and HCC risk
We only obtained five studies about the association between SNP TNF-α -857 and HCC risk with 716 cases and 1005 controls. The number of cases for TT and CT genotypes was reported together in the studies by Jin et al [35] which could only be used for dominant-model analysis (TT/CT vs CC; table 2). As the heterogeneity among studies for all models (I2) was less than 60% and p-value for the heterogeneity was more than 0.05, fixed-effects models were applied. A significant association between SNP TNF-α -857 C/T and HCC risk was found in dominant-model analyses (OR=1.31, 95% CI=1.06-1.62, p=0.01; figure 3). We did not observe any visual asymmetry in Funnel Plot analysis regarding the publication bias.
Association between SNP TNF-α -308 and HCC risk
Interestingly, we found most included studies analyzing the association between SNP TNF-α -308 G/A and HCC risk. The number of cases for AA and GA genotypes was reported together in the studies by Niro et al [42] and Jin et al [35] which could only be used for dominant-model analysis (AA/GA vs GG; table 2). In studies with heterogeneity among studies (I2) more than 60% and p-value for the heterogeneity was less than 0.05, random-effects models were applied. All five genetic models showed significant association with risk of HCC with OR for allele model=1.98, 95% CI=1.62-2.42, p<0.001; OR for dominant model=1.95, 95% CI=1.53-2.49, p<0.001; OR for recessive model=2.52, 95% CI=1.69-3.76, p<0.001; OR for codominant major vs minor homozygote model=3.14, 95% CI=2.06-4.79, p<0.001; and OR for codominant heterozygote vs major homozygote model=2.07, 95% CI=1.60-2.68, p<0.001; figure 4). As heterogeneity was found in the statistical analyses, we did sensitivity analyses to evaluate the sources of heterogeneity. We found that heterogeneity between studies was mainly caused by the studies by Ho et al [31] and Akkiz et al [38], as after these studies were excluded, no significant heterogeneity was found.
Association between SNP TNF-α -238 and HCC risk
Nine studies with 831 cases and 1293 controls were included to determine the association between SNP TNF-α -238 and HCC risk. Similar to the TNF-a -863 SNP, the numbers of cases for GA and AA genotypes were reported together in the study by Niro et al [42], thus it could only be used for dominant-model analysis (GA/AA vs GG; table 2). In studies with heterogeneity among studies (I2) more than 60% and p-value for the heterogeneity was less than 0.05, we applied random-effects models. The allele model of A versus G of SNP TNF-α -238 G/A was significantly associated with risk of HCC (OR=1.50, 95% CI=1.16-1.94, p=0.002). The codominant-model analysis (AA vs GG) also showed significant association between SNP TNF-α 238 G/A and HCC risk (OR=3.87, 95% CI=1.32-11.34, p=0.01). The recessive model analysis (AA vs GA+GG) proved significant association between SNP TNF-α 238 G/A and HCC risk as well (OR=2.67, 95% CI=1.17-6.10, p=0.02; figure 5). As heterogeneity was found in the statistical analyses, we did sensitivity analyses to evaluate the sources of heterogeneity. We found that heterogeneity between studies was mainly caused by the study by Teixeira et al [41], as after this study was excluded, no significant heterogeneity was found.
Sensitivity analysis
In this meta-analysis, there were no significant change in ORs by deleting a particular study, which indicated that no single study influenced the statistical significance of the overall results.