A meta-analysis of tumor necrosis factor-α-308 G>A polymorphism in gastric cancer

Pubmed and Google databases were used to extract all relevant literature on human studies, TNF-α polymorphism. Data from eligible studies were identified with respect to year of publication, first author name, country, source of sample, cases and controls, genotype frequencies, and reported associations. This study has a total of 12 articles on TNF-α-308 G>A and GC. Data on the author's last name, publication year, country of origin, study population source, genotypes, number of cases and controls, and TNF-α-308 genotyping method were extracted. Data were procured independently from each study using a predefined form and conflicts were resolved by discussion as shown in Figure 1.

Figure 1

The genotype frequency and distribution among controls and cases were analyzed by Fisher's test. P < 0.05 is considered statistically significant. Comprehensive meta-analysis V.3 employed for calculation of pooled odds ratios (ORs) and 95% confidence interval (CI). The Cochran Q-test and index (I²) were used to assess the heterogeneity within studies. A Q-test with P < 0.10 indicated significant heterogeneity. A fixed or random-effects model was used to calculate OR and 95% CIs based on heterogeneity strength. I² values range from 4% to 100%, where lower values indicate nonsignificant heterogeneity and higher values indicate a high degree of heterogeneity. Publication bias was investigated by the construction of funnel plots and Egger's regression analysis.

Twelve studies were analyzed in which 12 sets of data were compared with distributions of TNF-α-308 genotypes in patients and healthy controls.

Analysis using Fischer's exact test revealed that TNF-α-308 GA genotype was significantly prevalent in GC cases compared with healthy controls [P < 0.0001, OR (95% CI) = 0.82 (0.74–0.91)], and homozygous genotype AA did not show association with GC risk [P = 0.49, OR (95% CI) = 0.62 (0.44–0.86)], as shown in Table 1, thus indicating possible association of TNF-α-308 GA with the GC susceptibility. However, a meta-analysis of TNF-α-308 genotypes in GC cases and healthy controls revealed nonsignificant association, hence we evaluated 54 articles; after the screening of titles, abstracts and analysis of articles, 12 eligible studies were screened for meta-analysis, including 4,852 GC patients and 5,197 controls. Characteristics of studies and other relevant data, such as genotype frequency in cases and controls, are shown in Table 2. Q-test and I² statistics showed heterogeneity in the included studies, and therefore a random-effects model was employed for genotype comparison; however, nonsignificant heterogeneity was observed in TNF-α-308 AA genotype and therefore fixed-effects model was employed for genotype comparison among cases and controls (Table 3).

The TNF-α-308 GG genotype distribution in patients and controls was 3,200/4,300. Significant heterogeneity among studies with Q = 222.03, I² = 95.04%, and P = 0.00 is indicated in Table 3. Meta-analysis by random-effects model indicated nonsignificant results with pooled OR = 1.65 (0.98–2.76), Z = 1.9, and P = 0.056 as shown in Figure 1A.

TNF-α-308 GA genotype distribution in patients and controls was 891/806 and significant heterogeneity with Q = 109.14, I² = 90.83%, and P = 0.00 (Table 3). Meta-analysis of the TNF-α-308 GA genotype did not show significant results with pooled OR = 0.74 (0.48–1.15), Z = –1.30, and P = 0.19 using random-effects model as indicated in the forest plot in Figure 1B.

TNF-α-308 AA genotype distribution in patients and controls was 66/79. Nonsignificant heterogeneity with Q = 9.39, I² = 4.23%, and P = 0.40 is shown in Table 3. The meta-analysis overall showed nonsignificant results with pooled OR = 1.01 (0.71–1.43), Z = 0.57, and P = 0.95 by fixed-effects model.

TNF-α-308 GA + AA genotype distribution in patients and controls was 1,135/1,054. Significant heterogeneity with Q = 100.59, I² = 89.06%, and P = 0.00 is shown in Table 3. Meta-analysis showed nonsignificant results with pooled OR = 0.73 (0.51–1.04), Z = –1.73, and P = 0.082 by random-effects model.

Begg's funnel plot and Egger's test were used to assess publication bias and results did not indicate evidence of publication bias for TNF-α-308 polymorphisms in cases and controls [(TNF-α-308 GG Begg's test, P = 0.84; Egger's test, P = 0.39), (TNF-α-308 GA Begg's test, P = 1.0; Egger's test, P = 0.95), (TNF-α-308 AA Begg's test, P = 0.60; Egger's test, P = 0.56), and (TNF-α-308 GA Begg's test, P = 1.0; Egger's test, P = 0.95)] as shown in Figure 2.

Figure 2