Statistical analysis
Crude ORs and 95% CIs was applied to assess the association between theCYP1A1 T3801C and A2455G polymorphism with CRC risk. The following genetic models were used: a dominant model (CYP1A1T3801C: (TC + CC) vs. TT and CYP1A1 A2455G: (AG + GG) vs. AA), a recessive model (CYP1A1 T3801C: CC vs. (TT + TC) andCYP1A1 A2455G: GG vs. (AA + AG), an additive model (CYP1A1T3801C: TC vs. TT and CC vs. TT; CYP1A1 A2455G: AG vs. AA and GG vs. AA), and an allelic model (CYP1A1 T3801C: V vs. T andCYP1A1 A2455G: G vs. A).
Heterogeneity among studies was assessed to useP heterogeneity (P h) andI 2 values [24]. A fixed-effects model (Mantel–Haenszel method) [25] was considered if P ≥ 0.10 and/or I2 ≤ 50%; otherwise, a random-effects model (DerSimonian and Laird method) was regarded [26]. Subgroup analyses were calculated by ethnicity, geographic region, gender, location of CRC, and site of CRC. Sensitivity analyses were conducted to investigate the stability of results. The following two methods were considered: (1) One by one exclusion and (2) a data set was built only selecting high-quality and controls of Hardy-Weinberg equilibrium (HWE) studies. HWE was tested by a Chi-square goodness-of-fit test. IfP < 0.05, the controls were considered as Hardy-Weinberg disequilibrium (HWD). Begg’s funnel plot [27] and Egger’s regression asymmetry test [28] were employed to assess publication bias. A nonparametric “trim and fill” method [29] was selected if a obvious publication bias was observed. In addition, a meta-regression analysis was employed to explore the sources of heterogeneity among studies. Furthermore, a Bayesian false discovery probability (BFDP) was used to assess the false positive results [30]. A cutoff value of BRDP was set up to be a level of 0.8 and a prior probability of 0.001 to evaluate whether the significant associations were the false positive. All statistical analyses were calculated applying Stata 12.0 software (STATA Corporation, College Station, TX).