Discussion
Hayashi S et al. In 1992 (Supplemental Table 1) first reported an association on the CYP1A1 T3801C and A2455G polymorphisms and with CRC risk. So far, a lot of studies have performed the above two issues. However, the results are still inconsistent. Moreover, seven previously published meta-analyses [16 22] have arrived at significant disagreement results, especially in different races, as list in supplemental Tables 6 and 7. Moreover, As far as we know that this is the first meta-analysis to investigate the significant results by BFDP test to avoid confounding factors.
The CYP1A1 T3801C polymorphism was associated with rectal cancer risk in overall races and CRC risk in females; the CYP1A1 A2455G polymorphism was associated with an increased CRC risk in overall analysis, Caucasians, Europe countries populations, and males. Five genetic model and several subgroup analyses was conducted, this analysis came at the cost of multiple comparisons, in this instance, the pooled OR with their 95% CI must be adjusted to avoid the false positive results [32]. Wakefield et al. [31] provided a precise Bayesian measure of false discovery to adjust the positive results. The method especially suited for SNP and disease susceptibility study. Therefore, we applied BFDP value to evaluate the positive results in this study. After we further performed a sensitivity analysis and BFDP test, BFDP > 0.8 were found in all analyses.
Therefore, these positive associations might most likely result from false-positive results, rather than from true associations or biological factors. The reason may be because some small sample and low-quality studies were easier to accept if there was a positive report. They tend to yield false-positive results because they may be not rigorous and are often of low-quality. In addition, controls of HWD may be genotype errors and selection bias in molecular epidemiological studies.
Therefore, we performed a sensitivity analysis restricted to studies that only included high-quality articles and controls of HWE.
Supplemental Tables 6 7 list the results of seven previously published meta-analyses [16-22] on the CYP1A1 T3801C and A2455G polymorphisms with CRC risk, respectively. The largest sample size study [19] was performed in 2014 involving 18 studies (7,171 cases and 8,957 controls) for the CYP1A1 T3801C polymorphism; on the CYP1A1 A2455G polymorphism, The largest sample size study [16] was reported in 2016 including 20 studies (8665 cases and 8853 controls). Several studies should be excluded in the above two studies [16, 29]. The present study included 27 studies (9,672 cases and 12,689 controls) on the CYP1A1 T3801C polymorphism and 24 studies (8,933 cases and 11,692 controls) on the CYP1A1 A2455G polymorphism with CRC risk. In addition, four previous meta-analyses [16, 18, 19, 21] reported the association on the CYP1A1T3801C polymorphism with CRC risk, only a meta-analysis [18] observed a significant decreased CRC risk in Chinese population; another three meta-analyses found that the CYP1A1 T3801C polymorphism was not significantly associated with CRC risk; five published meta-analyses [16, 17, 20 22] have been performed to investigate the association on the CYP1A1 A2455G polymorphism with CRC risk. Of these, four studies [16, 20 22] reported a significantly increased CRC risk in Asians and Caucasians and one study [17] observed a significantly increased CRC risk in Asia and Europe, but not in America. These results of previously published studies were not consistent, especially in ethnicity. An obvious inconsistency was found in classification of ethnic groups among these published meta-analyses, cells with red color in supplemental Table 1. Moreover, we still found a significant difference on the HWE of controls in these published studies (cells with blue color in supplemental Table 1). In addition, previously published meta-analyses involved incomplete studies and some repeat studies. All studies did not adjusted significant results for multiple comparisons. Therefore, their meta-analyses are still incredible.
Though all our efforts were made to improve our research, the current study still exist several limitations. First, only published articles were included while positive results are known to be published more readily than negative ones. Second, several subgroup analyses only included small sample size, such as in smoking subgroups (Table 2 and Table 3), the results should be interpreted with caution and it is necessary that a well-designed large sample study to explore the true association. Third, we did not collect the adjusted data because the original study provided little information. Hence, a more precise analysis should be performed when enough data was available in future.
In summary, this study suggests that these positive findings may most likely result from false-positive results, rather than from true associations or biological factors.