To the Editor, We read with great interest the two recent meta-analysis of the association between Tumor necrosis factor(TNF)-a 308G/A polymorphism and breast cancer risk [1, 2] and the two related letters [3, 4] which commented on the two articles. The meta-analysis by Fang et al. [1] included 11 studies with 10,184 breast cancer cases and 12,911 controls, and the results suggested that TNF-a308G allele might be a modestly risk factor for the development of breast cancer among Caucasian and the overall. However, the other one by Shen et al. [2] comprised 13 studies involving 10,236 cases and 13,143 controls, and the results indicated that TNF-a308A allele might slightly decrease the risk of breast cancer among Caucasian, but not among the overall. The partial difference (for the overall population) on the conclusions between two studies is due to the eligible studies they identified. As commented by Chen et al. [3] and Zhou et al. [4], the article by Fang et al. contained one study not in Hardy–Weinberg equilibrium [5] and omitted two eligible studies [6, 7]; while the article by Shen et al. [2] included one duplicate study [8]. These findings suggested that the incomprehensive search of publications would bias the results. In addition, we have three further comments on the Rebuttal Comments by Fang et al. [3] and Shen et al. [4]. First, Fang et al. [3] mentioned that the study by Kohaar et al. [6] did not supply the frequencies of GA and AA genotypes. In fact, this information was provided in the main text (see the fourth paragraph of the ‘‘Result’’ section). Second, the opinions of Shen et al. [4] suggested that PubMed searching might be enough to indentify the eligible studies. We agree that PubMed database is one of the most comprehensive sources of medicine information worldwide, while its coverage is incomplete like any database [9, 10]. Other databases, such as ISI Web of Knowledge, Embase, and even non-English databases also should be searched to avoid the introduction of selection bias and language bias [10, 11]; or at least, both Medline and hand searching should be used when the meta-analysis is performed [10]. Third, the reference number in the ‘‘References for Rebuttal letter’’ section is not in accord with that in the main text of Rebuttal letter [4].
Read full abstract