LAUSR

Dear Editor, This letter reports some observations on the recent article entitled “Meta‐analysis of the prognostic value of CpG island methylator phenotype in gastric cancer” by Powell et al (2018) reporting that gastric cancers showing CpG island methylator phenotype‐high (CIMP‐H) were associated with poor 5‐year survival. The conclusion was reached by a well‐conducted meta‐analysis. However, as the authors claimed, there was significant heterogeneity among the 10 included studies (I = 88%, P < .001), but they applied a fixed‐effects model, which might limit the conclusion. As a result of a lack of a standardized definition of CIMP in gastric cancer, the authors noticed that the conflicting survival results might be caused by the choice of CIMP gene panel. Currently, gene‐specific methylation markers and genomewide DNA methylation profile were the 2 major methods to define CIMP. To limit the heterogeneity among studies and make the conclusion more precise, we regrouped the 10 included studies into 3 subgroups according to different methodologies and the similarities of gene panels, and then carried out subgroup analysis. The studies of Park et al and Shigeyasu et al were classified as the multiple gene panel group because over 15 CIMP marker genes were used in each of them. An et al, Ben Ayed‐Guerfali et al, He et al, Ksiaa et al, and Kusano et al were classified as the p16 or MINT‐based gene panel group, whereas the remaining 3 studies (Chang et al, Chen et al and Liu et al) were classified as the mixed gene panel group because there were no similarities among them. By subgroup analysis, we found that CIMP‐H was significantly associated with poor prognosis in the p16 or MINT‐based gene panel group, but not in the other 2 subgroups (Figure 1A). Interestingly, the overall effects of CIMP‐H in the p16 or MINT‐based gene panel group were almost the same as those in the total of the 3 subgroups (Z = 2.62, P = .009; Z = 2.61, P = .009) and the p16 or MINT-based gene panel group presents 51.5% in weight among the 3 subgroups, suggesting that the overall effects of CIMP‐H in the total of the 3 subgroups were mainly decided by the p16 or MINT‐based gene panel group. As a result of large heterogeneity, we then analyzed the data by the random‐effects model. Surprisingly, in all 3 subgroups, CIMP‐H did not have any association with poor 5‐year survival (P > .05) (Figure 1B). In summary, we may conclude that CIMP is not a prognostic marker in gastric cancer.