To the Editors: Our recent meta-analysis (1) detected a moderate but statistically significant association between IL1B-511T and gastric cancer risk in Caucasians, but not in Asians. This contrasts to the results of the meta-analysis by Kamangar et al. (2), which reported a weak, statistically nonsignificant association. To explain this contrast, it is worth considering differences in the two meta-analyses.
Both analyses examined population subgroups, but groupings differed. Kamangar et al. combined Central/South American studies with those from European/North American populations and contrasted them to Asian studies. We examined European/North American populations separately. Asian, Central American, and South American populations have an IL1B-511T allele frequency of >50%, in contrast to ∼33% in Europeans/North Americans. A real association between a genetic polymorphism and disease risk will seem weaker in populations with a high risk-associated allele frequency, when the measure of the association is relative risk; thus, grouping populations that vary in the relevant allele frequencies may obscure the association.
The two meta-analyses used different genetic models. Kamangar et al. compared IL1B511 CT versus CC and TT versus CC. We compared IL1B-511T carriers versus CC, a dominant genetic model, selected by the algorithm of Thakkinstian et al. (3). This algorithm allows the data to determine the genetic model and reduces the number of comparisons in the analysis.
The study by zur Hausen (included by Kamangar et al., as ref. 41, but omitted by us), described genotyping of cases from tumors, not from germline DNA. Because gastric cancers may undergo loss of heterozygosity in the chromosome 2q region containing IL1B (4), genotyping IL1B from tumor DNA may underrepresent heterozygotes.
In three European studies, Kamangar et al. noted overlapping samples, a duplication that we overlooked (refs. 4, 6, and 22 in Kamangar et al.). After removing two of the duplicated studies (see the table for revised data), the association between IL1B-511T and gastric cancer risk that we previously detected in Caucasians remains statistically significant, as does the association between gastric cancer risk and carriers of allele 2 of IL1RN versus all other genotypes, under the dominant model, for high-quality studies (random-effects odds ratio, 1.47; 95% confidence interval, 1.24-1.75).
Group of analysis for IL1B-51 . | P heterogeneity . | Random-effects OR (95% CI) . |
---|---|---|
Caucasians | 0.02 | 1.39 (1.05-1.85) |
Intestinal subtype cases | 0.02 | 1.55 (1.02-2.37) |
Noncardia cases | 0.42 | 1.63 (1.21-2.20) |
Quality score ≥5 | 0.45 | 1.77 (1.45-2.16) |
Group of analysis for IL1B-51 . | P heterogeneity . | Random-effects OR (95% CI) . |
---|---|---|
Caucasians | 0.02 | 1.39 (1.05-1.85) |
Intestinal subtype cases | 0.02 | 1.55 (1.02-2.37) |
Noncardia cases | 0.42 | 1.63 (1.21-2.20) |
Quality score ≥5 | 0.45 | 1.77 (1.45-2.16) |