Potential conflict of interest: Nothing to report.
Article first published online: 24 JUN 2011
Copyright © 2011 American Association for the Study of Liver Diseases
Volume 54, Issue 1, pages 375–376, July 2011
How to Cite
Guo, X., Johnson, R. and O'Brien, S. J. (2011), Reply:. Hepatology, 54: 375–376. doi: 10.1002/hep.24448
- Issue published online: 24 JUN 2011
- Article first published online: 24 JUN 2011
- Accepted manuscript online: 26 MAY 2011 09:34AM EST
- Manuscript Accepted: 14 MAY 2011
- Manuscript Revised: 11 MAY 2011
- Manuscript Received: 9 MAY 2011
We do appreciate that there are legitimate differences of opinion with respect to correction for multiple tests as the letter by Dr. Liu aptly illustrates. Our study1 was designed to replicate a previously published previous genome-wide association study.2 One opinion is, in this case, that a correction for multiple comparisons is unnecessary. In our study, however, we still address it. We state that a Bonferroni test may be too conservative given the nonindependence of adjacent single-nucleotide polymorphisms (SNPs) strongly associated by linkage disequilibrium3 (see Guo et al.1 for our detailed discussion of linkage disequilibrium statistics on genome-wide association studies). Nonetheless, even if we concede to the more stringent statistical correction offered by Dr. Liu in his comment (0.05/8 = 0.00625 for table 2 and 0.05/12 = 0.00417 for table 3), nearly all the associations for SNP alleles (table 2) or genotypes (table 3) still achieve statistical significance after correction. Only rs3135021 fails significance. We feel that our data are robust and statistically strong in support of the associations of the SNP alleles, genotypes, and haplotypes illustrated in table 2 to table 4 and figure 1.
The statistical power of our study for the additive model is 89%-100%, based on case number, control number, disease prevalence (0.08), significance level (0.05), disease allele frequency (0.334–0.797), and genotype relative risk (1.5). Also, the data reported in table 2 were actually from Armitage's trend test (computed with SAS Genetics software), which is most useful when there is an additive allele effect on the disease susceptibility.
As for the direction (i.e., the polarity of odds ratios [ORs]), this format is also a matter of taste. We chose to base our tables on the risk allele (protective allele as a reference) with OR < 1.0, which we state. For example, for rs2395309, the OR should be 1/0.71 = 1.41, and 95% confidence interval should be 1.16–1.69 (formula is 1/0.86 = 1.16, 1/0.59 = 1.69) for risk allele G. We realize that use of the term “risk allele” might be confusing, because some alleles protect whereas others are susceptible, as indicated by the OR. For this confusion, we stand corrected. We apologize for the confusion, but when one examines the OR, it seemed clearly stated to us.
For table 3, we list the reference group (OR = 1.0), and the most common genotype was used as a reference for each SNP.
- 1Strong influence of human leukocyte antigen (HLA)-DP gene variants on development of persistent chronic hepatitis B virus carriers in the Han Chinese population. HEPATOLOGY 2011; 53: 422-428., , , , , , et al.
- 2A genome-wide association study identifies variants in the HLA-DP locus associated with chronic hepatitis B in Asians. Nat Genet 2009; 41: 591-595., , , , , , et al.
- 3Accounting for multiple comparisons in a genome-wide association study (GWAS). BMC Genomics 2010; 11: 724., , , , , , et al.
Xiuchan Guo M.D.*, Randy Johnson Ph.D., Stephen J. O'Brien Ph.D., * HIV Incidence and Case Surveillance Branch National Center for HIV/AIDS, Viral Hepatitis, STD, and TB Prevention, Atlanta, GA, Laboratory of Genomic Diversity, National Cancer Institute–Frederick, Frederick, MD.