Are there conventions for marking up Odds Ratio?
I can't find from the study whether this is considered to fit recessive or dominant model, or what the effect of number of alleles is on the Odds Ratio. In addition, according to 1000 Genomes the risk-free genotype A/A is 0.027 in CEU and even in ASW only 0.102. When almost everyone carries at least one copy of the risk allele, shouldn't A/A actually be considered reduced risk rather than showing almost everyone increased risk?
In this particular case the study also states "The SNP rs12010175 in FAM58A is in modest LD with rs5945326 near DUSP9 (r2 = 0.35). Conditional analyses by including the two SNPs in one logistic regression model revealed that these two SNPs represent the same locus (P adjusted for rs5945326 = 0.02)." which seems ambiguous on whether the odds ratio of the two should be considered additive or overlapping. I already adjusted the GWAS box to show the new p-value, which makes it unlikely in any case.
I found Catalog of Genome-Wide Association Studies: Full Description of Methods which might be good guidelines, but opens some more questions. For example, "Focusing on risk alleles, we invert ORs < 1 and their associated confidence intervals, and report the opposite allele if available." so if 99% have the risk allele this is indeed reported as "increased risk". Also, "Where multiple genetic models are available, effect sizes (OR's or beta-coefficients) are prioritized as follows: 1) genotypic model, per-allele estimate; 2) genotypic model, heterozygote estimate, 3) allelic model, allelic estimate." however the genetic model doesn't seem to be indicated even on the original database. Some existing GWAS boxes have also showed homozygote estimate, which I assume to be in error.
Another quality issue is there is a bias against selecting studies where no association was found, and as I understand if no-association is mentioned Promethease will still pick them up, which may be undesirable. So pages end up listing two or three studies where statistically-insignificant association was found, and not the 10 studies where no association was found (on top of the publication bias where most studies report on an association).