I only read the abstract, and skimmed the tables, but their was no large Odds Ratio reported, most around 1-2. That small of an OR is not large enough to stratify patients or use as a diagnostic, you usually need an OR of at least 3, much better if its like 10 (Pepe 2004, Amer. Journ. Epid).

As the authors report -- what the paper will help with is identification of possible therapeutic targets for drug development. But the study in itself will not effect clinical practice.

There is a good editorial in the June 26th NEJM by Hunter, on the utility of genome wide association studies (or lack thereof) for clinical medicine.

you usually need an OR of at least 3, much better if its like 10

that's based on genotyping a single locus. that approach is not going to work for complex disease--any sort of classification is going to need to be based on genotyping many loci in the genome. if you were to genotype all 30 susceptibility loci for crohn's disease, you'd be able to get a decent guess as to where people fall in terms of liability to develop the disease.

...larger than that currently mapped for most (any?) other complex trait.

Well, pigmentation traits (for example) are generally much better-mapped than this. In terms of complex diseases: around 50% of the sibling recurrence risk of age-related macular degeneration is explained by five SNPs; just under 50% of the risk of male pattern baldness is due to variation in the androgen receptor gene; and for age at onset of dementia a single locus (APOE) explains around 10% of the population variance. But yes, for most common diseases and quantitative traits explaining 10% of the variance is still a way off (even for type 2 diabetes, despite ludicrous sample sizes).

A timely example: from this paper that looked at the top 18 known risk variants for type 2 diabetes. The top 1% most genetically at-risk individuals have around 4 times the risk of diabetes as the bottom 2%. It's still not great predictive accuracy (it adds only a marginal improvement to a model that simply incorporates age, BMI and gender), but it's a start.

dgmacarthur: I just read the abstract of the paper you mention, but they do not mention a measure of Fasting Blood Glucose or Hemoglobin A1C. Both are extremely easy to measure clinical parameters used for the diagnosis of diabetes. So, their results are probably even less applicable, than you already have stated.

p-ter: I'll believe it when I see it. My intuition is, that if you tried to build a classifier from their 30 loci (most, extremely weak), you would wind up with something doing marginally better than tossing a dart -- and clinically useless. Most patients with Crohn's will have low probability/risk scores (using those 30 alleles -- since 90% of the risk is still unexplained, and those patients will be misclassified). Crohn's is also a rare disease, making things worse.