In all honesty, I pulled those numbers out of a hat
Today my blog received a lot of traffic about this post from yesterday about the relative risk of disease genes. I wrote that the relative risk (RR) of the genetic variants we have discovered recently using genome wide association studies are rather small -- 1.1 to 1.5 -- and that such a small increase did not matter that much, all in all, and that I doubted that it would have much of an impact for us to know we have a gene that increases our risk that little.
All of this I stand by. The numbers for the relative risk are also consistent with the papers I've read, but I have not done a proper survey to see the actual distribution of the RRs. I am thinking about doing that now, but it is not quite as simple as it sounds to figure it out. There is something called "the winners curse" that essentially means that our estimates of the relative risk tends to be higher than the risk really is, because we estimate the risk from a biased sample: the sample where we discovered the risk in the first place. See Zöllner and Pritchard: Overcoming the winners curse for more on this.
I gave an example, however, where I said that increasing the risk of cancer from 0.1% to 0.15% -- a relative risk of 1.5 -- would have no consequence what so ever. Those numbers I just made up. I intentionally picked very small numbers to make a point, but it is a bit dishonest. I don't know what realistic numbers would be, to be absolutely honest, but these are probably way too small for any "interesting" disease.
If the risk of a disease, without "risk genes", is 0.1% I don't think we would bother with it in the first place. It would be pretty hard to find enough cases for a study anyway.
Realistic numbers might be 5% to 7.5% or 10% to 15%. I don't think it changes my point: people are not going to change their habits for such an increase in risk when the do not change their habit for much larger risks such as diet, exercise, smoking, drinking, etc. As Genome Technology Online puts it: That's Because Risk Is Small and Inertia Is Great.
Anyway, I shouldn't have made up numbers like that -- even as an informal example to make a point -- and I wouldn't have if I knew this many people would read it...
Now I should probably go figure out some accurate numbers so I don't make the same mistake again.
February 28th, 2008 at 7:48 pm
[...] I pulled those numbers out of a hat February 28, 2008 9:25 am ring worm medication diabetes Thomas Mailund had an interesting blog post (In all honesty, I pulled those numbers out of a hat).Here’s a [...]
February 29th, 2008 at 12:28 am
I think you (and other's out there posting on this subject) are missing part of the point. The risks that are measured at the moment are, in reality, marginal risks - that is, they are attempting to assign a risk value to an individual genotype. In reality, most diseases we look at currently (diabetes, cancer, cardiovascular disease, etc.) are not the result of single variants (or single loci), but the net effect of many (10's? 100's? possibly 1000's?) variants at multiple loci acting together. Thus, the relative risk is, in fact, a poor indicator of the actual risk attributable to genetic variation, because this measures the risk across the whole population of individuals carrying a single genotype, which understandably will be small (as only a portion of those will carry variants at other loci that together can account for the net change in risk). Eventually, with a better knowledge of the full spectrum of genetic variation that can influence a trait, we will be able to construct models which can predict, from a combination of variants at multiple loci, the individualized risk to a patient, and that will likely end up being much larger than even the 10-15% increase you've quoted above.
February 29th, 2008 at 4:55 am
Michael, I totally agree with what you write.
I am not saying that for polygenetic diseases the genetic effect is small, only that the marginal effect of any particular variant is small.
We know that there are major genetic contributors to these diseases. The relative risk of having cases in your family tells you as much. It is just that the variants we measure in personal genomics cannot explain much of this genetic effect.
Also, we only know of a few variants at this time. I don't think we will find more than a few with our current approaches that are, after all, based on the "common disease, common variant" hypothesis -- if our technology is only able to find common variants, I don't expect to find a lot.
Anyway, even with just a few variants, the total effect, if additive, can certainly contribute more than a small relative risk -- if you happen to have several of the disease variants.
Are they likely to have additive effects? Right now, we simply do not know.
Another possibility is that the genetic effect is explained by gene-gene or gene-environment interaction, as we still haven't figured out how to map for this. Yes, there are many suggestions -- I am currently finishing a manuscript myself ;-) -- but none of them have, to my knowledge, found any effects that have been replicated.
So in short, I am aware of your objections and I agree with them. But to measure and predict the effects you mention, we need more knowledge than we currently have.
At this time, knowing your genetic make up is not telling you much about your disease risk. I am convinced this will change in the future, but I am not optimistic that it will happen before we start doing re-sequencing studies.
I am pretty optimistic about when that happens, though. I will bet good beer that we see the first before 2010.