.

.

Thursday, June 13, 2013

p = 10 to the -13

I've been knocking myself out at the ESHG for the past few days.
At one of the booths they were handing out free issues of Nature Genetics, so I picked one up. I get Science, so Nature doesn't cross my path often. Just to browse, I mean – any specific paper I can look up on line.
The April issue is dedicated to big papers on big genome-wide studies looking for loci contributing to cancer risk. Worth a read.
Large-scale genotyping identifies 41 new loci associated with breast canceer risk!
3 new loci for ovarian cancer risk!
35 new loci for prostate cancer risk!
!!!
wow: just genes all over the place.
And the p-values. P-values of your dreams: 10-6 and better. (A p-value is one measure of how significant, or reliable, a statistical difference is. Usually statistical significance is any p below 0.5. p = 0.5 means there's a one in 200 chance that you're dealing with some random fluke, so 10-6 is one in a million.
But go beyond p, and look at a different measure of importance: the odds ratios. An odds ratio is what are the odds of getting the disease, given that you have the marker. You take an odds ratio and multiply it by the basic risk in the population to estimate a more personalized risk. An OR of 1 is neutral. An OR of 0.5 means the risk is only half; an OR of 2 means twice, and so on.
Odds ratios for the big genes in breast cancer are sometimes too big to multiply, because the risk would be a sure thing, and it's actually not. The moderate risk genes have ORs around 3.
So why did I tell you all that? What odds of cancer did the new genes give?
From 1.05 to 1.26 for the risk alleles, with most not more than 1.1
0.91 to 0.96 for the handful of protective alleles.
A 1.1-fold risk. Gee, I should go out and get that marker tested today.
I mean, really, what is there to get excited about? Why should we care? 1.1 here and 1.1 there; that's nothing.
Nature Frickin Genetics.
What good is it to know this?
Perhaps someday, if you had all the different genes identified, in the right populations because they're not all the same, and if you knew all the interactions between them so you could tell that this one is important only if that one is also present, or is some other marker not otherwise important at all is also present, and if you knew all the interactions with lifestyle and environment (because there too, a gene may only 'count' in women who have had children, or who have a bmi over some threshold, or...). Perhaps that day you could tell somebody she's really at risk of this cancer, much more than other people.
And even then I'm not convinced of the value of this sort of research. We've learned that cancer risk can be due to a mutation in one of a handful of genes, and now that dozens of little tiny effects can add up to something bigger. But for this latter part, is it useful to know that my risk is one in eight, or one in ten?
Let's move on.


2 comments:

The Bug said...

I can either live in fear, or I can accept that whatever is going to happen will happen & just try to do as many healthy things as I can. Is it fatalism to do the latter, or just common sense?

My mom died of kidney cancer - & no one has a clue why. Environmental (doubtful)? Genetic (probably not)? Too many diet Cokes (maybe)? Just dumb bad luck?

Dan Eastwood said...

It gets worse. p-values from large scale testing also produce a large number of falsely significant results (type I error) simply by chance. Methods such as False-Discovery-Rate can be helpful, but new statistical methods take a long time to come into general use.