Wednesday, March 17, 2010

SNPs for breast cancer risk? It Depends.

I hold in my hot little hands a copy of the NEJM, March 18th edition. In it there is an article which isn't even released yet.

"Performance of Common Genetic Variants in Breast-Cancer Risk Models"

Remember when we did this for heart disease risk? FAIL WHALE.....

Do you think it will happen again?

The Study

10 common genetic variants

I had to create a couple of pages on SNPedia for this list FYI.....

The Methods:
Cases and controls-WHI, ACS CPSII Nutrition Cohort, Nurses Health Study, Prostate/Lung/Colorectal/Ovarian Cancer Screening Trial, and Polish Breast Cancer Study.

Cases-Woman who had received diagnosis of invasive breast cancer.

Risk Models Used-A hybrid of the Gail model.....I.E. Not exactly the Gail Model.
1. First degree relatives with breast cancer
2. Age at Menarche
3. Age at first live birth
4. Number of Breast Biopsies

They acknowledge that they were unable to get atypical hyperplasia and Mammographic density. Both of which have improved Gail.

So, This Gail is a little hobbled and not the best predictive model.......

The studied models- 5 logistic regression models
I don't have the supplementary tables and methods yet.

The nongenetic model-Gail Model
The Demographic/Genetic Variant Count Model-included number of alleles.
The Demographic/Genetic Individual Variant-Accounted for individual effects of each SNP
The Inclusive Model-Gail, Genetics Demographics
The Demographic Model
And Random....

When we do these sorts of statistical analyses we look for a couple of things.

A. Number of people reclassified and how?
B. The Area Under the ROC Curve


1. The Inclusive Model Yielded and AUC of 61.8%
2. The Nongenetic Model yielded an AUC of 58%
3. The Genetic Individual Variant Yielded an AUC of 59.7%
4. The Genetic Variant Count Yielded an AUC of 58.8%
5. Breast Biopsy BY ITSELF Yielded an AUC of 56.2%

That is a 3.8% difference in Yield from Genes and without Genes integrated into the weaker Gail Model.

Lastly, they asked. Well, does this Inclusive Model do a good job of discrimination of High risk vs. low risk.

The Answer- It determines lower risk better than Gail. It does not determine higher risk better.

The authors of this study have stated that

"As in Diabetes and cardiovascular disease, the addition of the common SNPs added little to the predictive value of the clinical models. On the basis of theoretical models, Gail has shown that increases in the AUC similar to those observed here and not sufficiently large to improve meaningfully the identification of women who might benefit from tamoxifen prophylaxis or screening mammography"

Take Home

The addition of these factors only creates a minimal statistical increase that is of no useful clinical benefit.

The Sherpa Says: If the press says "gene tests fail to improve risk assessment" You can be assured that the DTCG industry is no longer the darlings. If instead they say "Improvement in risk model" well, then you have chance to woo them back! It Depends.......


Danny said...

Question: Does it matter if it does not improve yet? There is still some risk conferred and these are only with studies of common variants. In the next two years we will see many studies looking at whole genome sequencing and rare variants which will hopefully confer greater risk.

That's for the future, though. For now, however, why can't these SNPs be used? Why the genetic exceptionalism? They are merely just another risk tool to use concurrently with Gail. The more, the better right? I understand your fight against DTC genomics companies whose marketing overstates the value of these SNPs but your talk also does a disservice to the advancement of the field it seems by treating SNP scans as anything but just another, ADDITIONAL risk tool.


Steve Murphy MD said...

My answer was "it depends."

I don't think this is such a bad tool. I think it may help when we don't have all the data for the Gail risk.

I would like to see it go through the diagnostics process and be implemented. Just to test it and see.

The research for that needs to be done.

Listen, creating a diagnostic is hard work and requires significant investment. Of both Time and Treasure. These companies are not looking to do that.

Someone should. I would use this test once clinically validated. Provided I cannot get any Gail model info from the patient.


The whole field is great. And my shouting about these DTCG companies doing unscrupulous marketing (Not all are doing that, see Pathway and Counsyl)
is drowning out the good discovery. I am sorry, but your PhD brothers/sister looking to make a fast buck with PR wonks screwed it up for the real scientists, not me, them. Blame them.