I just received the May issue of Genetics in Medicine, only 24 days late. But it caught my attention for several reasons.
Monday, May 24, 2010
1. The issue is covering Adult Topics almost exclusively this month
Granted this article was a single author MBA, it was notable at the work she must have put in to this review.
1st she did an extensive analysis of the service 23andMe, Navigenics, deCodeMe, Gene Essence. She assembled the 20 multigenically evaluated conditions, reviewed website data, and deep dove into the studies, average pop lifetime risk, loci, genes, SNPs, Quant risk assessment, and methodologies.
That sounds very similar to what the FDA is requesting to do. In their case with non publically available data as well.
2nd she did a complete locus analysis which is available here.
213 conditions covered by DTCG companies, with only 9 conditions covered by all identified companies. 15 addition covered by 4/5 companies.
Lifetime average risk values of the same populations.
It turns out that the companies provide different life time risks for the same disease in the same populations.
This is not a big deal to me if you wiggle 2-4 points. But some vary widely
Glaucoma 1% in Navigenics while it is 15 for deCodeme
Heart attack 42% for Navigenics and 21% for 23andMe in Men
Heart attack in women 25% in Navigenics while it is 7% at 23andMe
DVT 3 percent for Navigenics 12% for 23andMe
23andMe does not provide references for their lifetime risk data.
Heterogeneous SNPs and Loci Assessed
No big surprise here, it turns out each company has their own way to make a Big Mac, each has their own special sauce and pickles/onions and even their own sesame seed bun. Thus you get different SNP risks given to customers.
A total of 224 loci are covered 401 SNPs for the 20 multigenic conditions. Of the 224 loci, 115 are only covered by one company. 63 are reviewed by only 2 companies.
For 12 conditions covered by all 4 companies, only 9 SNPs were covered by all. These 9 SNPs represent ONLY 3% of the total SNPs covered by all 4 companies and 18% of all loci covered.
Heterogeneous quantitative risk assessment
Once again, different risk assessment methods rule the day at these companies. Just like if I were to use Reynold Risk instead of Framingham risk but at least I have some data to base my conclusion. We have none of that with the DTC company risk models......
On Saturday I had a pissing contest with Thomas Goetz of Wired Magazine and his book called The Decision Tree,
He kept saying, It's My data. I kept saying. Fine, but the interpretation needs to be regulated.
I think we have a very decent reason why right here.
When you get a cholesterol of a blood pressure reading in the United States, you would hope the interpretation you receive is standardized in some way.
Further you hope at least the person giving you the interpretation of that data has some sort of licensing to assure quality and accuracy.
Unfortunately in this field there are many, many unknowns. This makes the risk prediction even less accurate. So it is no surprise these companies have widely variable assessments. But what does trouble me more, is the fact that they seem to not have done their homework with average lifetime risk populations.
That seems like they should be at least on the same page with this information. And why 23andMe has not listed reference articles for their quoted population risk is beyond me.
The Sherpa Says: Doctors go to medical school for 4 years, then go onto residency for 4-8 years and some do fellowship for another 2-5 years. And then we give risk assessment and diagnose and treat. Why do people forget that? Oh and we first operate under the principle of First Do No Harm. What doesn't Mr Goetz get about that?
Posted by Steve Murphy MD at 5:11 AM