Log in
My story with 23andmeMy story with diagnosed diseasesLatest DiscussionOpen CurationCuration Log

23andme vs. GeneKnot: My Atrial Fibrillation

I got 23andme results in early 2012. The only disease with high risk was Atrial fibrillation. 

Its odds ratio was 1.73. It has remained same since then. So my life time probability to get this disease is 27.2% (average) * 1.73 = 46.9%. When I checked this result, I had mixed feelings. 'Oh,, it's good. I don't have a major issue'. Then I thought 46.9% is pretty high and did some research about Atrial fibrillation. Coffee and alcohol may make it worse, and cholesterol and blood pressure should be monitored. OK.. I'd be more careful. Less caffeine and alcohol!

Then I started wondering whether this is only disease that I can find out from this DTC test. Is this 1.73 correct? I have other typical or low risk diseases. Are they really true?

23andme uses two SNPs for European. 

1. rs10033464 GG (my genotype) European: 0.92 

2. rs2200733 TT (my genotype) European: 2.57

The first question would be how 1.73 was calculated from 0.92 and 2.57. The customer support didn't give a clear answer and said the 23andme Odds calculator was used. 

The second question is how well these two SNPs represent Atrial fibrillation risk. 

GeneKnot uses GWASCATALOG data and reports 9 SNPs belows. The only overlapped SNPs is rs2200733 of which odds ratio (1.72) is different from 23andme's result (2.57)

GeneKnot reports: 

Atrial fibrillation
rs7193343-TT, rs2106261-TT, rs6817105-CC, rs10824026-AG, rs6843082-GG, rs17042171-AA, rs7164883-AG, rs2200733-TT, rs3807989-GG
1.21, 1.24, 1.64, 1.15, 2.03, 1.65, 1.19, 1.72, 1.11


23andme refers to two papers (Update on 8/7/14: Asian population often shows two Ts in  rs2200733 (~60% of frequency). Kääb et al. (2009) is based on European, so Asian users should not rely on this result):

Kääb et al. (2009) . “Large scale replication and meta-analysis of variants on chromosome 4q25 associated with atrial fibrillation.” Eur. Heart J. 30(7):813-9.
Gudbjartsson et al. (2007) . “Variants conferring risk of atrial fibrillation on chromosome 4q25.” Nature 448(7151):353-7.
GWASCATALOG refers to the same paper for odds ratio 1.72:
Gudbjartsson et al. (2007) . “Variants conferring risk of atrial fibrillation on chromosome 4q25.” Nature 448(7151):353-7.
23andme seems to use Kääb et al. (2009) to obtain 2.57. 
If you check the original paper, it says:
"Single nucleotide polymorphism (SNP) rs2200733 was associated with AF in all four cohorts, with odds ratios (ORs) ranging from 1.37 in Rotterdam [95% confidence interval (CI) 1.18–1.59; P = 3.1 × 10−5] to 2.52 in AFNet (95% CI 2.22–2.8; P = 1.8 × 10−49)."
"A meta-analysis of the current and prior AF studies revealed an OR of 1.90 (95% CI 1.60–2.26; P = 3.3 × 10−13) for rs2200733"
The values are not exactly consistent here, but 23andme might take the value around "2.52" since it has a really low p-value. But it's not still clear how 23andme ended up with 1.73. My genotype of rs22007333 indicates some risk in Atrial fibrillation.
Gwascatalog appears to support this conclusion:
Five SNPs between odds ratios 1.11 and 1.21 show p-values between 3E-17 and 4E-9, while 4 SNPs between 1.64 and 2.03 have p-values between 2E-74 and 3E-28. Therefere my "true" odds ratio is more likely to fall in between 1.64 and 2.03 rather than between 1.11 and 1.21. In this particular case, unfortunately to me, 23andme seems to do pretty good jobs even with one high risk SNP (rs22007333). However, if 9 SNPs are shown in the context of significance, it would be more convincing.
This odds ratio investigation is actually the first part of my story.
If you find my ID (geneknot) in a "GeneKnot" menu for Atrial fibrillation (https://geneknot.com/mds#images/Atrial_fibrillation_user_snp_scores.gexf; login required; Please find snapshot below; red:Diagnosed users; orange: users with high risk or related diseases), I'm really far away from other users (as of 8/18/2013) since I have relatively many "risk" alleles. I really hope that many other users can join and discuss their disease progression to find people with similar genetic risk and lifestyles. Ideally if enough users at different ages join this effort, the community might be able to get useful information to prevent diseases in next 10, 20 or even more years.
Thank you all.
Member of GeneKnot