Odd Ratio calculation in plink

Question

[dandanli0365]

Hi, I feel confused when i calculate odd ratio of per snp by plink. My problem is described as followings: when i run this command plink --noweb --bfile mydata --assoc ,I get this results:

CHR         SNP         BP   A1      F_A      F_U   A2        CHISQ            P           OR 
1   rs3094315     792429    G   0.1489  0.08537    A        1.684       0.1944        1.875 
1   rs4040617     819185    G   0.1354  0.08537    A        1.111       0.2919        1.678 
1   rs4075116    1043552    C  0.04167  0.07317    T       0.8278       0.3629       0.5507 
1   rs9442385    1137258    T   0.3723   0.4268    G       0.5428       0.4613       0.7966 
1  rs11260562    1205233    A  0.02174  0.03659    G       0.3424       0.5585       0.5852

when I use another command plink --noweb --bfile mydata --logistic, I get this results:

CHR         SNP         BP   A1       TEST    NMISS         OR         STAT            P 
   1   rs3094315     792429    G        ADD       88      2.061        1.381       0.1672
   1   rs4040617     819185    G        ADD       89      1.804         1.12       0.2629
   1   rs4075116    1043552    C        ADD       89     0.5303      -0.9272       0.3538
   1   rs9442385    1137258    T        ADD       88     0.8197       -0.687       0.4921
   1  rs11260562    1205233    A        ADD       87     0.5758      -0.5877       0.5567

So, why is odd ratio of the same snp different?

Answer 1

[discombobulated]

Basically they use different methods for the calculation of the odds ratio - logistic uses the logistic regression model and assoc uses allele counts. Your differences aren't that large but the rarer the allele the greater the difference you may see. Have a read of Plink 1.07 manual - it is outlined in this page.

[dandanli0365]

thank you very much! now, i still feel a little confused. I know how to calculate OR in '-assoc'. I wonder how to calculate in logistic regression. Here is my guess: coefficient is firstly calculated in logistic regression and then the exponent of the coefficient is used for OR. Is it right?

Answer 2

[discombobulated]

In the interest of learning you can test your theory by implementing the --beta flag into your analysis. This will return the regression coefficients. plink --bfile mydata --logistic --beta

[dandanli0365]

I have tried it and knew it well, thank you very much again!