Date:         Wed, 17 Oct 2007 09:58:06 -0500
Reply-To:     Mary <mlhoward@avalon.net>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Mary <mlhoward@AVALON.NET>
Subject:      treating ordinal (genetic) data as continuous in Proc Logisitic
Content-Type: text/plain; charset="iso-8859-1"

Hopefully Peter Flom, Kevin Viel, or Matthew Zack could have some comments, and I'd appreciate any input from statisticians as well.

I'm building models to predict disease, in that 1=has the disease, and 0= does not have the disease.  

Many of my variables are what are called SNPs- these are specific alleles within genes that I want to model against whether people have the disease or not.   I will later probably expand upon the model to include other variables that we have that are not genetic- which are primarily patient history variables like whether they smoke or not.

But my question now is how to deal with the SNP's.

A SNP is ordinal in nature- there are 3 possible values, along with missing.  These are:

a.   Homozygous for Allele 1  (example, AA)
b.   Heterozygous for Allele 1/Allele 2 (example (AB or BA)
c.   Homozygous for Allele 2 (example BB)

Allele 1 is designated to be the most frequent allele- thus most people in the population would expect to fall into category a.,
some would fall into category b, and some would fall into category c.

My question is, in Logistic regression, is it reasonable to treat these as continous variables without a class statement?   What I'd like to do is code a "2" for category a. (AA), a "1" for category b (AB or BA), and a "0" for category c (BB), and then use these as numeric variables in the model, like this:

proc logistic data=logistic_set;

model disease_flag= snp1 snp2;

run;



where I wouldn't have a class statement in my Proc Logistic.  



If any statisticians/genetics types could discuss the pitfalls/advantages to this approach over using a class statement (probably with a reference category to category a. (AA)), I'd appreciate hearing the discussion. Also how to interpret the odds ratios if I do this would be very helpful. 

Thanks!



 

Mary Howard

Research Assistant III

Dept. of Ophthalmology

Univ. of Iowa Hospitals and Clinics