Binary Logistic regression with multiple variables, identify effect of each variable?

[jongwook0712]

i am doing a binary logistic regression using multiple categorical variables to get a probability output:
The dependant variable being case or control and the independent variables being genotypes for 3 different SNPs. If i am obtaining a probability output,

Is it possible to determine the average contribution of each variable to the probability output?
The hope is to take regression results from a large sample set and design a prediction model based on the cohort results.
Thanks

[GretaGarbo]

This looks kind of familiar! Have you not already asked the same question in the “Statistics” forum?

So, it is a double post. That makes things confusing! Read the “sticky” in the “Statistics” forum. I can understand that you are eager to get a response. But one day is to short time to wait.

For me, one reason for not responding is that I have not understood the question. If you send me a question in Latin I will not respond. Not even if you repeat the question. Not even if you RAISE YOUR VOICE. I will still not understand.

If you add some new information it might be easier to understand what you are really after. If you give 10 rows of example of values for what is meant by SNP:s for example.

Then someone else might respond.

[Dason]

I deleted the other thread as you're talking about SNPs and genotypes and that implies biostats. GretaGarbo is correct - please only post a thread once.

[Link]

I'm not really understanding what you're saying either. Could you clarify?

[jongwook0712]

Sorry for the double post!
So Im looking at three genetic variants that are highly associated with drug induced toxicity. Using a forward logistic regression model using these 3 independant variants, I stratified the patients by risk (low,inter,high) based on the predictive values from the model.
For example,

Case/Control Gene1 Gene2 Gene3 G1 G2 G3 Predicted Probability
0 T_T A_A A_G 0 0 1 0.18944
1 T_T A_G G_G 0 1 0 0.28223
0 T_T A_G G_G 0 1 0 0.28223
0 T_T A_A A_A 0 0 2 0.28244
0 T_T A_G A_G 0 1 1 0.45582
1 A_T A_G A_A 1 1 2 0.91381
0 A_T A_G A_A 1 1 2 0.91381
1 A_T G_G G_G 1 2 0 0.79941
1 A_T G_G G_G 1 2 0 0.79941
1 A_T G_G A_G 1 2 1 0.89462
1 A_T G_G A_G 1 2 1 0.89462



The actual size of the cohort is around n=350. This predictive model proved to have a pretty high sensitivity and specificity when combining with signficant clinical variables.
HELP PART - The next step is to develop a prospective prediction algorithm for replication work. I guess the approach i thought would be feasible was to assign an average score for each genotype based on the probabilities of the logistic regression model. Any suggestions on how to do so???
Thanks!

[GretaGarbo]

Frankly, since you have been double posting I am reluctant to comment. I don’t want to encourage that. But now since you asked me..

Using a forward logistic regression model

If you have been using stepwise forward regression I would say that I am sceptical. Someone else here (who also had double posted so I did not say anything) planned to do stepwise regression on American football results. But I think that such a procedure can produce a lot of rubbish. (That does not mean that your stuff is incorrect but…) Someone here had a “sign” in the end “Think first, regress later”. I think he said that for a reason.

when combining with signficant clinical variables

I don’t know what the above means.

the approach i thought would be feasible was to assign an average score for each genotype based on the probabilities of the logistic regression

I am not sure of the consequences of such a procedure.