To Ioanna Gioni:
It would be worth while considering the transformation of the concentration by log10:
This would be a different model, and the corresponding odds-ratio would refer to the
effect of a ten-fold increase in the concentration.
A look at the distribution of the concentartions could give a hint to as whether
a few very hig concentrations carry most of the information in the linear formulation.
Whether the log-transform is relevant is however not only a statistical but also a
question of clinical interpretability and relevance.
Steno Diabetes Center
Niels Steensens Vej 2
tel: +45 44 43 87 38
mob: +45 30 75 87 38
fax: +45 44 43 07 06
[log in to unmask]
From: A UK-based worldwide e-mail broadcast system mailing list [mailto:[log in to unmask]] On Behalf Of Edmund J. Bini, M.D., M.P.H.
Sent: Friday, September 17, 2004 3:17 AM
To: [log in to unmask]
Subject: Re: Strange Odds ratios
Given the large range of your assay, it would be better to calculate the OR for a larger change in the concentration. The OR is 1.004 which means that the odds of having CAD increase by 1.004 for each 1 unit change in your assay. Since the range of your assay is very wide, you would be better off calculating a new variable for your assay by dividing the original concentration by 10 or 100 or some other larger number. If you divide the concentration by 100, then the OR of CAD would be for each 100 unit change in concentration. Hope this helps. Ed
From: A UK-based worldwide e-mail broadcast system mailing list [mailto:[log in to unmask]] On Behalf Of ioanna gioni
Sent: Thursday, September 16, 2004 12:17 PM
To: [log in to unmask]
Subject: Strange Odds ratios
I am a little bit puzzled about something and I was wondering if someone had
any suggestion related to this...
I have the concentration results of an assay which clinicians would like to
use in order to discriminate between subjects of CAD (Coronary Arterial
Disease) and subjects of no CAD.
The CAD group has 109 subjects and the no-CAD 53.
I try to perform logistic regression modelling for the probability of a
subject having CAD. However, the results are a little bit confusing.
Although, the model fits the data well and the p-value of the estimate of
the assay was highly significant (<0.001), the odds ratio of it came out to
be 1.004(1.002, 1.006)!!!
Therefore, I got confused.
Then, I implemented a ROC plot which gave me really nice plot of AUC of
0.895 (the two groups were found to have significantly different means with
the CAD group having bigger mean concentration than the no-CAD).
Then I thought that maybe it was the different group sizes (53 non-CAD
versus 109 CAD). I generated some data so that the group's size would be
quite similar (106 non-Cad versus 109 CAD this time.) Again, the estimate of
the assay was found to be significant (P<0.001) but the odds ratio
I am thinking that these strange results could be due to the very high range
of concentrations I have to deal with, (47µg/ml-10307.40µg/ml). Anybody has any suggestion..? Ioanna
It's fast, it's easy and it's free. Get MSN Messenger today!