different odds ratio with regression coding

[manthan]

hello, I wrote the following code to do logistic regression and obtain odds ratio for statin, diabetes meds,  blood pressure meds, antibiotics and cholesterol (output cholesterol where 1 = high 0 = low) and independent variables statin, bP meds, diabetes meds, antibiotics (1= received 0 = not received)

proc logistic data=work.import descending;
class statin (ref="0") diabetesMed (ref="0") bPmeds (ref="0") antibiotics (ref="0");
model cholesterol = statin diabetesmed bpmeds antibiotics;
run;

when I run the above logistic regression the odds ratio for BP meds is with the following 95% CI being (<0.001 and >999.999)..why is this happening?

<0.001

<0.001

>999.999

as a follow up I ran the same code above but limited the IV to just two : 

proc logistic data=work.import descending;
class statin (ref="0")   bPmeds (ref="0");
model cholesterol = statin bpmeds ;
run;

and the Odds ratio is : 

1.136

0.489

2.640

why is it different in both cases? Is it because the Top one is adjusted odds ratio with more variables? If that is the case why am I GETTIng <0.001 and >999.999. I also checked for missing values and its only 1 missing value in the frequencies analysis. 

thanks

[ballardw]

Very wide confidence limits tend to indicate that the model chosen doesn't "predict" the dependent values very well.

Do you have any evidence from literature that Antibiotics should affect cholesterol?

And what happens to that smaller confidence interval when you remove the Bpmeds? Since statins are designed to control cholesterol, I have a sneaking suspicion you model(s) are introducing lots of random noise and in effect hiding the statin behavior.

Of course there are so many different antibiotic, diabetes and blood pressure medications I am not at all sure I would be modeling in frames of, I am guessing here, "subject is on/not one one or more of medication type".

Statins are at least sort of in a single group of related meds if I understand them at all.

[PaigeMiller]

Multi-collinearity can cause this. What is the correlation between your predictor variables?