03-31-2016 02:11 PM
I have a data set of 274 people with 137 cases and 137 controls matched on w, x, y and z. I am doing a conditional logistic for multiple different exposures and testing effect measure modification by sex. I was going to check confounding too (related to x, related to y among unexposed and not on the causal pathway) but I am coming up with very odd values in my output. It says that the convergence criterion is satisfied and the data seems to be formatting correctly and there are no missing values.
This is my code:
proc logistic data=data; /*outcome = suicide. exposure = sex*/
model sample_index (event='1') = b_dm02; run;
Odds Ratio Estimates Effect
95% Wald Confidence Limits
03-31-2016 02:22 PM
The odds ratio is the ratio of two numbers. It looks like the denominator is zero or very close to zero. The estimate for the odds ratio is very big ("infinity") which is displaying as > 999.999.
The Wald confidence interval about 1 is similalry undefnied. Essentially you can interpret it as (0, infinity), which gives you no information.
03-31-2016 02:45 PM - edited 03-31-2016 06:38 PM
It is indicative of a larger problem, no? I should be able to easy test to see if sex is a predictor of suicide. I need to include it in all of my conditional logisit models (testing for effect measure modification). The variable works with McNemar's test so it doesn't seem to be a variable-level issue.