# Conditional logistic regression model does not converge but logistic regression model does

I am running an analysis where I have 2500 cases and 2500 controls. The cases have disease A, and the controls do not. I am trying to see if having disease A increases the odds of various diseases. For the sake of simplicity, we can focus on one disease, call it disease B.

D = 1 if disease B present, 0 otherwise

E = 1 if disease A present, 0 otherwise

I am also including in the model a measure of healthcare utilization.

F is a positive integer proportional to an individual's utilization of healthcare.

I am running the logistic regression model as such in R:

glm(D ~ E + F, family = "binomial")


Now, this works fine.

However, when I try to run conditional logistic regression, it gives me an error:

library(survival)
clogit(D ~ E + F, strata(matched.pairs))
Error in fitter(X, Y, strats, offset, init, control, weights = weights,  :
NA/NaN/Inf in foreign function call (arg 5)
In addition: Warning message:
In fitter(X, Y, strats, offset, init, control, weights = weights,  :
Ran out of iterations and did not converge


I have tried different strata, including dividing the individuals into quantile bins based on F. It does not seem to change anything. (note: pairs are matched on age, gender, race, and F)

This occurs only when I run it on a larger sample size. I ran this same analysis on a sample size of 200 (100 cases and 100 controls) and it worked fine. When I use a sample size of 5000, I get the above error.

I also made sure that at least 10 cases and 10 controls had the disease in question (disease B, for this example).

I am not sure why logistic regression runs fine when conditional logistic regression does not. Can anyone offer me any advice?

• This is strange... in case control studies don't we usually estimate odds of exposure as a function of case/control, not the other way around? After all proportion of cases vs. controls is a researcher choice. – Alexis Jul 24 '14 at 15:26
• Maybe the way I described it was unclear. The model is a function of case/control. It's the odds of having a disease if you are a case versus the odds of having a disease if you are a control. – user3821273 Jul 24 '14 at 15:41
• I see it is clear: "exposure" is another different disease condition. – Alexis Jul 24 '14 at 16:48
• Try using method="approximate". – purple51 Aug 22 '14 at 2:00