I did a matched case control study in Uganda looking at Determinants of Severe malaria in Children (325 age-matched pairs; severe malaria cases vs. uncomplicated malaria controls). I am doing a sub-group analysis of risk factors by type of severe malaria syndrome; cerebral malaria (59 age-matched pairs) and severe malaria anemia (141 age-matched pairs).
Using the clogit command I run a regression model for the cerebral malaria sub-group which resulted in some explanatory variables (EV) having OR with extremely wide CI. Based on the VIF scores, and the Conditional Index mullticollineraity doesn't seem to be a problem in the model. I therefore think the wide CI’s are attributable to small sample size numbers within the groups
I've used to approaches to try and over overcome this challenge:
1. Drop the EV; an undesired option for I lose the influence of the variable in the model.
2. firthlogit command; and the problem appears to be addressed
However, I am not sure if the firthlogit commands is applicable to matched data, and if not what are my options ?
Thanks
Arthur Mpimbaza
Email: [email protected]
Comment