I would like to identify the predictors of seizure recurrence after a first seizure. Should I use a Cox regression model yielding hazard ratios or should I use a log-binomial regression yielding risk ratios?
From what I understand, I shouldn't use logistic regression because it yields odds ratios, which often overestimate risk ratios.
There are a series of previous studies about "predictors of seizure recurrence after a first seizure" that may be helpful to you, for example:
Article Predictors of early recurrent seizure after first seizure pr...
Article Risk of seizure recurrence after a first unprovoked seizure ...
Article Risk of Seizure Recurrence After First Seizure
This comes down to whether the time-to-event component is clinically relevant or not. If it is, then you should do Cox regression, and otherwise logistic or as you suggest log-binomial.
For rare events, odds and risk ratios will be almost identical. For more common events, odds ratios will be higher but it isn't a case that they are 'overestimating' the likelihood, they are just a different effect measure. There is nothing wrong with using logistic regression in this instance either.
As Jack Henry said it depends on the research questions. I have attached 2 studies that may be helpful in answering the question. they contain one of each type.. Part of the answer is are risks important or is it prediction of a y. Best wishes, David Booth
Thank all of you for your help. I'll have a look at the studies you've sent me.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis