Similarly as in this question, I'm working with meta-analysis of studies reporting ORs of an event.

I have one study where I'm not sure if it should be excluded.

The frequencies of the outcome are: No outcome: 85% Outcome 1: 6% Outcome 2: 9%

Authors use 'multivariate multinomial logistic' regression using this outcome and report adjusted effects as RRR. This estimates refer to continuous variable coded 1 to 4 and range from 1.14 to 2.97 in different models and would be the estimated pooled to meta-analysis.

I know that in case of rare events (sometimes quoted as <10% incidence), RRs approximate ORs.. My question is – would that hold for multinomial model? And is 9% + 6% still rare? Effectively - can these RRRs be use as approximation of ORs?

  • $\begingroup$ Is there some reason, why one cannot calculate the odds ratio from the reports of the study (e.g. outcome vs. no outcome)? $\endgroup$
    – Björn
    Commented Dec 10, 2015 at 13:06

1 Answer 1


Pointers from Roberto Gutierrez on this answer on Statalistand the fact that the authors used multinomial model slowly make me believe that this study will be excluded.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.