I am doing a meta-analysis in which I have extracted data on ORs as the effect measure for presence of a condition against absence (reference category) of the condition. I one particular study, 4 levels of the variable of interest are presented instead of 2 as in the rest of the studies. ORs were estimated for 3 levels against one level as the reference category. By definition, adding the first and last two categories will give the two levels I need as in the other studies. However, the paper did not report frequencies that resulted in the OR estimates so the only thing left is the OR and 95% CI. How can I combine the ORs of the first and last two categories to produce the presence and absence comparison I need?
Hello. That’s not possible by any direct means, you got to have at least counts for each condition in order to remake the contingency table. Check for supplements of the article or maybe a published dataset of it, or else it won’t fit in your methods.
Since OR approximates RR,one approach could be to take the mean OR of the two highest categories (ORm1 say)and the mean OR of the two lowest categories(ORm2) and then calculate a new OR for a two level comparison ORm1/ORm2. From the 95%CI's given you could calculate the OR variances for the two highest categories and then the variance of their mean .From these you could estimate 95%CI s for the new two category comparison and use these estimates in your meta analysis..
- Badges
- Science topic
- Similar topics
- Statistics
- Survey