Tools for Those Who Summarize the Evidence Base
Good Afternoon All,
I am writing a protocol for a meta-analysis and was asked by the funder to include our plan for handling studies that provide data on a condinuous outcome (change from baseline) but no measure of variance. I have seen papers like Furukawa et al (J Clin Epidemiol 2006;59:7-10) suggesting you could borrow variance measures from other studies. I was interested if anyone had experience with this and could offer any advice.
William L. Baker, Pharm.D., FCCP, BCPS (AQ Cardiology)
Assistant Professor of Pharmacy Practice
University of Connecticut School of Pharmacy
69 N Eagleville Rd, Unit 3092
Storrs, CT 06269-3092
I have seen meta-analyses normalizing extreme variance estimates (under the presumption that that they were typos in the original published document), but not usually doing anything systematic to check the assumption.
Of course your studies would have to use the same measure of the key outcome (or you would have to put them on the same metric), but in theory you could infer SDs from those observed in the other studies. There are estimations in Lipsey and Wilson (2001), but I have not seen anyone use them in print. I did fool around using them once, but the experience did not leave me feeling confident in the estimates.
You could do sensitivity analyses to see how dramatically larger or smaller variance estimates affect results. You'd have to defend whatever you decide to do.
Hope that helps! -Blair