Dear Allstat,
I have been looking at a couple of rare cancers, where the only
published data is (and will only ever be) case-series. There a few
case-series from different centres at different times.
I am looking for a way to combine them all. I am familiar with the
principles suggested by the MOOSE (Meta-analysis of observational
studies) that suggest sensible quality controls to follow. However, in
terms of combining numerical parameters, I am less certain.
One suggestion has been to use a Random effects model, and check for
over-fitting [1]. More generally, people seem to concentrate on a
sensitivity analysis [2] to look for overly influential studies.
However, I am looking for something with a little more statistical
guidance - for example, how would I go about combining the estimates of
median survival?
I am assuming here that I will /not/ have access to all the original
patient data.
Thanks,
Matt
1: http://gateway.nlm.nih.gov/MeetingAbstracts/ma?f=102237190.html
2: http://gateway.nlm.nih.gov/MeetingAbstracts/ma?f=102274388.html
|