To tell the whole truth, we should say that the major part of doctors can't
really read a meta-analysis, and that they will never do it. I am the
director of an EBM course for hospital's doctors, and when they have just
participated to a lesson about reading meta-analysis, they always tell me
that they will never be really able to critically read a meta-analysis. It's
too complicated, they say, it's to difficult to evaluate the methodology,
it's to difficult to know if you can really transfer it in your practice.
Probably we should bear that in mind and try to produce more suitable stuff
for practical doctors. Doctors are not statisticians and they will never be.
But medical literature seems to forget it or to ignore it every day.
Danilo di Diodoro
Bologna, Italy
-----Messaggio originale-----
Da: Evidence based health (EBH)
[mailto:[log in to unmask]] Per conto di Piersante Sestini
Inviato: marted́ 17 febbraio 2009 18.06
A: [log in to unmask]
Oggetto: Re: making meta-analyses more consumer friendly
Emily DeVoto wrote:
> risks and benefits because they are more simple and meaningful than
> relative measures. (E.g., 25 out of 100 women in the exemestane group
> had a recurrence of breast cancer, compared to 35 out of 100 women in
> the tamoxifen group.)
this is wise... You are probably familiar with this paper:
Hoffrage U, Lindsey S, Hertwig R, Gigerenzer G. Medicine. Communicating
statistical information. Science. 2000 Dec 22;290(5500):2261-2.
>
> I don't suppose there's any way to back-calculate estimates of absolute
> risk and benefits from summaries of studies, is there? I don't see
Yes, if the paper reports all the figures along with the forest plot (as
the Cochrane reviews surely do). Just enter those figures in your
favorite meta-analysis program and ask for a plot of "risk difference"
instead of the osual OR or RR
If there is a proper way of obtaining correct pooled estimates of the
separate risk, I also would be very curious to know.
But please be warned that often studies are really "forced in" meta
analyses even if they are very heterogeneous, so to obtain a reasonable
estimate of baseline risk you may need to re-select the studies (and
possibly devise a new specific search strategy).
regards
Piersante Sestini
|