Thread by @eturnermd1

At first glance, it seems like the bottom line is "there's no publication bias in this field. But how can you study publication bias if you look only at studies that are published. What about the UNpublished studies?
#PLOSONE
dx.plos.org/10.1371/journa…

To be fair, they report having looked at study registrations. But it says they searched "clinicaltrials.com" (vs ".gov"), which is the URL for PPD, a contract research organization (CRO) that does Pharma trials. I hope this was a typo.

And they report having looked at funded studies, another good way of establishing an "inception cohort" of trials. They report finding zero unpublished studies. This seems almost too good to be true...

In our study of NIH-funded psychotherapy trials for depression, we found 1/4 of these trials weren't published journals.plos.org/plosone/articl…

Maybe this is due to the fields being different. If it's true that 527 studies were conducted and 527 were published, then that's amazingly good.

But even then, it doesn't tell us about outcome reporting bias. (We couldn't do that, either, in our paper on psychotherapy trials.) To determine that, you have to compare the results in the publications w/ results in a gold standard source.