Seriously, that "50% of all trials are never published" statement is really getting annoying...

Reposted from Zombiestatistics][
This post is in response to a really shoddy article on Pharmafile. If one of my trainees wrote something this badly researched, I would put them on a warning...
Article here.
Edit: I did add a comment on the original article, but it seems to have mysteriously disappeared. Glad I also posted it here.

It’s a shame that this article yet again perpetuates the “50% of trials are unpublished” zombie statistic by misrepresenting another not particularly robust article. The Thomson Reuters “Developments in clinical trials” article is not a new report or a new study. It simply cites (incorrectly) a recent article published in PLoS Medicine that looked for trials on CT.gov with posted results, then took a random selection of 600 trials from that subset to look for full publications. So actually, the results disclosure rate for the sample was 100%, as the results were available on a publically accessible website, and additionally, 50% of them had publications in academic journals (not all of which are publically available, as the headline suggests, because some will inevitably be behind paywalls). There are far better dissections of this PLoS Medicine article here and here.
The Thomson “non-study” also didn’t “find” anything about unregistered trials. It cited a BMJ paper on the non-publication of trials registered on CT.gov (71% publication rate in this one, more generally representative than the PLoS article because of the way the trials were selected) which included 7 references on studies on the quality of trial registration, only 1 of which was actually about unregistered trials. There’s a recent update to one of those papers that touches on the legislation about trial registration, if you’re interested.
What the Thomson article does report – if you actually read it – is that their search of the Cortellis Clinical Trials Intelligence database for trials ending between January 2000 and December 2009 showed that 40% of those trials had disclosed results (though I don’t know whether Cortellis just pulls in that information from CT.gov or whether there’s any additional publication searching done for that database). Running a crude search through that database corroborates the 40% (38% actually) but raises it to 43% if you eliminate Phase I studies. Without knowing how the “results” field gets ticked in that database it’s difficult to comment further on that number.

There are no latest data from Thomson Reuters. What this article does do is serve to perpetuate the "50% of trials are unpublished" zombie statistic yet again.

Post-post  update
Thomson Reuters Cortellis contacted me after I downloaded their article, as a sales call. When I explained why I had downloaded their article (and the issue) they said they would forward the issue to the editorial team. I wasn’t holding out much hope, but I have been contacted, and the result is that TRC have amended their article. Big thumbs up for information providers who are concerned with the accuracy of their content.
So, how do you kill zombies? You don’t. You take them apart, a piece at a time.

“These latest data from Thomson Reuters Cortellis will only serve to corroborate the stance taken by AllTrials, and fuel greater discussion over this divisive issue ahead of the European debate on the topic next month.”