Caution also needs to be taken regarding the fact that all but 1 of the trials had some form of pharmaceutical industry sponsorship.

I can't find the specific article I was looking for, but Ben Goldacre writes on this problem at http://www.badscience.net/

Short version: The main problem with pharma-sponsored research isn't that they've necessarily faked the data -- though that happens far too often -- it's the bias introduced by simply not publishing anything that doesn't support their preferred conclusion. Then any meta-analysis will show for instance, "19 out of 20 studies showed a positive correlation between foo and bar". Unknown (because they were never published) is that 47 other studies showed the opposite.