We sort of have to walk through this one.
Act One: Irving Kirsch, a psychologist formerly at UConn and now in Hull, UK, is one of my heroes for his great work on the placebo effect and expectancy. Kirsch is also a person for whom the drug companies must keep a personal voodoo doll into which they regularly stick pins. Kirsch has enraged a lot of folks by doing detailed meta-analyses of all the data on the newer generation of anti-depressant medicines (the SSRIs and their relations, drugs like Prozac and Zoloft). He has shown in a couple of meta-analyses in the past that it is very hard to tell the difference between what these highly touted drugs do for depression and what happens when people take sugar pills.
His latest meta-analysis is: http://medicine.plosjournals.org/perlserv/?request=get-document&doi=10.1371/journal.pmed.0050045. Kirsch and colleagues here show that it appears that there is really a tiny group of depressed folks for whom the drugs show virtually all their effectiveness-- those with the most severe depression. If you have moderate or even somewhat severely depressed people, you can find no difference between the drug and the placebo group in the average clinical trial. Moreover, it is perhaps not even the case that for those most severely depressed patients, it is the drug that makes the difference. The trial results, Kirsch and colleagues point out, show rather that it's not that the drugs work better for those patients. It's that the placebos do worse. The difference between drug and placebo is therefore greater, but only because there is less placebo response, not because there is any greater drug response.
Act Two: Of course the companies are not about to stand still for any of this nonsense, so GlaxoSmithKline shot off a news release right away, claiming that Kirsch's analysis was flawed because he analyzed "only a small subset of the total data available." If true, this would be a serious charge against any meta-analysis.
Act Three: Along come journalists Jeanne Lenzer and Shannon Brownlee, writing in BMJ: http://www.bmj.com/cgi/content/full/bmj.39504.662685.0Fv1. They ask the $64 question-- if Kirsch and gang reviewed only a subset of all the data, how come? Kirsch for instance filed a request under the Freedom of Information Act to get all study data for 6 antidepressants from the FDA. The FDA then refused to release data from 9 out of the 47 trials that they identified as relevant--they still have not explained why those 9 trials were withheld. (It later was revealed that all 9 trials had shown negative results, indicating why the companies might not have wanted them to see the light of day.)
In other words, say Lenzer and Brownlee, GSK may in fact be quite correct that people like Kirsch cannot do a valid meta-analysis because they don't have all the data. But whose fault is that? How about the industry that has done its level best to suppress and conceal any trial data that looked to be bad for marketing, especially regarding antidepressants, as we have seen before: