Peter Kramer critiques a new analysis of old data on antidepressants.
Posted Feb 26, 2008
Journalists have been contacting me today about a study (published online by Irving Kirsch, of the University of Hull in the UK, and others) that concludes that the new antidepressants—Prozac, Paxil, Effexor, and Serzone—confer very little benefit. The report is getting splashy play around the world—Google lists over 300 media reports, mostly abroad. One commentator, at a CBS News political blog site, has asked why the U. S. press has not shown more interest. Based on the phone calls, I suspect that we will get coverage here, but science writers who have given this new research a miss may simply have been showing good judgment.
Kirsch has obtained a full version of FDA data that he and others had already analyzed in partial form. (Those reports received a respectful response in the press here.) A recent New England Journal of Medicine article on published and unpublished FDA drug studies—it showed that trials with negative results rarely saw the light of day—used the same material; in fact, the Journal study covered more medications.
As I’ve written elsewhere, the FDA data is bad data. Pharmaceutical companies, rushing to get drugs to market under a system that ignored failed trials and rewarded successful ones, recruited subjects who may or may not have met criteria for disease. The result was, to quote the text of the Kirsch monograph, that "The response to placebo in these trials was exceptionally large." Actually, what’s at issue may not be true placebo responses, based on ill people’s hopes and expectations in response to a pill. It may be that fairly healthy people with low moods simply got better, due to a natural waxing and waning of their emotional states.
In the current study, Kirsch found that when compared to healthier subjects, severely depressed subjects got more benefit from the drugs both absolutely and relative to the placebo. (The absolute peak was at the low end of the severely depressed category.) That’s what critics of these studies, including those who believe the antidepressants are effective, had predicted: because the less depressed groups had non-patients mixed in, there the noise would drown out the signal. Kirsch thinks the emperor has no clothes—the drugs don’t work. But in demonstrating high placebo response rates, his analysis may merely be confirming that a number of the subjects in the drug trials didn’t have depression in the first place. Again, this analysis is less about the medications, which have proved moderately effective in numerous studies, than about the past politics of presenting data to the FDA.
We can always use more and better studies of antidepressant efficacy in various populations and for various disorders. But regarding the FDA data, perhaps it's time to say: Enough, already!