Here is the question that we need to ask ourselves: Does the essay further public understanding of what science is telling us about the merits of antidepressants? Or does it rely on a misrepresentation of the science in order to protect the image of the drugs?

In his essay, Dr. Kramer writes specifically about research conducted by Irving Kirsch, a psychologist at the University of Hull in the United Kingdom, who detailed his findings in his book The Emperor's New Drugs (which was also reviewed by Dr. Angell.) He also writes about a study by Robert DeRubeis, a psychologist at the University of Penn, which was published in JAMA in 2010.

First, Kirsch's work and Dr. Kramer's review of it.

The Emperor's New Drugs

In his  research, Kirsch analyzed the results of industry-funded trials submitted to the Food and Drug Administration for four antidepressants: Prozac, Effexor, Serzone, and Paxil. As Kirsch noted, these trials -- with one exception -- were conducted in patients who, at study entry, were severely depressed. In 34 of the 35 trials Kirsch reviewed, the mean baseline score for the patients was 23 or greater on the Hamilton Depression Rating Scale (HDRS), which is a score characteristic of "very severe depression."

One reason that pharmaceutical companies seek to enroll people who are very depressed into their clinical trials is because they know that it is in this patient group that their drugs are mostly likely to show a benefit over placebo. Once the FDA has approved their drugs, the pharmaceutical companies can then market them to people with mild depression, regardless of whether the medications are effective in that population. In most of the industry-funded trials of SSRIs, the patients had to have a baseline score of at least 20 on the HDRS, which meant that those with mild to moderate depression were explicitly excluded.

In his review of the FDA data for the four drugs, Kirsch found that symptoms in the medicated patients dropped 9.6 points on the HDRS, versus 7.8 points for the placebo group. This was a difference of only 1.8 points, and the National Institute for Clinical Excellence in Britain had previously determined that a three-point drug-placebo difference was needed on the Hamilton scale to demonstrate a "clinically significant benefit." Kirsch found that it was only in the very severely depressed patients -- basically those with a baseline HDRS score over 28 -- that the drugs provided a clinically significant benefit.

On page 31 of his book, Kirsch writes: "In examining baseline depression scores (that is, measures of how depressed the patients were before the clinical trial began), the first thing we noticed was that all but one of the trials had been conducted with patients whose scores put them in the ‘very severe' category of depression . . . in other words, our findings of a clinically insignificant difference between drug and placebo was based primarily on data from those patients who are the most severely depressed according to the APA and NICE classification scheme."

So how does Dr. Kramer, in his essay, "defend" antidepressants in the light of Kirsch's report? Let's go over this point-by-point.

First, he writes that Kirsch "found that while the drugs outperformed the placebos for mild and moderate depression, the benefits were small." This, of course, is not what Kirsch found at all. The studies didn't involve patients with mild to moderate depression (except for the one study.) What Kirsch found was that in the FDA trials, the antidepressants didn't outperform placebo, in a clinical meaningful way, for patients with severe depression. 

That, of course, is a finding that would cause readers to seriously wonder about the merits of the drugs. But rather than write about Kirsch's actual findings, Dr. Kramer crafted a sentence that tells of how the drugs provide a small benefit even in mild to moderate patients. As such, he is reassuring readers -- even though falsely so -- that antidepressants provide a benefit to the larger universe of patients that take these drugs. And the implication is that the benefit must be quite marked for the severely depressed. 

Having misrepresented Kirsch's findings, Dr. Kramer then writes that "the problem with the Kirsch analysis -- and none of the major press reports considered this shortcoming -- is that the FDA material is ill suited to answer questions about mild depression." The reason, Dr. Kramer explains, is that "companies rushing to get medications to market have had an incentive to run quick, sloppy trials," and in their haste, they "often [enroll] subjects who don't really have depression." It is these non-depressed patients who then become counted in the trial results as placebo responders, because, Dr. Kramer writes, "no surprise -- weeks down the road they are not depressed."

I have to confess that this is a paragraph that took my breath away. Dr. Kramer makes it seem that Kirsch's review focuses on mild to moderate depression (it doesn't);  then he explains that the reason that Kirsch found that the drugs provide only a slight benefit to those patients in the FDA trials is that the drug companies enroll patients who aren't really depressed at all (when in fact the study criteria required patients to be severely ill); and finally he concludes that when those non-depressed patients end up in the placebo arm of the study, they  show up as improved and thus as placebo responders. The "improvement" of the placebo group, Dr. Kramer writes, "may have nothing to do with faith in the dummy pills; it is an artifact of the recruitment process."

So, readers of the New York Times piece can only conclude this: The industry-funded trials used for FDA approval were in large part conducted in patients with mild depression, or in patients who weren't depressed at all, and that is why the drugs only slightly beat placebo. The results would  have been markedly different in patients who were really depressed. Plus, even in these flawed trials, antidepressants produced a small benefit in the mild-to-moderate group. 

Placebo Washouts and Biased Trial Designs

Now let's go to Dr. Kramer's analysis of the study by Robert DeRubeis and his collaborators.

As might be expected, the drug companies in fact design their trials in a manner expected to suppress the placebo response rate. This is done through what is known as a placebo washout period, which may last a few days to two weeks. All patients enrolled into the study -- who may have to be taken off an antidepressant they might have been on -- are given a placebo in single-blind fashion (the investigators know it is a placebo; the patients do not.) Those who get better on placebo in this washout phase are then excluded from the study. Only those who don't respond to a placebo are randomized into the trial. As such, trials with this design might be better described as "drug versus initial non-responders to placebo," and of course this is a design that is supposed to reduce the number of placebo responders in the final results.