A meme that continues to appear regularly in the media and in the comments section of this blog is “antidepressants are no better than placebo.” It is a false belief based on a faulty interpretation of the evidence. A new study has just come out that should put the final nail in its coffin. Antidepressants are indisputably effective when prescribed appropriately for moderate-to-severe major depression. Part of the confusion is that they have often been prescribed inappropriately for mild depression or sadness, where there is no evidence to support their use. They are serious drugs for a serious illness, not “happy pills.”
The “antidepressants are placebos” meme may have originated with Irving Kirsch. In a 1998 meta-analysis, he found that 75% of the apparent effects of antidepressants were placebo effects, and he suggested that the other 25% probably were too.
A 2008 New England Journal of Medicine article by Erick Turner et al. “Selective Publication of Antidepressant Trials and Its Influence on Apparent Efficacy” caused quite a stir. They did a systematic review of both published and unpublished antidepressant trials. Overall, only 51% of studies were positive, but 94% of published studies were positive. Because of publication bias, the published literature overestimated the benefits of antidepressants; it conveyed an effect size of 0.31, nearly one third larger than the effect size derived from the data as a whole.
Selective reporting deprives researchers of the accurate data they need to estimate effect size realistically. Inflated effect sizes lead to underestimates of the sample size required to achieve statistical significance. Underpowered studies — and selectively reported studies in general — waste resources and the contributions of investigators and study participants, and they hinder the advancement of medical knowledge. By altering the apparent risk–benefit ratio of drugs, selective publication can lead doctors to make inappropriate prescribing decisions that may not be in the best interest of their patients and, thus, the public health.
Commenters on that article tended to overlook one important fact: each drug, when subjected to meta-analysis, was shown to be superior to placebo. All the evidence agreed that antidepressants were better than placebo, just not as much better as the published literature would lead us to believe.
Kirsch et al. re-evaluated the data for four of the drugs covered in the Turner analysis, and they calculated an effect size of 0.32, almost identical to the 0.31 found by Turner. Despite this initial agreement, they reached opposite conclusions about efficacy based on different understandings of what these effect size numbers mean in terms of clinical significance. I wrote about their disagreement previously. As Turner elegantly explained, the published studies posited a glass containing 0.41 liters, and Turner found that the glass contained 0.31 liters, which he sees as a partially full glass, whereas Kirsch considers a glass containing 0.31 liters to be equivalent to an empty glass. For comparison, psychotherapy has an effect size of only 0.22.
A more definitive study
The new study was published February 21, 2018, in The Lancet. The lead author is Cipriani, but both Erick Turner and John Ioannidis (another familiar name, the author of “Why Most Published Research Findings Are False“) are among the co-authors. They looked at published data from 522 randomized double-blind controlled trials testing 21 different antidepressants, and they sought out unpublished data. In all, the data covered 116,477 subjects who had been properly diagnosed with major depressive disorder and were treated for at least eight weeks. Their criterion for “effectiveness” was a reduction of depressive symptoms of at least 50% on an observer-rating scale (better than subjective self-reporting). They found that each of the 21 antidepressant drugs was more effective than placebo.
The accompanying Forest plot (AKA “blobbogram”) is impressive:
The blue squares indicate the average odds ratio from studies of each drug, with amitriptyline leading the pack with an OR of 2.13 times compared to placebo and reboxetine in last place with an OR of 1.37. The black horizontal lines indicate the 95% confidence intervals. That means that while the measured value can’t be guaranteed to be precisely accurate, they are 95% confident that the actual value lies somewhere within that range. Note that none of the confidence intervals crosses the vertical 1.0 line that would indicate an equal response to placebo and drug. This is about as good a Forest plot as you could hope to see.
While all antidepressant drugs were effective, some were more effective than others. Fluoxetine (Prozac) was one of the less effective ones. They also looked at dropout rates to gauge acceptability. Clomipramine was the least acceptable. They have made their entire dataset available to the public. The information may help clinicians decide which drugs to choose for their patients.
Conclusion: A blow to antidepressant denialism
This new evidence is impressive. Only the most perversely recalcitrant antidepressant denialists will continue to say that the drugs are no more effective than placebo. The debate is over. Questions remain about effect size and prescription guidelines, but it has been satisfactorily established that antidepressants (all of them) work better than placebo. Depression kills; antidepressants can help prevent suicides, relieve suffering, and improve quality of life and functioning. Thanks for that, Big Pharma.