actively treated. Youâll also see that both psychotherapy and drug groups get significantly better. But, oddly, so does the placebo group. More bizarre still, the difference in improvement between placebo and antidepressant groups is only about 0.4 points, which was a strikingly small amount.
âThis result genuinely surprised us,â said Kirsch, leaning forward intently, âbecause the difference between placebos and antidepressants was far smaller than anything we had read about or anticipated.â In fact, Kirsch and Sapirstein were so taken aback by these findings that they initially doubted the integrity of their research. âWe felt we must have done something wrong in either collecting or analyzing the data,â confessed Kirsch, âbut what? We just couldnât figure it out.â
So in the following months Kirsch and Sapirstein analyzed and reanalyzed their data. They cut the figures this way and that, counted the statistics differently, checked what pills were assessed in each trial, and reexamined their findings with colleagues. But each time, the same results came out. Either they could not spot the mistake, or there simply was no mistake to spot. Eventually there seemed to be no other alternative than to take the risk and publish their findings that antidepressants, according to their data, appeared to be only moderately more effective than sugar pills.
âOnce our paper appeared,â Kirsch recalled, smiling, âthere was ⦠well, how can I put it ⦠controversy. The most significant critique was that we had left out many important trials from our meta-analysis. Perhaps an analysis that included those studies would lead to a different conclusion.â
Indeed, Thomas Moore, a professor at George Washington University, pointed this out to Kirsch by revealing that his meta-analysis had only assessed the published trials on whether antidepressants work. Their study had therefore failed to include the drug trials left unpublished by the pharmaceutical companies who conducted them. Kirsch and Sapirstein had been unaware that pharmaceutical companies regularly withhold trials from publication. When Kirsch looked into how many trials this amounted to, he was aghast at what he found: nearly 40 percent of all the trials on antidepressant drugs had not been publishedâa staggering amount by all accounts.
I asked Kirsch what he did next. âMoore suggested we appeal through the Freedom of Information Act to get the unpublished company studies released,â answered Kirsch. âOnce we were successful at that, we undertook a second meta-analysis which now included all the studiesâboth published and unpublished.â
As the results came in from this second meta-analysis, Kirsch grew even more alarmed. They showed that the results of his first study were plainly wrong: antidepressants did not work moderately better than placebos; they worked almost no better at all .
3
Moving away for a moment from the cozy sitting room in which Kirsch was recounting his unsettling series of discoveries, letâs backtrack a little so I can illustrate to you how the studies into antidepressants that Kirschâs meta-analysis surveyed were conducted.
To do this, imagine yourself in the following scenario: Youâve been depressed for at least two weeks (the minimum time needed to be classed as depressed, as youâll recall from Chapter 1). So you eventually decide to drag yourself to your doctor, who asks you if you would like to participate in a clinical trial. The aim of this trial is to test the effectiveness of a new antidepressant drug (which may cause some side effects if you take it).
Your doctor then explains how the trial will work: before you are given the antidepressant, your level of depression will be measured on something called the Hamilton Scale. This is a scale that runs from 0 to 51, and the task is to work out where you sit on this scale.
Your