In our recent study published in the journal Psychotherapy and Psychosomatics, we showed that adults who start antidepressants for depression are 2.5 times more likely to attempt suicide when compared to those starting placebo. This equates to 1 in every 200 people attempting suicide as a direct effect of taking the antidepressant. This number is significant given that around 7.4 million people were prescribed antidepressants in England alone last year, and given international estimates that suggest around 1 in 20 suicide attempts will end in death. We further showed that the suicide rate is about 3 times higher with antidepressants, but due to the very low incidence rate, we acknowledged that the evidence was weak.
In a response to our paper, Hayes et al. have today issued a critique. Far from undermining our findings, their critique provides us with a unique opportunity to address important issues that were not covered in our original paper, due to space limitations and firm opposition from one reviewer. Their critique also provides us with an opportunity to reassert, on even firmer ground, that our original findings were indeed correct.
So where do Hayes et al find fault with our paper? Their main point, which is fair, is that we should have used a meta-analytic method. However, as we explain in our response, not using such a method was our deliberate choice, given previous analyses of the FDA database did not use such methods either. These papers are highly cited and very influential, so we considered it important to apply a similar statistical model.
Hayes et al then proceed to present the findings of several meta-analytic methods they applied, only to conclude that there is no evidence that antidepressants increase the suicide risk. This conclusion, as we explain in our response, is an artefact of the models they chose to use – models that turn out to be grossly inadequate for these particular suicide data (very low event rates and many trials with zero events in both antidepressant and placebo groups).
Hayes et al were also unaware (or else, preferred to ignore) that several suicides were misreported in this database, despite this being very clear in the scientific literature. We wanted to address this serious issue in the original paper, but one reviewer firmly objected. In our response to Hayes et al we make up for this omission, by showing that in the paroxetine approval program two placebo-suicides were misreported, as these occurred during the lead-in phase (i.e. before the trial began), so they must be removed. There are other misrepresentations, which we detail in our response.
With the two placebo-suicides from the paroxetine program removed, most meta-analytic methods (we used only methods that are deemed appropriate for these particular data), we found clear evidence that the suicide rate is higher in antidepressant arms relative to placebo. Moreover, when the data from both the fluoxetine and the bupropion programs were included, the effect was even more pronounced, with odds ratios ranging from 2.5 to 6.3, depending on the meta-analytic model used. The various meta-analytic findings and the statistical code are available freely online via https://osf.io/qzjva.
Finally, and most crucially, Hayes et al did not present meta-analytic results for suicide attempts, as if suicide attempts did not contribute to the suicide risk. We should not have to point out to Hayes et al that suicide attempts are the single most important determinant of suicide. The more people attempt suicide, the higher the suicide risk. Even when based on the uncorrected data, our analyses reveal a significantly increased risk of suicide attempts (both fatal and non-fatal attempts combined) in antidepressant arms relative to placebo that was largely consistent across meta-analytic methods (see results here: https://osf.io/qzjva).
In sum, these meta-analytic findings are consistent with the findings from our original paper, helping to further indicate that there is an increased suicide risk with antidepressants in clinical trials submitted to the FDA. The next step would be to examine, whether this increased risk replicates in representative real-world patients treated in routine care settings.
Dr Michael P. Hengartner, School of Applied Psychology, Zurich University of Applied Sciences, Switzerland
Dr Martin Plöderl, Department of Clinical Psychology, Paracelsus Medical University, Salzberg, Austria