Robotic rejections of highly relevant letters to the editor of World Psychiatry about patient safety

By Peter C Gøtzsche

On 9 May, Bruce Arroll et al. argued in a letter to the editor of World Psychiatry, why patients with a depressive condition should not get a prescription for an antidepressant at their first visit to a general practitioner but at a later stage, if at all.

I agreed but had reservations with their arguments and argued why these drugs should not be used at all, for anyone, which I explained in a letter to the editor:

“They write that the effect of antidepressants in mild and moderate depression is small and may not be clinically relevant. They believe the effect is better in severe and very severe depression and give a number needed to treat (NNT) of only 4 for very severe depression. On this basis, they conclude that antidepressant medication should be considered in severe depression, but not at the first visit and as an alternative to or in combination with a psychological intervention.

In my freely available “Critical Psychiatry Textbook” I have explained why I believe antidepressants should not be used at all.

First, the reported effect is also small and irrelevant for patients with very severe depression, e.g. only 2.7 for patients with a baseline Hamilton score above 23 (page 73 in the book). The effect is 1.3 for milder degrees of depression, but this difference is likely just a mathematical artefact. Since the baseline scores for severe depression are larger than for mild depression, any bias will influence the measured result more in patients with severe depression than in those with mild depression. If we assume the bias caused by insufficient blinding because of the drugs’ adverse effects is 10% when estimating the effect in the drug group, and, for the simplicity of the example, that there is no bias in the placebo group and no improvement between baseline and the final visit, then a Hamilton baseline score of 25 would still be 25 after treatment. But because of the bias, there would be a 2.5-point difference between drug and placebo. If the baseline is 15, that difference would only be 1.5.

Second, NNT for antidepressants is highly misleading for at least five reasons (page 78 in the book). Most importantly, NNT is only about a benefit and completely ignores that the drugs have harms, which are much more certain to be experienced than their possible benefits. If benefits and harms are combined in a preference measure, it is therefore not likely that an NNT can be calculated. We can only calculate the number needed to harm (NNH). Dropouts during antidepressant trials illustrate this. Since 12% more patients drop out on drugs than on placebo, the patients prefer the placebo for the drugs, which is a net harm (with a NNH of about 25).

Third, antidepressants increase the risk of suicide while psychotherapy decreases the suicide risk (my textbook and our systematic review).”

The editor of World Psychiatry, Mario Maj, rejected my letter 8 days later:

“Unfortunately, although it is an interesting piece, it does not compete successfully for one of the slots we have available in the journal.”

That sounded familiar. Six years ago, it also took exactly 8 days for Maj to reject a letter Bob Whitaker and I had asked him to publish, with exactly the same text: “Unfortunately, although it is an interesting piece, it does not compete successfully for one of the slots we have available in the journal for letters.”

I wonder if the journal used artificial intelligence already in 2017 instead of an editor when rejecting letters and articles.

Back then, we wanted Maj’s help in elucidating why so many young people with schizophrenia died so young, referring to a large cohort study he had published. We wrote that his help “would be a great service to psychiatry, the patients, and everyone else with an interest in this vitally important issue.” But Maj was unmoved and when we appealed his decision, he did not reply. Perhaps the robot was out of order.