Since their introduction in the late 1980s, the benefits and harms of selective serotonin reuptake inhibitors and serotonin-norepinephrine reuptake inhibitors have been the subject of huge debate. Because their harms – in particular suicidality – have often been explained away as if they were disease symptoms or only a problem in children, 7 – 10 we wished to quantify the risk of suicidality and violence when these drugs are given to adult healthy volunteers.

A 2012 systematic review of 33 trials in healthy volunteers documented various effects of selective serotonin reuptake inhibitors but only mentioned the adverse events in a few words. 6 The review was based on published articles and none of these lived fully up to the CONSORT guideline for good reporting.

The reporting of harms in drug trials is generally poor, with inadequate explanation of how they were collected, and often harms are missing altogether. 1 , 2 From 2011 to 2012, the Nordic Cochrane Centre received unredacted clinical study reports on antidepressants filed for regulatory approval at the European Medicines Agency and the UK Medicines and Healthcare products Regulatory Agency. We demonstrated selective reporting of major harms in the published articles of duloxetine and inconsistencies between the protocols and the clinical study reports. 3 Furthermore, we noticed that the adverse events tables in clinical study reports had hidden suicidal events due to the medical dictionaries and coding conventions used. 4 Based on the 70 clinical study reports we received, we found that antidepressants more than doubled the risk of suicidal and aggressive behaviour in children and adolescents. 5

Uncertainties during data extraction were resolved by discussion between the authors. If the reporting of harms was unclear, we contacted the authors of the trials for clarification. We assessed the risk of bias in the trials focusing on randomisation and blinding.

We defined harms as adverse events that were either suicidality or violence or were considered precursor events to suicidality or violence in the literature, with a particular focus on the list of criteria used by the Food and Drug Administration for its 2006 meta-analysis of suicidality in 100,000 patients ( Table 2 ). 12 – 18 We categorised the harms as suicidality, violence, activation events, psychotic events and mood disturbances.

We also included relevant clinical study reports on antidepressant drugs we received from regulatory agencies for our projects in this area. 3 – 5 We checked whether the clinical study reports had been published by searching for investigator names and drugs on PubMed and Embase.

We searched PubMed (1966 to December 2015) and Embase (1974 to December 2015) using the search strings specified in Table 1 . The searches were conducted in collaboration with an information specialist to ensure the inclusion of drugs that were not indexed in PubMed’s MeSH thesaurus. One researcher (AØB) screened all search results by reading titles and abstracts, and all excluded studies were rescreened by the same researcher who also read the included studies and extracted relevant trial data, e.g. design, population size, harms, discontinuations and funding, to a spreadsheet. A second researcher (PBD) extracted data on a randomly selected sample of 30 articles; as there were no discrepancies, we did not perform data extraction in duplicate.

According to our prespecified protocol (available from the authors), we included double-blind, randomised placebo-controlled trials of selective serotonin reuptake inhibitors or serotonin-norepinephrine reuptake inhibitors (called antidepressants throughout this paper) in adult healthy volunteers with no signs of a mental disorder. There were no language restrictions. We excluded trials in abusers of tobacco, drugs or alcohol and also imaging studies, as these have another objective and furthermore are of poor methodological quality. 11

Treatment of adult healthy volunteers with antidepressants doubled their risk of harms related to suicidality and violence, odds ratio 1.85 (95% confidence interval 1.11 to 3.08, p = 0.02, I 2 = 18%) ( Figure 2 ). The number needed to treat to harm one healthy person was 16 (95% confidence interval 8 to 100; Mantel-Haenszel risk difference 0.06). Two clinical study reports and one published trial reported the severity of the harms.

Adequate methods for sequence generation and concealment of treatment allocation were described for six trials A3–A6,A8,A11 and for concealment of allocation for two trials. A7,A12 The methods were not specified in five trials. A1,A2,A9,A10,A13 Adequate blinding methods were mentioned for seven trials A5–A8,A10–A12 and not specified in six trials. A1–A4,A9,A13 We did not look at attrition because the subject of our research was side effects, not beneficial effects.

Thus, we included 11 trials from our literature searches A1–A11 and two from the regulatory agencies A12,A13 in our meta-analysis. The drugs were citalopram, escitalopram, fluoxetine, paroxetine, sertraline or venlafaxine, in all cases given orally. Four trials were not industry sponsored and did not have industry employees among the authors. A2,A3,A5,A7 Trial characteristics, length of treatment and our predefined harms with reported severity are shown in Table 3 . The harms we found all occurred within the randomised phase of the trials, not during the withdrawal phase after the trials were finished. The median age was 30 years; 40% of the volunteers were women, and the median publication year was 2008.

We received 29 clinical study reports from the regulatory agencies, two of which fulfilled our inclusion criteria for the meta-analysis. None of them had been published. Most of the remaining clinical study reports described non-blinded bioavailability studies or drug interaction studies.

Thirteen of the 130 trials reported on at least one of our predefined harms, 10 of which were parallel group trials and three crossover trials. We could not include two of the crossover trials, as there were no data for each period separately. We included the first period of the third crossover trial, which ended prematurely due to carryover effects despite a four-week washout period. A1

Reporting of adverse events was generally inadequate; 63 trials (48%) did not report any adverse events or stated that there were none; 43 trials (33%) reported at least one adverse event; while 24 trials (18%) reported only the most frequently occurring adverse events or those leading to discontinuation. The source of funding was industry in 29 trials (22%), non-industrial sources in 47 trials (36%), mixed in 17 trials (13%) and not reported in 37 trials (28%).

We excluded 174 of these articles because the studies were crossover trials using antidepressant inhibitor, a placebo and a third drug, or because the studies were parallel trials that used a drug other than an antidepressant as an interaction drug. This left 142 articles on a total of 130 trials that fulfilled our inclusion criteria, 52 of which were two-period crossover trials. Ten different antidepressants were tested in the trials; citalopram ( n = 33) and paroxetine ( n = 27) were most commonly used.

Discussion Section: Choose Top of page Abstract Introduction Methods Results Discussion << Conclusions References CITING ARTICLES

The century-old belief that patients with depression are at heightened risk of suicide as they begin to recover and their energy and motivation return21 is being propagated everywhere, e.g. in the 2003 practice guideline from the American Psychiatric Association, which states that ‘clinical observations suggest that there may be an early increase in suicide risk as depressive symptoms begin to lift but before they are fully resolved’.22

Because of this deeply ingrained idea, many psychiatrists believe that when patients become suicidal on an antidepressant drug, it is not an adverse effect of the drug but a positive sign that the drug starts working.7,10 However, a systematic review from 2009 showed that the research that has been carried out contradicts this belief,21 and our review also suggests that it is wrong. We found that antidepressants double the risk of suicidality and violence, and it is particularly interesting that the volunteers in the studies we reviewed were healthy adults with no signs of a mental disorder. Our results agree closely with a review of paroxetine trials in both adults and children with mental disorders using regulatory data released after a court case. It included events both during treatment and in the subsequent withdrawal phase and found a doubling in hostility events (odds ratio 2.10, 95% confidence interval 1.27 to 3.48).23

While it is now generally accepted that antidepressants increase the risk of suicide and violence in children and adolescents5,12 (although many psychiatrists still deny this10), most people believe that these drugs are not dangerous for adults. This is a potentially lethal misconception.7,10,15,24

As far as we know, our review is the first of the risk of suicide and violence in healthy volunteers. It was inspired by David Healy’s work.7 In 2000, Healy published a study he had carried out with 20 healthy volunteers – all with no history of depression or other mental illness – and to his big surprise, two of them became suicidal when they received sertraline.25 One was on her way out the door to kill herself in front of a train or a car when a phone call saved her. Both volunteers remained disturbed several months later and seriously questioned the stability of their personalities.

In one of the two crossover trials we excluded because we did not have data on the first period separately, a healthy volunteer committed suicide, which was mentioned in both published articles.A14,A15 She had received duloxetine in increasing doses for 16 days, tapered off the maximum dose of 400 mg daily very quickly (in just four days according to the design of the study) and killed herself four days later while on placebo. The authors, several of whom were employees of Eli Lilly or owned stock in the company, judged her suicide ‘to be unrelated to study drug treatment’,A15 although it is well known that the suicide risk is high when an antidepressant is stopped abruptly.10,23 There was no more information about the suicide in the articles, and it was not included in the listing of adverse events we acquired from Eli Lilly, which only mentioned a woman who reported suicidal ideation twice while on placebo. As we do not know if this was the same patient, we asked Eli Lilly for access to anonymised data for the volunteer who committed suicide and the detailed person narrative, as we also wanted to know how it could be possible to state that the suicide was not related to duloxetine, but the company refused to give us the data.

In another of Eli Lilly’s studies, a healthy 19-year-old student who had taken duloxetine in order to help pay her college tuition hanged herself in a laboratory run by Lilly.26 It turned out that missing in the FDA’s files was any record of the college student and at least four other volunteers known to have committed suicide, and Lilly admitted that it had never made public at least two of those deaths.26

In the other crossover trial we had to exclude, an unknown number of volunteers discontinued paroxetine due to restlessness, tremor and other adverse events.A16 We contacted the corresponding author of the study who referred us to the first author, but despite several attempts of making contact via two different email addresses and phone (to the doctor’s assistant), this author never responded.

Exploratory analyses of the clinical study reports Although only two of the 29 clinical study reports were eligible for our meta-analysis, e.g. as the studies needed to be double-blind, two researchers (AØB and PBD) read them all (2224 pages) and extracted data independently, as we wanted to explore possible selective reporting of harms in the published articles. Nineteen clinical study reports reported on the harms we investigated and nine of these were published, but less than half of the harms were reported in the articles (21 of 50 events on antidepressants and two of four events on placebo). One of these studiesA17 was mentioned by David Healy7 who had spoken with the study investigator, Ian Hindmarch. It was a crossover trial of the interaction between sertraline and diazepam that was terminated due to unexpected adverse events after only four days, before the first phase had been completed and before any of the volunteers had received diazepam. All five volunteers in the sertraline group became agitated and four of them anxious, while one of seven volunteers in the placebo group became aggressive, agitated and anxious. The study was never published.