Suicide and Antidepressants: What is the Evidence?

There has been increasing media scrutiny and criticism about the use of antidepressant drugs and their possible association with suicidal behaviours [1, 2]. This arose from initial concerns of Teicher et al. [3] in regard to fluoxetine, and there have been other reports [4–6] which have led authorities such as the UK Medicines and Healthcare Products Regulatory Agency [7] and the American Food and Drug Administration [8] to issue warnings about the use of selective serotonin re-uptake inhibitor (SSRI) antidepressants because of the presumed risk of increasing the likelihood of suicide.

It is timely to review the best currently available evidence, and in doing so it is necessary to provide a balanced view of the data, as well as an appreciation of its limitations. It is also important to appreciate that the science and art of medicine is to balance potential bene-fits and potential risks and nowhere is this more necessary than in the management of those who may have the potential to be suicidal.

Limitations of randomized controlled trials

While it has long been presumed that the treatment of depression would ameliorate suicide rates, it has been noted that ‘no single intervention has been shown in a well conducted randomized controlled trial to reduce suicide’ [9]. Such a comment is at first impression sobering, but it must be appreciated that there are formidable obstacles in proving the effectiveness of antidepressant measures in preventing suicide. For example, there is the ethical problem of including suicidal people in a placebo arm of a trial, and, perhaps even more daunting, is the fact that although suicide is dramatic and tends to remain in a clinician or researcher's mind, the reality of the situation is that it has a low base rate and it is impossible to select sufficient numbers of high-risk persons to participate in a randomized control trial. To illustrate this, it has been calculated that if one were to utilize the randomized controlled trials of antidepressants which have been published, then it would have been necessary to have recruited about 1.9 million subjects to detect a 20% decrease in suicide [10].

There are other research methodologies besides randomized controlled trials, particularly those which provide naturalistic real-world observations of depression and its treatment, and its impact upon suicidal behaviours. It is therefore pertinent to examine these.

Naturalistic studies

There have been few very long-term studies of the treatment of mood disorders. Indeed, the reports of Angst et al. stand alone in the length of their follow-up, as they described 40 and 44 years' follow-up of 406 patients with mood disorders, with and without long-term medication [11, 12]. They found standardized mortality ratios (SMRs) for suicide for untreated unipolar and bipolar patients of 38 and 29, respectively, compared with the treated patients' SMRs of 12 and 6. The difference was highly significant at the p < 0.001 level. They analysed their data further and demonstrated that the ‘logistic regression over all unipolar and bipolar patients who had died (n = 305) showed a significant suicide reducing effect of antidepressants’. They went on to state that ‘the effect was very strong in patients treated by antidepressants, by antidepressants plus neuroleptics, or by lithium plus antidepressants and/or neuroleptics.’

Such follow-up studies may be persuasive to clinicians, but there is a need for larger population studies. Probably the first such population depression intervention study was that on the Swedish island of Gotland in which a depression recognition and management program for general practitioners and the community resulted in a reduction in depression morbidity and fewer suicides [13, 14]. There has been debate as to whether the reduction in suicide was statistically significant, but Rihmer et al. demonstrated a statistically significant reduction in those who died by suicide who had had a depressive diagnosis [14]. Further work pertinent to this issue was that of Isacsson who had predicted a 25% decrease in suicide with a fivefold increase in antidepressant prescribing [15]. With the introduction of the SSRI antidepressants there was a marked increased in prescribing to the extent that antidepressant use in Sweden increased from 1% to 3.4% of the population between 1991 and 1996. This allowed a natural experiment to address Isacsson's hypothesis and he found that there was a 19% reduction in suicide between 1991 and 1996 and there were no groups (age, gender, county) where the rate decreased in the absence of increased antidepressants. He also analysed data for Finland, Denmark and Norway and found similar results.

Since then there have been further analyses from New Zealand [16], the US [17–19], Australia [20], Italy [21], Slovenia [22], Norway [23], Ireland [24] and Denmark [25]. All but the Slovenian and Italian data, and the Irish data for those under the age of 30 years, have supported the hypothesis that antidepressant prescribing has been associated with the reduction of suicide. Furthermore, a comprehensive recent analysis of SSRI use in 27 countries by Ludwig and Marcotte demonstrated a strong association between an increase in antidepressant prescribing and a reduction of suicide in those over the age of 15 years, although the data were not conclusive for those aged 10–15 years [26]. They noted that an increase of one pill per capita was associated with a 2.5% reduction in suicide rates, or, put another way, one suicide could be averted for every 300 000 SSRI antidepressant pills sold.

It is acknowledged in research that associations such as these do not necessarily imply a causal relationship. Indeed, while such studies may again be persuasive for most clinicians and researchers, information from other naturalistic research designs is also necessary.

Probably the most comprehensive study yet published has been that of Jick et al. who analysed data from the United Kingdom General Practice Research Database (GPRD) which contains over 35 million patient years of information [27]. Using a very rigorous research methodology, they delineated between 35 000 and 50 000 people who had had antidepressant treatment initiated with one of either two tricyclic antidepressants, amitriptyline or dothiepin, or one of two SSRI antidepressants, fluoxetine or paroxetine. Jick et al. reported that there were similar risks of emerging suicidal behaviour after starting each of these antidepressants. They also found that those who had been on the antidepressant less than 10 days were more than four times as likely to have their first suicidal behaviour compared with those who had been on their antidepressant for greater than 90 days before their suicidal behaviour, and the risk of suicide was even greater in that first 10 days, being 38 times more likely than those who had been on their antidepressant for more than 90 days. They reported that there was no trend between the drugs and they also noted that the gradation of risk was consistent with antidepressants not being immediately effective, with an increased risk in those who were newly diagnosed and newly treated.

In an accompanying editorial, Wessely and Kerwin [28] noted that ‘the results confirm that antidepressant prescription is indeed associated with suicidal behaviour, and strongly so. This simply means that antidepressants are being prescribed for the right indication, and that they do not immediately eliminate suicide risk. That we knew.’ They also noted perceptively that such results would be unlikely to allay community concerns, particularly ‘as society becomes increasingly obsessed with questions of risk and then blame’.

It is important to note that Jick et al. extended their study to examine adolescent suicide. They found no suicides in people aged 10–19 years who were on one or other of the study drugs. However, they did report that there were 15 people in that age group on the GPRD who had died by suicide, and none had received an antidepressant drug. One might, perhaps provocatively, ask why the media have not picked up on that and criticized practitioners for not using antidepressants in those people.

That the Jick et al. data were not simply an artefact of British practice has been demonstrated by subsequent reports from three other countries. Isacsson et al. [29] in Sweden reported a toxicological study of over 14 000 suicides between 1992 and 2000 and compared them with control subjects. They found that SSRIs were underrepresented compared with other antidepressants in those who died by suicide (OR = 0.83, 99% CI: 0.77–0.90). They also addressed the issue of suicide in younger people and found in the 15–19 years age group that SSRIs had a lower relative risk (0.14, 95% CI: 0.05–0.43) for suicide compared with non-SSRI antidepressants, and in those under the age of 15 years, there were 52 suicides and no SSRIs were found by toxicological screening, although there were seven who had other antidepressants detected. A recent study from the US has provided similar results, with Moskos et al. [30] reporting that in a series of 49 suicides aged 13–21 years, ‘not one demonstrated therapeutic or subtherapeutic levels of psychotropic medications upon autopsy.’ Furthermore, in as yet unpublished research presented at the International Association for Suicide Prevention meeting in Durban in September 2005, Sondergard et al. [25], in a pharmacoepidemiological study of youth suicide in Denmark between 1995 and 1999, found that none of the 42 suicides aged 10–17 years had been treated with SSRIs in the 2 weeks prior to their deaths, and none of the 37 boys who died by suicide had ever been treated with antidepressants at any time during the study period.

With such data from four different countries providing similar results, it is quite clear that there is no evidence from large population studies to implicate antidepressants in the precipitation of suicide. Indeed, the question arises about the risks of not treating depression in young people, and again one might raise the issue of criticism for not using SSRI antidepressants in this younger age group.

Notwithstanding the re-assurance from the previously noted studies, there are data which emphasize the need for clinical vigilance in regard to suicidal behaviours and the prescription of antidepressants. Thus Fergusson et al. analysed 702 randomized controlled trials of antidepressants involving 87 000 patients and, although they found no association between SSRIs and suicide, there was an association between SSRI prescriptions and increased attempted suicide [31]. However, there was no difference between SSRIs and tricyclic antidepressants. They calculated the number needed to treat to harm and derived a figure of 684 patients.

A similar meta-analytic study of pharmaceutical company data presented to the UK Medicines and Health Care Products Regulatory Agency was reported by Gunnell et al. [32]. In a review of over 40 000 patients in 477 randomized controlled trials, they found no evidence of increased suicide or suicidal thoughts. However, they did find what they described as ‘weak evidence… possible increased risk of non-fatal self harm’, and they calculated the number needed to treat for non-fatal self-harm and derived a figure of 759 patients. They extended their research to include a measure of the benefits of treatment and, on the basis of the meta-analytic work on fluoxetine data of Bech et al. [33], they reported that the number of individuals needed to treat for a response defined as both self-report and clinician report of much or very much improved was between four and seven patients.

It is fair to note that these numbers needed to treat for both harm and benefit do not necessarily apply to children and adolescents. Indeed, it is acknowledged that the data for the effectiveness of antidepressants in young people is not as robust as for adults, as is illustrated in the 27-country review of the impact of antidepressants on suicide by Ludwig and Marcotte [26]. Furthermore, it has recently been reported by Bridge et al. [34] that emergent suicidality in the medication-free psychotherapeutic treatment of depression in adolescents was 12.5%, a figure ‘comparable to rates observed in antidepressant trials’, and this suggests that one must be particularly cautious in the interpretation of data in this complex clinical area.

Notwithstanding the limitations in the studies reviewed, there are now convincing data to repudiate claims of increased risk of suicidal ideation or suicide with antidepressants, be they tricyclic or newer preparations. Nevertheless, the question arises of why there should be any increase of attempted suicide associated with such use, as is demonstrated by the Fergusson et al. [31] and Gunnell et al. [32] reviews. Naturally one can never preclude the possibility that there will be individuals who have an idiosyncratic response to antidepressants, or indeed to any medication. However, the answer may be more prosaic, and related to the natural history of recovery from depression. Thus, as long ago as in 1812 Benjamin Rush [35] cautioned about suicide occurring following apparent improvement; in 1892 George Savage [36] noted that ‘with the entry on convalescence suicidal attempts are common’; and in the mid 20th century, just before the advent of antidepressant medication, Mayer-Gross et al. [37] stated that ‘with beginning convalescence, spontaneously or after convulsion treatment, the risk of suicide once more become serious as retardation fades.’ These observations appear to be particularly important, as clearly the risk of suicide in the recovery phases of depression has been well documented for almost 200 years, and, with an historical perspective, it can be interpreted as a risk that is inevitable, rather than necessarily being associated with a specific treatment.

As a result of these recent studies it can be seen that both the potential risks and benefits can now be more fully appreciated and seen in perspective. This has contributed to a reassessment of some cautions. For example, there has been modification of the US Black Box warning and that has probably been brought about by the endorsement of over a dozen national professional and patient/carer advocacy organizations of the American Academy of Child and Adolescent Psychiatry and American Psychiatric Association submissions to the Food and Drug Administration [38]. Thus, the warning language now is confined to indicating that there has been an increased risk of suicidal thinking and behaviour in short-term studies of adolescents and children, and not that the drugs actually could cause suicidal behaviour. This difference may appear to be subtle, but it is an indication of a shift in interpretation of the data.

It has also been observed, particularly because of the strong association between child and adolescent mood disorders and suicide [39] and the finding of the absence of use of antidepressants in young people who die by suicide in the UK [27], Sweden [29], the US [30], and in Denmark [25], that there is the potential for legal action for not at least having a trial of antidepressant medication in young people with severe depression if nonpharmacological measures are ineffective [40].

Conclusion

It is evident that there will never be randomized controlled trials to demonstrate a reduction of suicide with antidepressants because of the constraints of the low base rate and the enormous number of subjects needed for such a study. However, despite the methodological challenges involved in clarifying this important clinical issue, there are increasing data about the effectiveness of antidepressants in reducing suicide. Furthermore, there are now calculations available to compare the numbers needed to treat for the emergence of any suicidal behaviour as opposed to the numbers needed to treat for clinical improvement.

It is important to recognize that all treatments have potential risks and we need to consider all of the evidence, including the historical perspective, with full disclosure and discussion with patients and their relatives. This includes evaluating the risk of not treating, as well as balancing the potential risks and benefits of treatment. We are now better informed in this regard, and clinical decisions can be made accordingly.