The Fluoxetine and Suicide Controversy

A Review of the Evidence

David Healy

Academic Sub-Department of Psychological Medicine, North Wales Hospital, Denbigh, Chwyd, Wales



1. A Review of the Evidence

2. Toxicity Indices

3. Catecholamines and Suicide

4. Serotonin and Suicide

5. Databases versus Case Reports

6. A Plausible Mechanism?

7. Conclusions and Management Recommendations


Evidence is emerging that a range of psychotropic drugs may precipitate akathisia and/or reactions in predisposed patients. The use of fluoxetine has hitherto been the most potable example of this occurrence. These reactions may foster the genesis of suicidal ideation in a small proportion of patients. At present, it is not clear what biological mechanisms may underlie this finding. The serotonin (5-hydroxytryptamine: 5-HT) system may be involved in these reactions.

The best management of such reactions will involve counselling patients beforehand about the possibility of these reactions, stopping treatment with the agents if such a reaction is suspected, or adding an agent with 5-HT1A antagonistic properties (e.g. propranolol) to the treatment regimen.

In February 1990, a report by Teicher and colleagues appeared in the American Journal of Psychiatry claiming that 6 patients who had been taking the selective serotonin (5-hydroxytryptamine: 5-HT) reuptake inhibitor (SSRI) fluoxetine, had become intensely suicidal. The present article reviews this and other evidence in an attempt to answer the question: can fluoxetine lead to the emergence of suicidal ideation? In addition, the appropriate management for such an occurrence is discussed.

1. A Review of the Evidence

The article by Teicher et al. caused considerable controversy, probably for 3 reasons.

First, the newer generation of antidepressants, such as fluoxetine, were specifically designed to be less toxic in overdose than older agents. Therefore, they were presumed to be less likely to cause fatalities through suicide. In contrast, the observations of Teicher and colleagues suggested that these agents somewhat perversely cause suicide fatalities, albeit for a different reason.

Secondly, this report appeared to go against observations that suggested that antidepressants acting on the noradrenergic system were more liable to lead to suicide than those acting primarily on the serotonergic system. Thirdly, the findings contradicted research that points to the existence of a serotonergic deficit in individuals who commit suicidal acts.

There was immediate criticism of the article by Teicher et al. The main criticisms are listed in table I. Subsequently, a meta-analysis of clinical trials involving 3065 patients who had received fluoxetine was performed. These analyses suggested that fluoxetine was unlikely to lead to the emergence of suicidal ideation.

Nevertheless, a series of case reports involved 1 to 6 patients have continued to emerge implicating fluoxetine in the generation of suicidal ideation. In addition to these case reports, a reanalysis of the data presented by Fava and Rosenbaum was carried out. This has suggested that the data can be interpreted such that fluoxetine does indeed lead to the emergence of suicidal ideation. The frequency of suicide ideation in patients receiving fluoxetine was somewhat higher than that in recipients of tricyclic antidepressants.

2. Toxicity Indices

In the mid 1980s, the use of mianserin was being questioned because of reports of drug-induced agranulocytosis. As part of a programme aimed at 'saving' the agent, considerable research was done on the number of deaths that occurred per million prescriptions of antidepressants. In the course of this and other work, the notion of a fatal toxicity index was developed.

This index indicated that a number of commonly prescribed tricyclic antidepressants, particularly desipramine, dothiepin and amitriptyline, were associated with a great risk of death from overdose than other antidepressants. Compared with these 'unsafe' compounds, mianserin was found to be a relatively safe agent. This was despite that fact that it induced agranulocytosis with a marginally higher frequency than other antidepressants (a claim that has not been fully substantiated).

A considerable part of the initial appeal of the SSRIs lay in their relative safety in overdose. While there have been insufficient prescriptions as yet to construct fatal toxicity indices for each of the SSRIs, these drugs have been specifically designed to be safe in overdose. As a result, it can be expected that their toxicity index will be low.

A standing comment in presentations over recent years has been that if the tricyclic antidepressants were to be developed now, they would not be approved because of their toxicity. This suggestion has led to the proposal that any programme aimed at suicide reduction should encourage the replacement of tricyclic antidepressants as first-line treatments for depression, with SSRIs. However, some have argued that in practice this would make very little difference to the rates of suicide in general practice.

It is against this background that the questions raised by the fluoxetine controversy must be judged. When considering the issue, a number of authorities have pointed to the clear benefits gained from the use of fluoxetine in terms of its relative safety in overdose and in patients with a range of other medical conditions, such as cardiovascular disease. Such safety is clearly desirable for most depressed patients. However, this property is unlikely to carry much legal weight if some individuals are inevitably led to take their own or others' lives as a direct consequence of receiving fluoxetine. The fact that pertussis vaccines reduce the overall likelihood of brain damage from the pertussis virus does not prevent legal action in the event of vaccine-induced brain damage.

Moreover, it has been pointed out in media reports that antidepressants are often developed using clinical trials conditions that differ substantially from those found in the real world. It can be argued that the clinical trials performed in the development of the early tricyclic antidepressants (some of which reached the market place without any clinical trial programme at all) were biased. This was because they focused heavily on inpatient samples of severely depressed patients. However, criteria for entry into current antidepressant trials specify that patients should not be suicidal, should not have any co-existing medical conditions, should not be receiving any other treatment, and should not have bipolar disorders. Increasingly, the patients who enter such trials are recruited from primary care settings and have depressive episodes of relatively recent onset and relatively mild to moderate severity.

Improvements have been made in the conduct of clinical trials, and there is now a broader definition of drug-induced adverse events. Despite this, it can be argued that the use of sophisticated post-marketing surveillance methods is the only accurate way of assessing the possibility of suicidal ideation induced by the newer agents. Available post-marketing surveillance data from individual countries do not suggest that any of the SSRIs are particularly liable to cause aggressive, impulsive or suicidal acts. If it is later established that they or other antidepressants do cause such effects, the question would arise as to whether this is a direct or indirect effect.

Teicher at al. appear to have initially believed that the effects they described arose in a manner that was not open to psychosocial intervention but later have perhaps modified their position.

3. Catecholamines and Suicide

As early as the mid 1960s, it was suggested that monoamine oxidase inhibitors (MAOIs) might be more likely than tricyclic antidepressants to lead to suicide.

It was thought that this might occur via a MAOI-induced increase in catecholamine [dopamine, noradrenaline (norepinephrine) and adrenaline (epinephrine)] levels. This effect might increase drive and arousal more rapidly than the more sedative tricyclic antidepressants. Activating patients in this way might enable them to act on their latent suicidal intentions. This idea never received any clear empirical support; however, it has never fully gone away. Indeed, there have been suggestions that the newly launched reversible monoamine oxidase A-selective agent, moclobemide, might be particularly likely to induce suicidal ideation and lead to suicide attempts.

More specifically in 1988, Damluji and Ferguson reported the occurrence of desipramine-induced dysphoria and suicidal ideation in 4 patients. As desipramine had a preferential effect on catecholamine rather than serotonin reuptake inhibition, this appeared consistent with the ideas noted above.

These ideas also found some support from a trial by Rouillon and colleagues. These investigators studied the long term outcome of patients maintained on maprotiline, a catecholamine reuptake inhibitor, compared with those on placebo. They found that maprotiline alleviated symptoms of depression and reduced the risk of relapse more effectively than placebo. However, use of the agent also led to a greater number of individuals attempting and completing suicide during the follow-up year.

These data have interfaced with ideas that serotonin is implicated in the control of impulsive behaviours, and that SSRIs might be associated with reduced rates of impulsivity and attempted or completed suicide. It is clear that SSRIs have effects on appetitive behaviours that appear loosely linked to an 'impulse system'; hence the potential usefulness of SSRIs in eating and sexual disorders. Unfortunately, the field of impulse disorders remains unstandardised, and it is not clear which disorders qualify as impulse disorders. As a result, it is not known whether SSRIs have any specific effect on impulsivity.

4. Serotonin and Suicide

In 1973, Van Praag and colleagues claimed that probenecid reduced cerebrospinal fluid (CSF) levels of 5-hydroxyindoleacetic acid (5-HIAA) the metabolite or serotonin, in depressed individuals more than it did in nondepressed individuals. This finding found some support from other investigators. Asberg and colleagues also reported a reduction of 5-HIAA levels in the CSF of depressed patients. This latter finding, however, was not clearcut and was not associated with any one category of depression. Subsequent investigation has suggested that reduced CSF 5-HIAA levels is not a feature of depressive illness.

Nevertheless, investigation in this area continued. In 1981, Traskman, Asberg and colleagues found that depressed patients as a group do not display a particular reduction of 5-HIAA levels. However, they claimed that there was a disorder of serotonin metabolism in those who are depressed and suicidal. Subsequently, the same group reported that the correlation between reduced 5-HIAA levels and suicidal behaviour was one that cut across diagnostic categories.

In supporting this claim, the investigators presented data suggesting that altered 5-HIAA metabolism might predict future suicide attempts. Others have pointed to correlations of 5-HIAA levels with previous suicide attempts or aggressive acts. Still others have accounted for heterogeneity in the data by appealing to distinctions between intentional and impulsive acts and distinctions between self-injury and suicide proper.

It would appear that initial claims regarding reduced 5-HIAA levels in the CSF were readily accepted by the biological psychiatry community. In contrast, subsequent failure to replicate the original findings have not been. For example, Annitto and Shopsin reported that the majority of 14 studies of CSF 5-HIAA levels that had been performed by 1979 did not support the conclusion that depression was associated with reduced 5-HIAA levels.

This did not prevent investigators at the National Institute of Mental Health predicting responsiveness to antidepressants based on alternations in CSF 5-HIAA levels. It was hypothesised that agents that supposedly increase serotonergic function, such as SSRIs, should be more effective in depressed patients with reduced levels o CSF 5-HIAA. It followed that agents that preferentially act on catecholamine reuptake might be expected to be of greater benefit in individuals with reduced catecholamine metabolism. However, this has not been found to be the case.

Quite apart from these negative findings, in general, studies of 5-HIAA have not been vigorously controlled methodologically. They have not adequately controlled for the effects of physical activity or alcohol intake, both of which are clearly implicated in suicidal or aggressive states. Indeed, the basic question of whether 5-HIAA levels in CSF taken from lumbar samples reveal anything more than the state of serotonin metabolism in the lumbar spinal cord has not been settled.

Efforts to approach the problem by looking at other aspects of serotonin metabolism have not clarified the situation. Evidence in favour of 5-HT2 receptor and [3H]-imipramine binding site abnormalities in post mortem brains of suicide victims has been cited. Unfortunately, few if any of these findings have been replicated. In addition, just as with work on 5-HIAA, post mortem brain research is subject to many confounding methodological variables. The contaminating influence of such factors has not been well controlled for.

Despite all these caveats, the idea that there is an abnormality in serotonin metabolism in suicidal states has gained hold. Indeed, in recent months reports of cholesterol-lowering agents possibly precipitating aggressive or suicidal acts have been greeted with interest. There have been a number of proposals linking this effect with an association between suicide and serotonin deficits, by means of an effect of lipid lowering on serotonergic metabolism. As Owens has noted, however, the existence of a biologically plausible explanation for this finding may prevent further indepth analysis of the situation. In a broadly similar manner, were the existence of an association between reduced serotonin levels and suicide to be confirmed, far from solving the problem, this finding would risk compromising the analysis of the fluoxetine and suicide data.

Even if the findings of altered serotonin metabolism in suicidal states were to be replicated, it is not clear that treatment with SSRIs would produce or reduce suicidal ideation. Even when depressed individuals with reduced 5-HIAA levels are selected for treatment with SSRIs, responses to treatment do not appear to be particularly enhanced or compromised.

5. Databases versus Case Reports

In reply to the case reports of fluoxetine-induced suicidality. Beasley and colleagues scrutinised the Eli Lilly database for evidence of increased suicidality in patients receiving fluoxetine. No such evidence has been found. These data from several thousand patients, and the evidence that fluoxetine reduces suicidal ideation, must on any scientific scale outweigh the dubious evidence of a handful of case reports.

There are, however, problems with the use of such a database. The evidence collected as part of the clinical trial programme for fluoxetine in the early 1980s did not specifically focus on the question of suicidal ideation. There was no reason to have such a focus. Accordingly, scales that might be particularly sensitive to variations in mental state and associated induction of suicide were not used. The primary variable used in the database was ratings on item 3 of the Hamilton Rating Scale for Depression (HRSD). This is a insensitive item, particularly when it comes to rating possibly emergent suicidal ideation.

Furthermore, clinical investigators will tend to mark down any scores on this item in line with a general marking down of HRSD scores as patients in a trial improve. Even in patients in whom suicidal ideation may have emerged or increased, there may be a simultaneous improvement in other aspects of their mental state. This effect would have masked increased ratings on item 3.

In addition, patients have a natural bias to minimise cognitive dissonance and so would be unlikely to volunteer information regarding emerging suicidal ideation. This is particularly true if they were experiencing improvements in other aspects of their mental state. Unless the investigators were suspicious of the induction of suicidal ideation, it is unlikely that they would have picked up such emerging tendencies.

Case reports are clearly an unreliable form of information. Several criteria have been proposed to enhance the validity of conclusions that may be drawn from such reports (see table II). To these we might add the criterion that there should be a clinical plausible mechanism that might make sense of the claim of an adverse effect.

While the initial reports by Teicher et al. of suicidal ideation after fluoxetine administration were clearly compromised on many of these criteria, subsequent reports have not been. Across all reports there is a broad correspondence regarding the time of emergence of suicidal ideation - about 10 to 14 days after initiation of fluoxetine treatment. It is clear from many reports that some individuals were prescribed fluoxetine only and in currently recommended doses. The effects have been clearly documented. Although suicidal ideation is a manifestation of depressive illness, intense suicidal ideation of the form reported is less commonly encountered. Furthermore, it certainly would not be expected to emerge frequently in individuals taking placebo.

With this particular adverse effect, there are problems with a rechallenge. Few investigators or patients would be anxious to revisit a compound that had caused suicidal ideation. Rothschild and Locke, however, did just this in 3 patients. They found that whilst suicidal ideation cleared on discontinuation of fluoxetine, it reemerged with readministration of the agent. However, readministration of fluoxetine was not done in a double-blind manner, and hence these reports are open to some bias. Creaney et al. found reemergence of suicidal ideation in 2 individuals who had experienced these effects after receiving fluoxetine when they were rechallenged with SSRIs other than fluoxetine. This suggested that the problem of a possible induction of suicidal ideation was not specific to fluoxetine.

6. A Plausible Mechanism?

The induction of suicidal ideation by fluoxetine seems to be in opposition to prevailing knowledge of the causes of suicide and mechanism of action of SSRIs. However, a possible mechanism mediating such an effect can be postulated.

It has commonly been suggested that individuals who experienced fluoxetine-induced suicidal ideation had developed akathisia. Independent of such reports, there have also been case reports implicating fluoxetine in the development of akathisia. Furthermore, fluoxetine seems to be associated with a higher incidence of extrapyramidal adverse effects than other antidepressants. It has also been noted that the induction of akathisia by antipsychotics may lead to a precipitation of suicidal attempts.

The phenomenon induced by Fluoxetine may not be accurately described by 'akathisia'. Akathisia traditionally refers to a visible restlessness and pacing. While this can be observed in patients receiving fluoxetine, it is not clear that the akathisia referred to in the above case reports has involved this type of effect. The states being referred to might also be labelled agitation or dysphoria rather than akathisia. Unfortunately the phenomenology of such drug induced restlessness/dysphoria is not well clarified.

There are 2 mechanisms, possibly related, whereby drug-induced 'akathisia' may lead to suicidal ideation. First, it would seem likely that some depressed individuals might misattribute the development of drug-induced dysphoria or agitation to a worsening of their mental state. Seeing this as a worsening of their depressive illness despite drug treatment, they may conclude that their case must be hopeless.

An alternative mechanism has recently emerged. In 1989, Weissman and colleagues reported an association between suicidality and panic disorder. In their analysis of findings derived from the epidemiologic catchment area study, panic disorder appeared particularly likely to be associated with both suicide attempts and completed suicides. In response to criticism from others, these authors have revised their estimates downwards of how likely panic is to lead to suicide. However, they still argue that the data point strongly toward an association between panic and suicide, particularly when the symptoms of panic occur during an episode of depression.

Indeed, panic may provide a final common mechanism through which a number of unrelated states lead to suicidal ideation. For example, Creaney and colleagues reported 1 patient who became suicidal following a dissociative reaction to SSRIs. This reaction was characterised by pronounced depersonalisation and derealisation. Such a mechanism may furthermore underline instances of paradoxical aggression which have been reported following administration of a variety of antidepressants.

Against this background, the notion that the precipitation of agitation by an antidepressant might lead to suicidal ideations/suicide attempts seems less remarkable. In addition, it has recently been recognised that persons who are anxious are particularly likely to develop a 'jitteriness' syndrome when treated with antidepressants.

Power and Cowan have argued that such adverse effects may be related to a serotonin behavioural syndrome. This can be blocked by propranolol, a -adrenoceptor blocking agent that also has 5-HT1A antagonistic properties. As a consequence, propranolol might be particularly effective at alleviating any restlessness/akathisia caused by SSRIs. Indeed, Rothschild and Locke reported that propranolol appeared to relieve the akathisia/suicidal ideation precipitated by re-exposure to fluoxetine.

If suicidal or homicidal ideation develops in patients taking SSRIs or antipsychotics as a consequence of the development of akathisia or panic and aggravated by a process of misattribution or other cognitive process, such states would not be directly drug-induced. Whether or not such states occur is still a matter that needs to be determined conclusively. Nevertheless, there appears to an increasing consensus that these states arise via some secondary indirect process rather than directly. As such, it is likely that the burden of responsibility for drug-induced injury, should any cases be contested, is likely to shift from the makers of SSRIs to the prescribers and to issues of clinical practice.

The induction of significant akathisia or panic during antidepressant treatment is infrequent and unpredictable. Furthermore, there is a range of variables that might confound the interpretation of causation. Therefore, large scale clinical trials are unlikely to settle the question of whether and how often such reactions occur. A more promising line of investigation might be to investigate generic factors in influencing pharmacokinetic and behavioural responses to psychotropic drugs.

Brunner and colleagues have recently identified a point mutation in a structural gene for monoamine oxidase A. This mutation is associated with an increased incidence of impulsive aggression, arson, attempted rape and exhibitionism. Thus, it might be possible to investigate those individuals who have significantly unusual responses to psychotropic agents for a range of mutant genes. If the genes were found to be abnormal, family members could be invited to undergo a double-blind challenge with the psychotropic drug of particular concern in their relative. Such a method might lead to conclusive findings relatively rapidly, and might in due course lead to the establishment of diagnostic tests to determine whether there are particular psychotropic drugs that certain individuals should avoid.

7. Conclusions and Management Recommendations

In the opinion of this author, the volume of case reports and other studies is sufficient to demonstrate that antidepressants and antipsychotics may induce suicidal ideation in certain individuals under certain conditions. These reports must be set against a general recognition that the peak time for suicide attempts is shortly after individuals start treatment with a new antidepressant or antipsychotic. Therefore, it seems a reasonable proposition that, in some patients, taking a new psychotropic agent may lead to the emergence of suicidal ideation.

The fluoxetine controversy, for a number of reasons, has served to highlight this issue. Some of the case reports regarding fluoxetine-induced suicidal ideation do appear sufficiently well documented and detailed to sustain an argument that fluoxetine may lead to the emergence of suicidal ideation. However, this effect is uncommon.

Suicidal ideation associated with fluoxetine is probably not a direct effect of fluoxetine, but is mediated through the induction of akathisia/agitation/panic. This may occur through changes in the 5-HT1A or 5-HT2 receptors. The effects may be successfully blocked by propranolol.

Inappropriate reactions to akathisia, agitation or panic may also play a significant part in any subjective distress that emerges. Any deleterious effects of such reactions will be prevented by a clear awareness on the part of the therapist and patient that such reactions may develop. Patients should be instructed to report any such effects promptly, so that potential misattributions by the patient may be dealt with.

Such reactions to fluoxetine are rare, and undue concern over them may lead to more suicides by virtue of patients giving up treatment. It is this author's opinion that acknowledging that such reactions may occur with all psychotropic drugs, and developing management guidelines for such reactions, is essential for all parties involved in therapy.

Table I. Major criticisms of the data provided by Teicher et al. concerning fluoxetine-induced suicide ideation


No patients had major depression

Some patients had a diagnosis of borderline personality disorder (a condition associated with prominent suicidal ideation and impulsive behaviour)

No patients were treated with fluoxetine alone

No patients were treated with fluoxetine at currently recommended dosages

All patients were or had recently been receiving other medications

All patients had a history of suicide ideation or attempted suicide



Table II. Criteria required for a case report of a drug-induced adverse effect to be considered reliable.


* There should be a close temporal relationship between administration of the drug and occurrence of the adverse effect

* The adverse effect should not be a manifestation of the underlying illness

* The individual should not be taking other medications

* The effect should be clearly documented

* Discontinuation of the drug should lead to cessation of the adverse effect

* The adverse effect should reemerge on readministration of the compound