Concerns regarding treatment-emergent suicidal ideation and behaviour have resulted in some restrictions being placed upon the use of some antidepressants in children and adolescents. In 2004, following analyses of data from clinical trials to assess the risk of suicide in patients receiving various antidepressants,[1,2] the US FDA and the UK Medicines and Healthcare Products Regulatory Agency Committee on Safety of Medicines (CSM) issued warnings regarding the use of paroxetine, sertraline, citalopram and venlafaxine in children and adolescents.[3–5] The concerns prompted reviews of safety data for the newer antidepressants in adult use, in order to detect any increased risk of treatment-emergent suicidal ideation and behaviour in adult patient groups.[1,6–8] Treatment-emergent suicidal ideation and suicide-related behaviour have been postulated with other drugs that affect the CNS, and also with some drugs that are not primarily neuropharmaceuticals.

As a consequence, the assessment of suicidality in both pre- and postmarketing clinical trials of neuropharmaceuticals has become an issue. Prevention of suicide, particularly suicide in young people, is a health priority in a number of countries;[9–11] hence, regulatory agencies would be particularly concerned at the prospect of treatment-emergent suicidal ideation. The prospect of black box warnings or of having to withdraw drugs from the market would be of great concern to pharmaceutical developers, given the considerable investment involved in bringing new drugs to market. However, experience with the assessment of suicidality in relation to drugs has highlighted a number of serious methodological difficulties in investigating treatment-emergent suicidal ideation and suicide-related behaviour.

The aim of the present article is to identify reports of associations between drug exposure and suicidal ideation, suicide-related behaviour and completed suicide, and then to review the manner in which such relationships have been determined.

To achieve this aim, literature searches of PubMed to 31 December 2006 were performed to identify published articles dealing with the relationship of drug exposure to suicidal ideation, suicide-related behaviour and suicide. The search terms used were: (suicide OR suicidality OR suicidal-ideation OR suicide-related-behaviour OR self-harm) AND (drug or medicine) AND (clinical trial). The bibliographies of articles dealing with the topic were searched to identify further studies that had not been identified by the literature searches. Internet searches were performed for government agency reports using Google and specific searches of the FDA, CSM and European Medicines Evaluation Agency (EMEA) websites. Specific PubMed searches were also performed for drugs mentioned in government agency reports. Reports of associations between drugs and suicidal ideation, suicide-related behaviour or suicide were included in the review.

1. Drugs Linked to Suicidality

1.1 Antidepressants

Meta-analyses of randomised controlled trials have demonstrated that compared with placebo, SSRIs and other newer antidepressant drugs are associated with treatment-emergent suicidal ideation and suicide-related behaviour in both children and adults,[2,3,6,7,12–14] but the evidence is more convincing for children and adolescents than for adults. In children and adolescents, SSRIs as a group, compared with placebo, have an increased risk of treatment-emergent suicide-related events, with an odds ratio (OR) of around 1.7.[12,13] However, there were no completed suicides in children and adolescents in any of the clinical trials reported to date.

There have been no assessments, in either children or adults, of whether the tricyclic antidepressants (TCAs) or the monoamine oxidase inhibitors are associated with an increased risk of suicidal ideation, suicide-related behaviour or suicide.

The strength of the association with treatment-emergent suicidal ideation and suicide-related behaviour in children and adolescents varies between the individual SSRIs and the newer antidepressant drugs. In clinical trials of paroxetine in children and adolescents with major depressive disorder, there was an increased risk, compared with placebo, of treatment-emergent suicide-related events.[3,15–17] In the FDA analysis, which included fluoxetine, paroxetine, sertraline, mirtazapine, citalopram and venlafaxine, venlafaxine had the greatest increase in the risk of definitive suicidal behaviour or ideation compared with placebo, with an OR of 8.84 (95% CI 1.12, 69.51) in patients with major depressive disorder and 4.97 (95% CI 1.09, 22.72) for all indications.[2,12] For fluoxetine, although reported meta-analyses and individual clinical trials do not support an increase in suicide-related behaviour, TADS (Treatment for Adolescents with Depression Study) indicated an increased risk of self-harm but not of suicidal ideation.[2,3,17–21] Citalopram, fluvoxamine, sertraline, mirtazapine and escitalopram did not have statistically significant increases in suicidal behaviour or ideation compared with placebo, but their lack of demonstrated efficacy in the paediatric population has resulted in cautions against their use in that population.[3,17]

In adults, meta-analyses of controlled trials indicate an increase in suicide-related behaviour with SSRI treatment.[6,7] This increase was demonstrated in comparison with placebo, but compared with TCAs there was no significant difference in risk.[6] Analysis of individual SSRIs did not demonstrate an increase in the risk of suicide-related events for fluoxetine, compared with either placebo or active treatment.[1] The data for paroxetine are conflicting, with analysis of different subsets of clinical trials indicating either an increase in suicide attempt or a decrease in suicidal ideation, but overall no increase in risk compared with placebo.[1,22,23] Although not individually statistically significant, citalopram, escitalopram, fluvoxamine, mirtazapine and sertraline all had increased ORs for suicide-related events in the CSM analysis.[1] Although SSRIs have the same mode of action, they have differing pharmacokinetic characteristics and it is debatable whether these drugs should be included together in a pooled analysis.[24]

However, the rate of completed suicide has not been demonstrated to be increased in adult patients treated with SSRIs.[8] In an analysis of 48 277 patients included in controlled trials of SSRIs, where 77 patients completed suicide, there was no significant difference between the rates of suicide in the SSRI-treated group, in those treated with other antidepressants and in those treated with placebo.[8] The OR for SSRIs of completed suicide has been estimated as 0.95 (95% CI 0.24, 3.78) compared with placebo, 1.08 (95% CI 0.28, 4.09) compared with TCAs and 0.59 (95% CI 0.16, 2.24) compared with therapeutic interventions other than TCAs.[6,25] The number of completed suicides during industry-sponsored randomised controlled trials is small, given that the CSM review examined data that included 16 completed suicides for 40 826 patients included in the trials under review.[1] This is reflected in the wide confidence intervals.

1.2 Antiepileptic Drugs

The relationship between antiepileptic drugs (AEDs) and suicide-related behaviour is controversial and is currently undergoing review by the FDA.[26] The earliest reports of an AED being associated with suicide-related behaviour were for phenobarbital — an observational study found the rates of depression and suicidal ideation in patients treated with this drug to be higher than those observed in patients treated with carbamazepine.[27,28]

Patients with epilepsy have a higher rate of depression and suicidality than the general population, particularly in the adolescent age group.[29,30] In particular, complex partial seizures involving the temporal and frontal lobes have been associated with depression. In a meta-analysis of controlled trials of vigabatrin, the OR for depression was increased [3.7 (95% CI 1.9, 7.3)] and, although not statistically significant, four suicide attempts occurred in the vigabatrin-treated groups compared with none in the placebo groups.[31] A systematic review of clinical trials found that there is a higher rate of ‘behavioural events’ in patients treated with levetiracetam for epilepsy than in those treated with placebo.[32] Although the behavioural events included intended overdose and attempted suicide, the incidence of these events was <1% and there was no statistically significant difference between the treatment groups. A spontaneous report of a patient treated with topiramate for migraine who developed depression and suicidal ideation resulted in a watching brief on the drug being issued by the New Zealand Centre for Adverse Reactions Monitoring, and in the prescribing information for the drug being updated.[33]

Hence, although there are scant data linking AEDs with treatment-emergent suicidal ideation and suicide-related behaviour, sufficient data are available to justify a watching brief.

1.3 Atomoxetine

Atomoxetine is a nonstimulant medication used for the treatment of attention-deficit hyperactivity disorder (ADHD) in children and adolescents.[34] Its exact mechanism of action is unknown but it is thought to inhibit noradrenergic transporters.[34] It has been demonstrated to be effective, in comparison with placebo, in the treatment of ADHD.[35,36] However, in 2005, both the FDA and the New Zealand Medicines and Medical Devices Safety Authority (Medsafe) issued a caution to be placed on atomoxetine and its use in children and adolescents.[37] The caution was issued in response to the sponsor’s notification of a pooled analysis of the results of 12 trials of 6–18 weeks’ duration. The pooled analysis included data for 2208 children and adolescents, and it found that 5 of 1357 had suicidal thoughts and one had suicidal behaviour compared with none of the 851 individuals taking placebo.[37] There were no completed suicides reported in the studies.

These data have been interpreted as indicating an increased risk of treatment-emergent suicidal ideation and suicide-related behaviour during early treatment, with a resultant black box warning. It is not known whether the risk increases or diminishes with long-term treatment, in a condition that usually requires a treatment duration of many years.[38]

1.4 Isotretinoin

The association of isotretinoin with depression and suicidal ideation is controversial.[39] Isotretinoin is a vitamin A analogue used in the treatment of acne. Over an 18-year period, the FDA received reports of depression, suicidal ideation, suicide attempts and completed suicide for 431 patients exposed to isotretinoin.[40] These data included 37 patients who completed suicide, the majority (62%) of whom had a previous psychiatric history or other possible contributing factors. Dechallenge and rechallenge supported isotretinoin as a causative agent for depression in 25 patients.[40] Hence, applying the Naranjo criteria[41] could support isotretinoin as a causative agent for depression. However, analysis of prescribing data has not demonstrated an increase in the relative risk of depression, suicide or attempted suicide in patients prescribed isotretinoin compared with that in patients treated with antibacterials for acne.[42] Similarly, an increase in the rate of antidepressant prescription in patients who had previously commenced isotretinoin has not been demonstrated.[43]

A physiological or pharmacological mechanism for isotretinoin exposure leading to depression and suicidal ideation has not been elucidated. A recent paper has suggested that suicidal ideation is common among patients with dermatological conditions.[44] Acne occurs primarily in adolescents, a population with a higher incidence of suicide-related behaviour, and may in itself be associated with disturbances in mood. Hence, confounding by indication may also occur.

1.5 Interferon-α

An association between suicidal ideation and treatment with interferon-α2b has been postulated, based upon case reports and surveys.[45] In a cohort study of 42 patients with hepatitis C who were treated with interferon-α2b, 18 (43%) had suicidal ideation, including 11 with treatment-emergent suicidal ideation.[45]

A postulated mechanism for depression and suicidal ideation in patients treated with interferon-α2b is through modulation of neurotransmitters;[46] however, interferon-α2b is used to treat chronic medical conditions that may in themselves be risk factors for depression and suicidal ideation.[47]

1.6 Ethanol and Substance Misuse

Although suicidal ideation is prevalent in patients with alcohol dependence, it is not clear whether the alcohol exposure precedes the suicidal ideation or whether both are symptomatic of another disorder.[48] Early-onset substance misuse also appears to correlate with suicidal ideation, but this most likely represents a common aetiology rather than one being causal for the other.[49] Overall, there is a high prevalence of suicidal ideation in patients with substance misuse disorders, with 63% of women and 47% of men reporting suicidal ideation or suicide related behaviour.[50,51] The interaction between substance misuse, psychiatric co-morbidity and suicidal ideation is intricate and could easily result in confounding.

1.7 Other Drugs

β-Adrenoceptor antagonists (β-blockers) have previously been associated with treatment-emergent depression;[52] however, this has not been supported by level 1 evidence. In a meta-analysis of 15 clinical trials involving more than 35 000 subjects, β-blockers did not have a significantly greater risk of depressive symptoms than placebo or active treatment.[52]

Although lipid lowering medications have been suggested to increase the risk of depression and suicidal behaviour, a meta-analysis of available clinical trials has not demonstrated an association.[53,54]

2. Methodological Issues in Assessing the Influence of Drug Therapy on Suicidal Ideation and Suicide-Related Behaviour

2.1 Overall Findings

Concerns regarding treatment-emergent suicide-related behaviour and suicidal ideation have initially been raised for a number of drugs by observational studies, and these concerns have either been confirmed or refuted by better designed studies. Observational studies are useful for signal detection, but experimental study designs may be required for the determination of causality. This is particularly so when the outcome of interest is independently associated with the indication for treatment.

A further issue in study design is that the sample size requirements for experimental studies in the field preclude the use of completed suicide as an outcome measure. Methodological issues also include the definitions used for suicide-related behaviour and suicidal ideation, and the methods for detecting and quantifying these outcomes.

These issues will be addressed individually in sections 2.2 to 2.6.

2.2 Confounding by Indication

Confounding occurs when factors independently associated with the outcome are also associated with the exposure.[55] In observation research when potential confounders cannot be measured, the data analysis cannot correct for potential bias; hence, where suicidal ideation, self-harm or suicide are related to the indication for treatment, biases may occur in the estimation of effect size, or may result in false associations or the masking of associations.[56] The risk of suicide attempt appears to be increased by chronic disease states, even when adjusted for depression and alcohol abuse;[47,57] hence, the role of observational studies is to generate rather than to test hypotheses. Experimental study data, such as that from randomised controlled trials, are required to overcome confounding by indication.

2.3 Sample Size Requirements for Studies Using Suicide and Suicide-Related Behaviours as Outcome Measures

Suicide is a relatively rare outcome. Worldwide, the estimated age-adjusted suicide rate is 15.5/100 000 persons per year, with a rate of 18.6/100 000 per year for males and a rate of 12.7/100 000 per year for females.[58] Although rates of suicide are higher in patients with co-morbidities, such as chronic disease states and psychiatric disorders, even in high-risk conditions suicide is uncommon.[30,47,57] This has implications for sample size requirements in studies that would seek to use suicide as an outcome measure. For example, it has been estimated that to detect a 20% decrease in the risk of suicide (with 80% power and an α of 0.05) would require 1.9 million participants, whereas to detect a 20% decrease in the risk of nonfatal self-harm would require 220 000 participants.[7] Hence, the sample size requirements preclude the use of suicide as an outcome measure, and the use of nonfatal self-harm as an outcome measure will likely require meta-analysis of randomised controlled trials.

2.4 Effects of Age Groups/Incidence/Prevalence Upon Study Design and Interpretation

The epidemiology of suicide and suicide-related behaviour may impact upon both the design and interpretation of studies.

There are differences in age- and gender-specific rates for suicide and suicide-related behaviour. Overall, the rate of presentation with self-harm is around 300/100 000 per year for males and 400/100 000 per year for females.[59] In adolescents and young adults, the presentation rates are higher (800/100 000 per year for females and 500/100 000 per year for males) than in the older age groups, e.g. 100/100 000 per year for persons >55 years of age.[59] For suicide, there are different effects for age and gender than for self-harm. Suicide rates are consistently higher in males than females, but the effect of age varies between countries, suggesting that region-specific effects may interact differently with age.[60]

The risk of suicide and self-harm may vary between patient groups and indications for treatment. For example, in patients treated with antidepressants, the risk of suicide is 1 in 3000 acute treatment episodes, and the risk of serious suicide attempt is 1 in 1000 acute treatment episodes.[61]

Hence, the age, gender and disease profiles of treatment groups need to be considered when interpreting individual studies and when pooling data from different studies for analysis.

2.5 Definitions of Suicidality

There is lack of consistency of terminology in relation to suicidality.[62] Terms such as ‘suicide-related behaviour’, ‘suicidal behaviour’, ‘self-harm’, ‘attempted suicide’, ‘suicide gesture’, ‘suicide plans’, ‘suicide ideation’ and ‘suicide’ have been used, with varying definitions, in the literature. The resulting ambiguity creates problems with the conduct and interpretation of meta-analyses.

In the FDA analyses of suicide-related behaviour and antidepressants, methodology was developed to enable comparability between studies.[2] The FDA used a standardised assessment process, where individual patient data were used and categorised by a panel of expert reviewers using a coding scheme. The reviewers also underwent training in order to calibrate their responses. However, this methodology does not overcome the problems of data capture, whereby there is variability between trials in the notification and recording of suicide-related behaviour and ideation. Such meta-analyses, using standardised classification schemas, can only be performed where there is access to individual patient data.

Suicide-related behaviour and suicidal ideation are not strongly related to completed suicide. Hence, an association of a drug with suicide-related behaviour and suicidal ideation should not be interpreted as an association with completed suicide. While suicide is a rare outcome, suicidal ideation is common and most people with suicidal ideation will not commit suicide. In a general population, the prevalence of suicidal ideation is around 3%.[63] The corresponding prevalences of suicidal plans and suicidal attempts are 1% and 0.5%, respectively.[63] The presence of a plan for committing suicide appears to increase the risk of self-harm. One in five people with suicidal ideation had formulated a plan for committing suicide in the preceding year, and of those who had formulated a plan, half had proceeded to either a ‘suicide gesture’ or a ‘suicide attempt’.[63] In the absence of a plan, 9% progressed to either ‘suicide gesture’ or ‘suicide attempt’.[63] It is not clear whether these patients represent different conditions or different severities of the same condition.

The progression from suicidal ideation to self-harm and then to suicide is by no means absolute. Self-harm may escalate and this may be a marker for subsequent suicide;[64] however, of those patients who present to hospital with self-harm, around 7% will have completed suicide over a 9-year period of follow-up.[65] The suicide rate appears to be higher amongst those patients who abscond from medical care or who state their self-poisoning to be accidental.[66] In a group of patients presenting after suicide attempt, those patients who took precautions against discovery had a higher risk of subsequent completed suicide.[67] This implies that there is a group of patients who disguise their suicidal intent.

Hence, suicidal ideation and behaviours should be analysed as a separate outcome to suicide.

2.6 Detection and Quantification of Suicidality

None of the randomised controlled trials discussed in sections 1.1–1.7 were specifically designed with suicide-related behaviour, suicidal ideation or suicide as primary outcome measures. These outcomes were included in the safety analyses as adverse events and some studies assessed these behaviours using post hoc analysis of individual patient data. In addition, there is no general agreement on the analysis of events that occur after the treatment period, after a patient has dropped out of a study or in a lead-in phase of a clinical trial. Study design would be improved by the inclusion of appropriate a priori measures for suicide-related behaviour and suicidal ideation.

Clinical trials of drugs in nonpsychiatric conditions, such as acne or even epilepsy, are not well designed to investigate psychiatric endpoints. The outcome measures of these trials naturally concentrate on physical measures, such as severity of acne or frequency of seizures. Although self-harm and suicide would ordinarily be recorded as serious adverse events, suicidal ideation would not normally be captured. In addition, important information about episodes of suicide-related behaviour may not be obtained and recorded. Hence, measures to detect and evaluate suicidal ideation and suicide-related behaviour will need to be developed and incorporated into clinical trial design.

A number of questionnaires have been developed that, either primarily or secondarily, detect and measure suicidal ideation. Examples of these include the Inventory of Suicide Orientation-30 (ISO-30), the InterSePT Scale for Suicidal Thinking (ISST), the Clinical Global Impression of Severity of Suicidality (CGI-SS), the Suicide Ideation Questionnaire and the General Health Questionaire-28 (GHQ-28).[68–71] These tools vary in the number of dimensions of suicidal ideation that are addressed and in their ability to measure the intensity of suicidal ideation. In the case of the ISO-30, five belief dimensions are each measured using six questions, each on a 4-point scale.[69] Such detail may be more appropriate for the specific study of suicidal behaviour rather than for use in clinical trials for indications unrelated to suicide. The inclusion of questions designed to detect suicidal ideation within general or disease-specific quality of life tools is a more practical approach;[68,72] hence, other tools have been validated for measuring suicidal ideation that can be incorporated into clinical trial design. For example, the CGI-SS (a 5-point scale) and the ISST (a 12-item scale) were developed for use in a clinical trial comparing clozapine and olanzapine in the prevention of suicide.[73] The ISST has good inter-observer reliability and internal validity in a population of patients with schizophrenia and schizoaffective disorder, but the validity of the ISST may not extend to patients with other psychiatric diagnoses or to the general patient population.[71] In the InterSePT study, neither the CGI-SS nor the ISST were better predictors of suicide behaviours than features of the patients’ clinical history, such as the number of hospitalisations to prevent suicide in the preceding 36 months, number of previous suicide attempts and substance misuse;[74] hence, although these instruments may be valid in detecting and measuring suicidal ideation, this does not translate into a utility in predicting suicide-related behaviour or suicide.

As mentioned in section 2.5, the link between suicidal intent and the endpoints of self-harm and suicide are tenuous. The Suicide Intent Scale is a measure of the degree of suicidal intention associated with a specific act of self-harm.[75] However, the positive predictive value of the Suicide Intent Scale for suicide (in patients presenting with self-harm with a mean follow-up of 5.2 years) is only around 4%, and its sensitivity and specificity have been reported as 76.7% and 48.8%, respectively.[76] Even in a high-risk group of males presenting with suicide attempt, the Suicide Intent Scale did not predict subsequent completed suicide.[77] The Suicide Intent Scale may be more valid in predicting repetition of self-harm.[75]

Whilst the CSF level of 5-hydroxyindoleacetic acid may be predictive of completed suicide in a high-risk group of males presenting with self-harm, this test is not practical for a clinical trial, particularly for an unrelated indication.[77]

Measurements of lethality and intent, although correlated, are not the same dimension and do not contribute significantly to the risk of repetition;[78] hence, suicidal ideation is not a suitable surrogate outcome measure for self-harm or suicide. Detection of these endpoints should also be incorporated into clinical trial design.

In addition to recording the presence of self-harm, it is also important to measure and record the seriousness, intent and planning of the self-harm. These measures are used in the interpretation of interventions for self-harm and are also useful in the interpretation of risk factors.[79] Factorial analysis of the Suicide Intent Scale has indicated that both intent and planning, in relation to an episode of self-harm, may be measured.[80] The Suicide Intent Scale may also be interpreted in dimensions of expectancies and attitudes, premeditation, precautions against intervention and oral communication.[81] The medical seriousness of a suicide attempt does not necessarily reflect the degree of intent; instead the patients’ expectations of lethality correlate better with intent.[82] The measurement of seriousness requires a clinical assessment of the episode, with a judgement of the likelihood of death or injury resulting from the episode. This may require standardised assessment by a panel, such as in the FDA reviews.[2] Reaction to survival could also be measured following a suicide attempt, as an indicator of severity.[83]

3. Conclusion

The demonstration of an association between a drug and treatment-emergent suicidal ideation and suicide-related behaviour has depended upon randomised controlled trials and meta-analyses. Observational studies have been useful in generating hypotheses, but because of methodological flaws, such as confounding by indication, such studies are inadequate in establishing causality.

In the case of antidepressants, available data from randomised controlled trials and meta-analyses indicate that SSRIs and newer antidepressant drugs, with the exception of fluoxetine, result in increased suicidal ideation and suicide-related behaviour. An association with increased risk of completed suicide has not been established for SSRIs, either because of inadequate sample size or because suicide is a different outcome to suicidal ideation and suicide-related behaviour. Causal associations have not been established between AEDs, isotretinoin, interferon-α and substance abuse and suicidal ideation, suicide-related behaviour or suicide, despite observational studies proposing associations.

To facilitate the comparison of studies and pooling of results in meta-analyses, there is a need for uniform requirements for the detection and recording of suicidal ideation, suicide-related behaviour and suicide in all randomised controlled trials for all treatments.