Every year, nearly one million people die from suicide, including nearly 58,000 within the European Union. Suicide is among the top 20 leading causes of death globally for all ages, and in 15–39-year-olds, suicide is the second leading cause of death after accidents (WHO 2014).

It has been estimated that 90 % of completed suicides occur in people with diagnosable psychiatric disorders (Cavanagh et al. 2003). Bipolar disorder and severe recurrent major depression have the highest rates of suicide of any psychiatric condition (Harris & Barraclough 1997). For patients with affective disorders, the lifetime risk of committing suicide is estimated to be between 6 % and 10 % (Harris & Barraclough 1997) or even up to 19 % (Latalova et al. 2014). This is ten times higher than in the nonpsychiatric population. In an analysis of 28 studies involving 823 suicides among 21,484 patients with bipolar disorder at risk for an average of about 10 years, Tondo and colleagues showed that the pooled, weighted mean annual incidence of suicide was 0.39 % (390/100,000), approximately 28 times higher than the international rate for suicide in the general population (which is about 14/100,000/year, i.e. 0.014 %) (Tondo et al. 2003). The higher risk of suicide in patients with bipolar disorder is also supported by a review of the medical records of 2,826 patients with any major mood disorder according to DSM-IV, recruited in Sardinia (Italy) (Tondo and Baldessarini 2007). The authors found on average a greater risk of suicide among patients with bipolar rather than unipolar disorder. Among the 843 bipolar patients, the annual risk of suicide was 150/100,000, more than 15 times higher than in the general population and three times greater than in patients with recurrent major depressive disorder.

Suicide attempts, defined to be any “intentional act of self-inflicted poisoning, injury or self-harm which may or may not have a fatal intent or outcome” (WHO 2014 pg. 12), are much more difficult to quantify than completed suicides. Rates of suicide attempts are however also known to be higher among people diagnosed with bipolar disorder than in the general population. It has been estimated that 25–50 % of bipolar disorder sufferers will at some point make an attempt on their lives (Latalova et al. 2014). Lethality of suicide attempts varies by a number of factors, for instance, age—the ratio of suicide attempts to suicide death in youth is estimated to be approximately 25:1, compared to about 4:1 in the elderly (www.​afsp.​org/​understanding-suicide/​facts-and-figures). A history of attempted suicide is reported to be the biggest predictor of later completing suicide (WHO 2014). Moreover, lethality is suggested to be increased in some psychiatric populations; for example, in a study by Baldessarini and colleagues published in 2006, the ratio of attempted suicides versus completed suicides (an index often used to assess lethality of suicidal behaviour) was 8.6 to 1 in bipolar patients and 9.6 to 1 in patients with major depressive disorder (Baldessarini et al. 2006), which, if compared to an estimated risk ratio of 20–30 to 1 in the general population (Kessler et al. 2005), indicates higher lethality of suicide attempts among mood disorder patients. This suggests that patients with bipolar disorder are both more likely to attempt suicide and more likely to complete. Bipolar sufferers who are male, or who have a first-degree family member who has completed suicide, are thought to be at increased risk of suicide (Schaffer et al. 2015). The rate of completed suicides has been reported to be as high as 26 % in men admitted to psychiatric hospital with bipolar disorder and a history of deliberate self-harm (Nordentoft et al. 2011).

Notwithstanding the known risk of suicide among those suffering from mental health problems, a recent retrospective study reported that 60 % of suicide completers were not receiving mental health treatment at the time of their suicide, despite a previous psychiatric diagnosis and history of suicide attempts (Bakst et al. 2014). Such findings demonstrate a real need to increase the awareness of risk profiles for suicide and to use effective treatments to reduce suicide rates.

Suicide and suicidal behaviour have serious social and economic implications. Reducing suicidal ideation and activity is an essential starting point in the recovery of people suffering from affective disorders. The prevention of suicide and treatment of suicidal behaviour in patients with mood disorders has therefore become the focus of therapeutic activities at the individual and the social level. Establishing effective treatments that can reduce the risk of suicide (or possibly prevent it) is nowadays one of the main priorities for the health system (http://​www.​nhs.​uk/​Conditions/​Suicide/​Pages/​Prevention.​aspx) and for society in general (http://​www.​nrls.​npsa.​nhs.​uk/​resources/​?​entryid45=​65297). There are, however, currently very few treatments that effectively prevent suicide, and the treatment of patients with suicidal behaviour is one of the most challenging tasks for healthcare professionals. The high mortality, morbidity and costs related to attempted suicide mean that the development of treatment and prevention strategies for suicidal behaviour has been the focus for much of the research on suicidality. In their recent report on suicide prevention, the World Health Organization (WHO) has urgently called for more research into effective therapeutic strategies for treating suicidal behaviour (WHO 2014).

Medication currently plays a relatively minor role in most suicide-prevention strategies (Department of Health 2002; US Department of Health Human Services 2001), although its place may have been underestimated (Saunders and Hawton 2009). Interestingly, however, since the early 1970s, a series of international studies (mainly observational) have suggested the anti-suicidal effect of lithium.

One of the earliest suggestions of lithium’s effect on suicidal behaviour was reported in 1972 (Barraclough 1972). In a subsequent report of his retrospective investigation of 100 suicide cases, Barraclough showed that a high proportion of the suicides had occurred in patients with long histories of treatment for psychiatric illness and that 40 % of the sample had previous episodes of affective disorder. Reviewing these affective disorder cases individually and looking at the chronology of illness and treatment regimes, he suggested that 25 % of these cases had been prescribed ineffective treatments at the time of their death. He reported that lithium treatment may have prevented relapse in at least half of these cases and concluded that, with the use of lithium, one fifth of the 100 cases of suicide may have been prevented (Barraclough et al. 1974).

More recent examples of observational reports of lithium’s anti-suicidal effects come from a study based on data collected by a health maintenance organisation in the USA, using a sample of 20,638 patients with diagnosed bipolar disorder. The report identified a 2.7 times greater risk of suicide for patients prescribed valproate compared to those prescribed lithium (Goodwin et al. 2003). A similar study by Collins and McFarland (2008) found comparable results in a sample of 12,662 patients with a diagnosis of bipolar disorder. By the end of the study, there had been 11 completed suicides and 79 reported suicide attempts. Results suggested lithium to be superior in reducing rates of both suicide attempts and completed suicides, in comparison to the other anticonvulsants studied (divalproex, gabapentin and carbamazepine) (Collins and McFarland 2008). Consistent findings continue to be reported in the literature; for instance, in a 2014 report of post-hospitalization suicide rates in Finland, Isometsä and colleagues (2014) reported that of three types of pharmacotherapy investigated (lithium, valproate and antidepressants), only lithium had a significant effect in reducing the risk of suicide. Also, in 2015, similar findings were reported in a study following, for 3.5 years, a naturalist cohort of 826 bipolar patients who had previously been hospitalized as a result of a suicide attempt (Toffol et al. 2015). The study authors explored the effect of treatment with a variety of drugs on suicide and all-cause mortality in the sample. Results suggested that several medications were associated with an increase in suicidal attempts: valproic acid (relative risk (RR) = 1.53, 95 % confidence interval (CI) 1.26–1.85, p = 0.001), antidepressants (RR 1.49, 95 % CI 1.23–1.80, p = 0.001) and benzodiazepines (RR 1.65, 95 % CI 1.37–1.99, p < 0.001). Lithium, in contrast, was associated with fewer suicide attempts, although this difference was not significant (RR 0.82, 95 % CI 0.65–1.02, p = 0.09). Lithium treatment was however, associated with significantly reduced numbers of completed suicides (RR 0.39, 95 % CI 0.17–0.93, p = 0.03) and decreased all-cause mortality by 49 %. The authors concluded that, among such high-risk bipolar patients, lithium should be considered a suitable treatment to help to prevent suicide.

Trying to make use of all available evidence, meta-analyses have been carried out to combine data and provide pooled estimates from studies from a range of different sources. Baldessarini and Tondo in 2006 analysed data from 31 studies (eight of which were randomized controlled trials) of participants with bipolar disorder and other major mood disorders and found suicide-preventative effects of lithium (Baldessarini et al. 2006). This meta-analysis reported a fivefold reduction in suicidal acts for participants treated with lithium in comparison to those without lithium treatment (RR 4.91, 95 % CI 3.82–6.31, p < 0.0001) and a risk reduction of suicide by approximately 80 % following treatment with lithium. Subsequent studies by the same team of authors found less positive, but still clearly favourable, results for lithium. In a second meta-analysis of six studies, including more than 30,000 patients who were treated with lithium or with an anticonvulsant, Baldessarini and colleagues observed a nearly threefold superiority of lithium over other mood-stabilizing anticonvulsants (carbamazepine, divalproex or lamotrigine) in terms of reducing suicidal behaviour (RR = 2.86, CI: 2.29–3.57, p < 0.0001) (Baldessarini et al., 2009).

Although these studies provide a very positive indication of a link between lithium treatment and reduced rates of suicide, their findings lack proper scientific rigour due to their observational nature. Nonrandomized studies, especially those based on lithium clinic samples, are likely to produce biassed findings. Any beneficial effect might reflect compliance with medication being greater in patients attending a clinic. There may also be selective prescribing to patients at risk. Furthermore, deaths from other causes are often not considered in such studies. Theoretically, given the potential physical side effects of lithium on thyroid and renal function, it is possible that a benefit of lithium in preventing suicide might be offset by an increased risk of death from physical disorders. More rigorous research methods are thus needed to provide more reliable data about lithium’s preventative effect on suicide.

Randomized controlled trials (RCTs) represent the gold standard in research methodology. Just 1 year after Barraclough’s initial observational report, two of the first RCTs investigating the anti-suicidal efficacy of lithium were published in the scientific literature (Prien et al. 1973a, b). In the first of these reports, an RCT with a sample of patients with bipolar disorder (n = 44) and unipolar depression (n = 78), participants were randomly assigned treatment with imipramine, lithium or placebo (Prien et al. 1973a). One suicide was reported in the placebo group. In the second report of a placebo-controlled, randomized control trial, allocating patients with bipolar disorder (n = 208) treatment with either lithium or placebo, the author observed no completed suicides in the lithium-treatment group, in comparison to two completed suicides in the placebo group (Prien et al. 1973b).

The authors concluded that there was a suicide-preventative effect of lithium treatment.

During the 1990s, Greil and colleagues contributed three further RCTs to the literature surrounding lithium’s anti-suicidal effects (1996, 1997a, b). All three of these trials involved a relatively long involvement period, following patients for 2.5 years, as well as having reasonably large sample sizes (81, 144 and 90 participants, respectively). Moreover, the open-label design of these trials allows for a more pragmatic result, producing findings that are potentially more relevant to everyday clinical practice. In the first trial, Greil and colleagues compared lithium treatment to amitriptyline in a sample of 81 participants with unipolar depression. One suicide was reported in the amitriptyline group, in contrast to no completed suicides in the lithium group (Greil et al. 1996). Similar effects were reported in a paper in 1997, comparing lithium treatment to carbamazepine, in patients with bipolar disorder (n = 144) (Greil et al. 1997a). Again, no suicides were observed in the lithium group, in comparison to one completed and one attempted suicide in the carbamazepine group. In a separate study comparing the same treatments (lithium and carbamazepine), but with a sample of participants with schizoaffective disorder (n = 90), the same team reported no suicides in the lithium group, in comparison to four attempted suicides in the carbamazepine group (Greil et al. 1997b).

RCTs comparing suicide rates following treatment with lithium to other mood stabilizers continued to appear in the literature in the following years. In 2000, Bowen published an RCT comparing lithium treatment to valproate and placebo. In a 1-year follow-up, no completed or attempted suicides were reported in any group. In 2003, two randomized controlled trials of lithium in comparison to lamotrigine were published. In an open-label, randomized, placebo-controlled trial of 463 patients with bipolar I disorder, Calabrese and colleagues (2003) compared treatment with lamotrigine, lithium or placebo in monotherapy for a period of 18 months. Here one death was seen in the lamotrigine group, in comparison to no deaths in the placebo group or in the lithium treated group. Similarly, in a double-blind placebo-controlled trial of 175 patients allocated to treatment with lamotrigine, lithium or placebo, the same authors again suggested a suicide-preventative effect of lithium treatment, reporting one suicide attempt in the lamotrigine group, versus none in the lithium group. Additionally, more evidence of suicidal ideation (reported on the Hamilton Depression Rating Scale) was observed in the lamotrigine group in comparison to the lithium group (Bowden et al. 2003).

More recently, in a study of relapse prevention in bipolar disorder, Geddes and colleagues added to these findings by looking at the benefits of using a combination treatment of both lithium and valproate. In this open-label randomized trial, 330 patients were randomly allocated to treatment with lithium, or valproate in monotherapy, or a combination of both lithium and valproate. Participants were followed for a total of 24 months. Findings from this study suggested that using the treatments in combination therapy resulted in the best prophylactic effects. There were two patient deaths from all causes in a lithium-treated group, whilst there were three patient deaths in the valproate group (Geddes et al. 2010).

All these studies suggested a possible effect of lithium in terms of suicide prevention; however, considering that suicide is a rare event, individual studies did not have enough power to assess whether this association was statistically significant. Pooling data through the use of meta-analyses allows for limitations such as low event rates to be minimized. In 2005, the first systematic review investigating evidence on possible anti-suicidal effects of lithium, which included only randomized studies (32 RCTs), was added to the literature (Cipriani et al. 2005). This review combined results from RCTs lasting at least 3 months, in patients with mood disorders, in which there was comparison of lithium versus placebo or versus any active drugs. These included 1,377 patients randomized to lithium and 2,052 to other compounds. Nineteen trials compared lithium with placebo, nine with carbamazepine, three with lamotrigine and one with divalproex. The other trials were comparisons with antidepressants, namely, imipramine, amitriptyline, maprotiline and mianserin (multiple comparisons were included in some trials). Seven trials reported suicide as an outcome and an event occurred in at least one arm of the trial. In these trials 2 out of 503 people prescribed lithium died by suicide, compared with 11 out of 601 patients prescribed placebo or comparator drugs (Peto OR 0.26, 95 % CI 0.09–0.77). Thus there was evidence of a significant reduction of suicide in those prescribed lithium (although this finding was associated with significant heterogeneity). Surprisingly few trials included data on deliberate self-harm/attempted suicide. Therefore, results for these outcomes were combined with those for suicide. There was a lowered risk of fatal or nonfatal suicidal behaviour, this occurring in just 6 out of 670 patients in the lithium-treated groups compared with 18 out of 781 in the groups of patients receiving comparator drugs (Peto odds ratio 0.21, 95 % CI 0.08–0.50), although this result was also associated with significant heterogeneity. Finally, deaths from all causes were examined. There was a much-reduced overall risk of death from any cause in patients treated with lithium than in those in the comparison groups (Peto OR 0.42, 95 % CI 0.21–0.87), corresponding to 9/696 versus 22/788 events, respectively. Although this result was again associated with significant heterogeneity, the apparent beneficial effect of lithium on suicide risk was not offset by an increase in deaths from other causes. In post hoc analysis, there was no significant difference for the above outcomes between the results of trials that compared lithium with placebo and those that compared lithium with active drugs. Whilst the effects of lithium on suicidal behaviour were less in this meta-analysis of RCTs than reported in nonrandomized studies, lithium appeared to reduce the risk of suicide by approximately 60 %. The reduction for suicide and deliberate self-harm/attempted suicide combined was approximately 70 %.

Overall, the reduction in deaths from all causes was of the order of 60 %.

The results of such a meta-analysis offered more compelling evidence of an association between lithium treatment and reduced rates of suicide in patients with mood disorders. A methodological problem however remained in that the evidence of lithium’s anti-suicidal effect in the RCTs discussed came from incidental findings, where suicide was not an outcome measure in the trial’s design (Tondo and Baldessarini 2009). The opportunity to conduct placebo-controlled, randomized control trials to investigate the effects of lithium treatment on suicidality is, however, limited. Clearly, there are ethical implications for designing studies that would offer a placebo arm to participants experiencing suicidal ideation, particularly if one hopes to create a trial investigating suicide as a primary outcome measure. Investigating suicidal behaviours, even as a secondary outcome measure, has been impeded by the fact that suicidal ideations or actions are often exclusion criteria for large-scale clinical trials. Robust research evidence from placebo-controlled RCTs specifically designed to investigate the effects of lithium of suicidality has therefore been more challenging to accumulate than in other treatment areas. For instance, despite the high incidence of suicide and self-harming behaviour among adolescents, there are to date no randomized control trials to investigate the influence of lithium, or in fact any pharmaceutical agents on suicide, in younger populations (Ougrin et al. 2015). There are, however, now a number of RCTs investigating suicide in adults that have overcome these methodological and ethical barriers by comparing lithium to other mood stabilizers, antidepressants and antipsychotic treatments and therefore ensuring that all arms of the study offer treatment to participants. Researchers have called for more trials using this method and suggest that using suicide as a primary outcome measure is ethically feasible if all arms offered to the participants consist of some treatment—for example, antidepressant alone versus antidepressant plus lithium (Guzzetta et al. 2007).

The first of these placebo-controlled RCTs to specifically investigate lithium’s relationship with suicidal behaviour was completed by Lauterbach and colleagues in 2008. This study explored the use of lithium as an adjunctive therapy, over a 1-year treatment period, versus placebo, to reduce suicidal behaviour in patients with depressive disorders (n = 170). The findings of a survival analysis stated that there was no significant difference in suicidal acts, or attempts between the groups, with seven reported attempts in each group (hazard ratio (HR) = 0.517; 95 % CI 0.18–1.43). Post hoc analysis, however, revealed that there was a significant difference in incidence rates of completed suicides (p = 0.049), with three completed suicides in the placebo group, compared to no completed suicides in the lithium-treated group. There were a high number of dropouts, and the total number of suicidal events was just 17, such that these results should be interpreted with caution, but the study does appear to support a role for lithium in protecting those with affective disorders against suicidal acts with a fatal outcome. Khan and colleagues (2011) similarly conducted a randomized controlled trial comparing suicidality in depressed patients, assigning them to 4 weeks of treatment with citalopram and lithium or citalopram and placebo. In this study, a therapeutic dose of lithium (>0.5 mEq/L) was associated with prevention of suicidal thoughts and behaviour.

Oquendo and colleagues (2011) also completed an RCT using this comparator model, with a 2.5-year follow-up trial of patients with bipolar disorder and past history of suicide attempts (n = 98). Patients were randomly assigned to treatment with lithium or valproate, in a double-blind trial. Over the 2.5-year follow-up, 45 suicide events were reported in 35 participants, including 18 suicide attempts (six suicide attempts in a lithium-treatment group and eight suicide attempts in a valproate group). An intent-to-treat analysis suggested there was no significant difference between treatment groups in terms of suicide event or time to attempt.

An additional methodological difficulty in this field is the fact that suicidal behaviour is statistically rare, making it difficult to achieve power in many of the studies, due to limited sample numbers and short follow-up periods. Many of the papers published in this area report findings comparing only one incidence of suicide in the participant group, compared to no incidences in another. This can create challenges, as small differences in count can have a large impact on the estimated effect size. It has been suggested that this low event rate may be reflective of the fact that suicidal patients are often excluded from participating in trials in the first place (Simon et al. 2006). Although Oquendo and colleagues’ (2011) study did not show any significant effects of either treatment on suicidal behaviour, the use of patients with a history of suicide attempts did lead to an increased number of suicide events during the follow-up period than had been seen in previous studies. The authors describe this choice of participants with a higher baseline rate of the outcome of interest as a practical method for overcoming the challenges of studying such a low-frequency event as suicide.