Suicide risk assessment and intervention in people with mental illnessBMJ 2015; 351 doi: https://doi.org/10.1136/bmj.h4978 (Published 09 November 2015) Cite this as: BMJ 2015;351:h4978
- James M Bolton, associate professor1,
- David Gunnell, professor of epidemiology2,
- Gustavo Turecki, professor of psychiatry3
- 1Departments of Psychiatry, Psychology, and Community Health Sciences, University of Manitoba, Winnipeg, MB, R3E 3N4, Canada
- 2School of Social and Community Medicine, University of Bristol, Bristol, UK
- 3Departments of Psychiatry, McGill University, Montreal, QC, Canada
- Correspondence to: J M Bolton
Suicide is the 15th most common cause of death worldwide. Although relatively uncommon in the general population, suicide rates are much higher in people with mental health problems. Clinicians often have to assess and manage suicide risk. Risk assessment is challenging for several reasons, not least because conventional approaches to risk assessment rely on patient self reporting and suicidal patients may wish to conceal their plans. Accurate methods of predicting suicide therefore remain elusive and are actively being studied. Novel approaches to risk assessment have shown promise, including empirically derived tools and implicit association tests. Service provision for suicidal patients is often substandard, particularly at times of highest need, such as after discharge from hospital or the emergency department. Although several drug based and psychotherapy based treatments exist, the best approaches to reducing the risk of suicide are still unclear. Some of the most compelling evidence supports long established treatments such as lithium and cognitive behavioral therapy. Emerging options include ketamine and internet based psychotherapies. This review summarizes the current science in suicide risk assessment and provides an overview of the interventions shown to reduce the risk of suicide, with a focus on the clinical management of people with mental disorders.
Suicide is a major international public health problem, claiming one life every 40 seconds.1 It is the second leading cause of death in people aged 15-29 years and was responsible for 39 million disability adjusted life years in 2012.2 At least six close relatives or friends are bereaved by every suicide, and these people also have an increased risk of depression and suicide.3 4 5 For every death from suicide, 30 people attempt suicide; in the United States, this amounts to one million people each year.6 The economic cost of suicide and self harm is considerable, with estimated annual costs (direct and indirect) of $41bn (£27bn; €36bn) in the US.7 Improved understanding of who is at risk of suicide and the development of interventions to reduce suicide in key high risk groups are priority targets of national research agendas and government suicide prevention strategies.8
Several reviews have examined the prevention and epidemiology of suicide, as well as the risk factors.9 10 11 12 This review will complement these by examining two facets of the clinical care of people at risk of suicide—the assessment of suicide risk and interventions that can reduce that risk. We will focus on people with mental health problems and will not deal with broader public health strategies to prevent suicide. We will begin by describing the epidemiology of suicide and suicidal behavior in clinical populations, highlighting specific patient subgroups that are at higher risk. Risk assessment approaches will be discussed in detail, with a review of specific assessment tools and how the science of risk assessment is evolving. Interventions that have been shown to reduce suicidal behavior or prevent suicide, such as pharmacotherapy, psychotherapy, and follow-up care, will be reviewed. Given the prominence of suicide as a health problem and the demands on clinicians to manage this challenging condition, this review will provide an overview of evidence based assessment and treatment approaches to help guide clinical work with this at risk population.
Sources and selection criteria
This review included material compiled from a comprehensive literature review of PubMed, as well as papers selected from reference lists of the articles obtained. We also performed a broader search with Google to identify relevant articles and guidelines not listed on PubMed. Systematic reviews and meta-analyses were prioritized. Randomized controlled trials were included regardless of size or quality. We limited searches to English language studies published between 1 January 1990 and 1 February 2015. Search terms included suicide, self harm, risk assessment, guidelines, intervention, outreach, follow-up, medication (and specific types), pharmacotherapy, electroconvulsive therapy, psychotherapy (and specific types), and specific terms to identify clinical populations (emergency department, inpatients, discharge). Searches were conducted by the authors and three research associates. We reviewed and summarized the collected articles and guidelines and then determined the most clinically relevant articles that would be included in the review.
Suicide is currently the 15th most common cause of death worldwide. In 2012, 804 000 people died by suicide, accounting for 1.4% of deaths worldwide and an average population rate of 11.4 per 100 000.1 In high income countries suicide rates are around three times higher in males than in females, and key risk factors include previous self harm, depression, alcohol misuse, physical illness, low socioeconomic position, and relationship breakdown.10 11
People with mental health problems have a substantially higher risk of suicide and self harm than that found in the general population (fig 1⇓).13 14 In a 36 year observational follow-up study of the Danish population, the cumulative risk of suicide in people who had inpatient or outpatient clinical contact with specialized mental health services was 4% in men and 2% in women.15 The rate varied across disorders and was highest in people with comorbidity and a history of self harm. This risk varies across different clinical populations, with psychiatric inpatients showing the highest risk of suicide within the next year.13 The risk of suicide seems to be greatest during the first few months after diagnosis across all mental disorders.16 17 18 The risk of suicide is also influenced by treatment factors, which will be described later in the review.
Inpatients and recent discharge
Although admission to hospital is often intended to provide a safe environment for the suicidal patient, the risk of suicide while an inpatient is high. The rate of suicide has been reported at five per 1000 occupied beds each year.19 A meta-analysis of 27 studies reported a rate of 147 suicides per 100 000 inpatient years, with individual studies reporting figures as high as 860 suicides per 100 000 inpatient years.20 Suicide tends to occur early during the course of an admission, with 40% occurring within the first three days.21 A suicide attempt preceding the admission significantly increases the risk of suicide while an inpatient.
The risk of suicide is high in the first week after discharge from a psychiatric hospital admission, remains high for the first few months after discharge, and then slowly decreases (fig 2⇓).22 23 24 In a UK national study of suicide among psychiatric patients, a quarter of suicides occurred within three months of discharge.25 Almost half of patients who die by suicide within three months of discharge die within the first month, often before their first follow-up contact.26 The risk of suicide after discharge is especially high for psychiatric patients who were admitted to hospital with a suicide attempt. In a Swedish observational study, about a fifth of men with schizophrenia or bipolar disorder died by suicide within one year of admission for a suicide attempt.27
The emergency department
The emergency department is a common point of contact between suicidal people and treatment providers. Rates of future suicide among people presenting to the emergency department with self harm are high: 2% of these people will kill themselves within one year, and the five year estimate of suicide is 4%.28 This risk is more than 50 times greater than that seen in the general population and is associated with a 40 year reduction in average life expectancy.29 30 Rates of repeat self harm after contact with the emergency department are 10% at one month and as high as 27% at six months.31 32
Higher suicide risk shortly after clinical contact
Recent discussions have cited the need to improve the prediction of suicide in much shorter time intervals (hours, days, and weeks).33 34 This is especially important in the emergency department and other acute care settings. This line of reasoning corresponds with current dialogue on targeting prevention efforts in high risk periods, particularly after clinical contact.35 This places clinicians in a difficult position because patients are arguably most vulnerable in the phase after the interaction has ended. Future research is needed to develop methods of more accurately assessing suicide risk within hours or days; this will allow informed delivery of treatment services to those at highest risk.35
Suicide risk assessment
The concept of suicide risk assessment is controversial and much debated in suicide research. National guidelines provide recommendations for risk assessment yet there is no widely accepted standard of care. What constitutes a risk assessment is also an important question. While risk assessment is often synonymous with risk assessment tools or scales, at its most basic it represents a clinical encounter where a patient is asked about suicidal thoughts and plans. Although risk assessments are often conducted in emergency departments and specialty mental health settings, many occur in primary care—30% of American adults who die by suicide have seen their primary care provider in the month before suicide.36 Frameworks for approaches to suicide assessment are generally consistent; they suggest collecting information on previous suicidal behavior, current suicidal thoughts and plans, hopelessness, stressors, the presence of mental disorder symptoms, themes of impulsivity and self control, ready access to highly lethal methods (such as firearms), and protective factors.37 38 39
A collaborative, therapeutic alliance between clinician and patient is important when conducting the assessment.40 Even a single mental health assessment in the emergency department has been associated with a reduced risk of repeat suicidal behavior that may be as high as 40% in the short term.41 The Collaborative Assessment and Management of Suicidality (CAMS) is one such therapeutic relationship approach that has been shown to enhance treatment retention and reduce suicidal ideation initially and at one year follow-up.42 This approach involves a range of four to 12 sessions during which the clinician and patient collaboratively engage in a structured assessment of the patient’s suicidal thoughts and treatment planning.
Does asking about suicide make a patient more likely to act on it?
Patients and general practitioners think that patients with depression should be asked about suicidal thoughts, yet many GPs have received no formal training in suicide assessment.43 A further barrier to assessment is the belief held by some clinicians that asking about suicidal thoughts will induce such thoughts in patients. A non-systematic review published in 2014 examined 13 studies published between 2001 and 2013 that investigated this question and found that none reported a significant increase in suicidal ideation in patients who were asked about suicide.44 This review included studies of adults and adolescents, with methodologies ranging from small cohort studies to larger randomized controlled trials (RCTs). Of note, one RCT showed that asking about suicide did not increase distress or suicidal thoughts, in both a general group of students and those with depression or previous self harm.45
Challenges in risk assessment
There are considerable difficulties in obtaining an accurate suicide risk assessment. A fundamental challenge is determining which people determined to be “high risk” will later die by suicide (“true positives”). The positive predictive value (PPV) of any risk tool that assesses an event with a low base rate in the population, such as suicide, is likely to be low even when sensitivity and specificity values are high (see box 1 for definitions of statistical tests).46 As such, many people will be inappropriately labeled “high risk” and provided with resources that they may not have needed (such as inpatient admission). Another challenge is the traditional reliance on subjectively reported information, which can be misleading when assessing the risk of suicide. One study found that almost 80% of people who eventually died by suicide denied suicidal thoughts in their last verbal communication.47 This has prompted the search for alternative measures of assessment, including computer based implicit association tests that are more sensitive than both patient self report and clinician opinion.48
Box 1: Definitions and parameters of statistical tests in relation to suicide prediction
Sensitivity: The proportion of people who later die by suicide who are identified by an assessment tool as being at risk of suicide. Sensitivity is inversely related to the false negative rate
Specificity: The proportion of people who do not die by suicide who are identified as not being at risk of suicide. Specificity is inversely related to the false positive rate
Positive predictive value (PPV): The PPV statistic measures “true positives”: the proportion of people who later die by suicide compared with people who are both correctly and falsely identified by a test as a positive risk for suicide (suicides/positive test results). High false positive rates or a low incidence of suicide (or both) will lower the PPV
Negative predictive value (NPV): The NPV statistic measures “true negatives”: the proportion of people who do not die by suicide compared with people who are both correctly and falsely identified by a test as a negative risk for suicide (non-suicides/negative test results). High false negative rates will lower the NPV. NPV tends to be high given the low incidence of suicide
Shifts in the science of risk assessment scales
Traditional suicide risk scales were not developed using empirical evidence. Numerous scales and tools have been created over several decades, but unfortunately few were tested statistically,49 and several scales have never been tested on their predictive ability for suicidal behavior. Consequently, a systematic review of risk assessment tools in the emergency department found that only 12 studies qualified for inclusion.50 Current research has therefore shifted to statistical derivation and evaluation of risk assessment tools. Another important shift in the science underlying risk assessment scales has been the move from interview dependent tools to interview independent tools. The table⇓ lists risk scales that have been evaluated in their ability to predict suicide together with published psychometrics (specifically sensitivity, specificity, PPVs, negative predictive values (NPVs)).
Conventional risk assessment scales
A conventional approach to the development of suicide risk assessment scales has been to collate likely risk factors for suicide based on concepts of face validity or content validity. The most commonly studied risk scales include the Beck hopelessness scale (BHS), the Beck depression inventory (BDI), the Beck scale for suicide ideation (BSS), the suicide intent scale (SIS), and the SAD PERSONS scale.
Beck hopelessness scale (BHS)
The BHS has been studied in psychiatric inpatients and outpatients, and these studies examined its ability to predict suicide over long follow-up periods (5-20 years).51 52 54 63 BHS scores differentiated those who died by suicide from those who did not, so the BHS was deemed to be a successful prediction tool in some studies; however, specificity was low and one American study reported a PPV of 1%.54 A subsequent meta-analysis of 10 BHS studies, four of which investigated suicide and six non-fatal self harm, found similar results for suicide and self harm—the BHS had a sensitivity of 78-80% but a specificity of 42%.56
Beck depression inventory (BDI) and the Beck scale for suicide ideation (BSS)
The BDI and BSS also performed with low precision in the American study mentioned above, with PPVs of 2% and 3%, respectively,54 for predicting future suicide.
Suicide intent scale (SIS)
A review of the SIS found that it has been assessed in 158 studies.64 Of these, 13 assessed its ability to predict suicide and 17 studied its ability to predict self harm. Regarding suicide prediction, most studies reported no significant associations between scale scores and later suicide, and only two reported predictive statistics.57 58 PPV values were low in both studies (4% and 10%) but higher in the study that performed a secondary analysis on the subset of people aged over 55 years (23%).57 In a subsequent Swedish study that assessed the ability of the SIS to predict suicide over 10-15 years in outpatients with a history of suicide attempts, specificity was 52% and the PPV was 17%.59 Together, these studies illustrate the tendency for scales to have low predictive accuracy, as well as how the duration of follow-up (more time at risk) and more suicidal case mix (higher risk people) can influence PPV.
SAD PERSONS scale
The SAD PERSONS scale is a mnemonic of 10 items where each letter corresponds to a potential risk factor for suicide.65 It is one of the most widely used tools for suicide assessment even though it was never formally tested during its development. In a representative sample of hospitals in England, it was the most commonly used scale.66 A recent systematic review showed that only three studies have examined its ability to predict suicide outcomes, and none of these found it to be predictive.67 The PPV of the SAD PERSONS scale to predict repeat self harm in emergency department patients at six months was 5%, and its sensitivity for predicting suicide at one year follow-up was 23%.60 68
Suicide assessment scale (SUAS) and Karolinska interpersonal violence scale (KIVS)
These two scales have also been used to assess suicidal intent.62 69 The 20 item SUAS assesses a range of emotions, personal characteristics, and suicidal thoughts and behaviors. In the one study that used it to predict suicide over one year, it showed 75% sensitivity and 86% specificity, but the PPV was 19%.61
The KIVS features two subscales that assess exposure to, and expression of, violence in childhood and adulthood. Its use as a risk assessment tool is based on observations that people attempting suicide often have higher levels of aggression and impulsivity. When administered to people who had attempted suicide, both subscales predicted suicide within four years with 80-88% sensitivity but lower specificity (60-65%) and a low PPV (range 7-14%).62
Newer suicide risk scales
The Columbia-suicide severity rating scale (C-SSRS), the suicide trigger scale (STS), and the suicide probability scale (SPS) are based on conventional approaches.
The C-SSRS measures four domains of suicidal thoughts and behavior including the severity of ideation, intensity of ideation, type of suicidal behavior, and lethality of suicide attempts.70 One study has shown that it predicts future suicidal behavior with sensitivity and specificity of 67% and 76%, respectively.71 In this mixed sample of 3776 people (some with major depression and post-traumatic stress disorder and others with epilepsy and fibromyalgia), 201 (5.3%) reported suicidal behavior during the follow-up period. A positive baseline test had a PPV of 14% and NPV of 98% of predicting future suicidal behavior during a mean follow-up of nine weeks.
The STS scale is based on five areas linked to suicide: suicidality, mood symptoms, trauma, impulsivity, and attachment style. In a study of psychiatric inpatients that compared the STS with the C-SSRS and BSS, only the STS was predictive of future suicide attempts within six months.72 This study is one example of a limited number of studies directly comparing performance of risk assessment scales.
Empirically derived tools
Several newer studies have used the more robust methodology of empirically deriving a prediction tool using a development dataset and then testing it in a separate validation dataset.
Manchester self-harm rule
This tool was the end result of statistical selection from a list of 50 candidate variables assessed in almost 10 000 people presenting to the emergency department with self harm.73 The assessment of four clinical variables (with endorsement of one or more being classified as moderate-high risk) yielded high sensitivity (97%) for self harm or suicide within six months. The rate of repeat self harm (including suicide) in this population was 17%. The four variables, which have the advantage of being interview independent, were: history of self harm, previous psychiatric treatment, presents with a benzodiazepine overdose, and current psychiatric treatment. Specificity was low (26%), as was the PPV (22%). This rule was also tested in a Swedish population and predicted repeat self harm with similar psychometric scores.74
ReACT self-harm rule
A subsequent study from the same UK study group and dataset yielded a slightly different predictive model named the ReACT self-harm rule.32 It was based on a larger number of presentations (almost 30 000), using the same list of candidate variables and using a derivation and validation dataset. Statistical based selection of predictor variables identified four: recent self harm, living alone or homeless, cutting as a method of self harm, and treatment for a current psychiatric disorder. High risk classification (based on presence of one or more of these four variables) predicted suicide within six months with high sensitivity (88%) and NPV (almost 100%), but low specificity (24%) and PPV (0.5%).
Repeated episodes of self-harm (RESH) score
This statistically derived tool was developed in Australia using administrative databases.75 The authors investigated the risk of repeat self harm (including suicide) within six months in all people admitted to hospital with a baseline self harm episode in two states. As opposed to the UK studies, which involved emergency department presentations, this study looked only at self harm incidents that involved admission to hospital. Using separate development and validation samples, a tool measuring four variables (number of previous self harm episodes, time between episodes, diagnosis of a mental disorder in the past year, psychiatric admission in the past year) showed fair prediction. Sensitivity was lower than that seen for the UK tools, but at higher scores (>16) it showed high specificity (98%) and PPV (82%).
A statistically derived tool also examining the prediction of future suicide attempt in the emergency setting was developed from a Canadian sample.76 It used a baseline sample of all people who presented to the emergency departments of two hospitals over four years (for self harm and non-self harm) who received a psychiatric consultation. The tool was developed on the basis of the outcome of suicide attempts within six months. Of 16 candidate variables, only two significantly accounted for variance in future suicide attempts: suicidal ideation at presentation and history of psychiatric care or suicide attempts. The presence of at least one of these variables showed high sensitivity (91%) and NPV (99%) but low specificity (24%) and PPV (3%) for future attempts. An important consideration when interpreting the results of assessment tools completed in clinical situations is that the psychometric performance of the tools probably varies according to the availability and effectiveness of local mental health services and the response of clinicians to the assessment results.
Novel methods of risk assessment
Advances in suicide risk assessment include forays into the use of implicit thoughts and neurocognitive functioning. Implicit thoughts of suicide represent an appealing substrate for risk modeling because they overcome the inherent challenge of patients denying (or even being unaware of) their true suicide intent.
The implicit association test (IAT) is an established psychological test that measures a person’s unconscious beliefs on a subject or motivations towards a specific behavior. A computer displays an image (in this case relative to suicidal behavior) or a neutral image, and the subject presses a key to indicate whether or not they view the image as related to self. The reaction time in this task provides the measure of their propensity for suicidal behavior. A cohort study of 157 patients in a psychiatric emergency department showed that a specific death-life IAT test predicted future suicidal behavior within six months, independently of the person’s voiced intention and the clinician’s belief of future suicidal behavior (sensitivity 50%, specificity 81%, PPV 32%, NPV 90%).48 This finding was replicated in a Canadian emergency department cohort study of 107 patients, which showed that the death-life IAT independently predicted suicidal behavior within three months (sensitivity 43%, specificity 79%).77 Of note, the accuracy of the IAT was improved when used with other variables (previous self harm, education level, depression with psychosis, and non-poisoning self harm).
Several neurocognitive tests have been examined for their ability to detect suicidal behavior. A recent meta-analysis showed that only three of these tests significantly correlate with a history of suicidal behavior: the Stroop test, the verbal fluency test, and the Iowa gambling task.78 Conceptually, these results match with observations that suicidal people tend to fixate on suicidal thoughts (deficits in attentional shifting), they have difficulty communicating a need for help (deficits in verbal fluency), and they are prone to impulsive and risky behavior (poor decision making). However, almost all studies in this area have tested people in cross section, so the predictive accuracy for future suicidal behavior is currently unknown. The one exception is a prospective US study that administered the Stroop test to 124 people in a psychiatric emergency department.79 Attentional bias towards suicidal themes independently predicted suicide attempts within six months even after controlling for a variety of factors including history of suicidal behavior, severity of suicidal ideation, and prediction of suicidal behaviour by both clinicians and patients.
Is suicide risk assessment worth while?
The World Health Organization recommends that all people over the age of 10 years with a mental disorder or other risk factor should be asked about thoughts or plans of self harm within the past month.1 The use of scales or tools is more controversial. National Institute for Health and Care Excellence guidelines encourage risk assessment and needs assessment of patients but oppose the use of risk assessment tools to determine patient disposition and treatment.80 Similarly, a review for the US Preventative Services Task Force found insufficient evidence to support screening tools in primary care.81 Opponents of tools argue that the low precision of assessment tools renders them useless. They are consistent in their low specificity and PPV, which limits their predictive utility and can result in the inappropriate allocation of sparse resources.82
Furthermore, one of the most consistent predictive variables for future self harm is a history of previous self harm, and this subgroup may already be in treatment. The prediction of incident self harm is a high clinical priority but extremely challenging given its relatively low incidence. It may be a worthwhile approach to optimize NPV in risk assessment because the identification of true negatives could help preserve health resources. One observational UK study found that the process of assessment itself correlated with a lower likelihood of future suicidal behavior.83 This speaks to an often overlooked aspect in risk assessment: that doctor-patient contact can provide an important therapeutic effect. Education of trainees in the assessment of suicidal risk is needed yet difficult to achieve when assessment practices are highly variable. It is imperative that standards of care are consistent across and within regions, and in an effort to achieve this the National Action Alliance for Suicide Prevention recently published training guidelines for clinicians.84
Guidelines for risk assessment
Several organizations have published clinical practice guidelines for assessing and preventing suicide. These include national guidelines from various countries and practice guidelines from specific organizations. Box 2 details a selected list of guidelines available online. A review published in 2014 noted that although some components are similar across most guidelines, inconsistencies include how to stratify risk, recommendations for outpatient management and means restriction, and guidelines for training.85 Consistent aspects of suicide prevention guidelines include the importance of assessing risk and protective factors and of clarifying the degree of suicidal intent, as well as recommendations regarding interventions.
Box 2: Selected clinical guidelines for suicide prevention available online
Self harm and suicide. 2015. www.who.int/mental_health/mhgap/evidence/suicide/en/
International Association for Suicide Prevention (IASP)
IASP guidelines for suicide prevention. 2015. www.iasp.info/suicide_guidelines.php
National Institute for Health and Care Excellence (NICE)
Self-harm: the short-term physical and psychological management and secondary prevention of self-harm in primary and secondary care. 2004. www.nice.org.uk/guidance/cg16/chapter/1-recommendations
New NICE guidance for the longer-term management of self-harm. 2011. www.nice.org.uk/guidance/cg133/resources/new-nice-guidance-for-the-longerterm-management-of-selfharm
American Psychiatric Association (APA)
Practice guideline for the assessment and treatment of patients with suicidal behaviors. 2010. http://psychiatryonline.org/pb/assets/raw/sitewide/practice_guidelines/guidelines/suicide.pdf
Royal Australian and New Zealand College of Psychiatrists (RANZCP)
Australian and New Zealand clinical practice guidelines for the management of adult deliberate self-harm. 2004. www.ranzcp.org/Files/Resources/Publications/CPG/Clinician/CPG_Clinician_Full_DSH-pdf.aspx
Substance Abuse and Mental Health Services Administration (SAMHSA)
SuicideAssessment Five-step Evaluation and Triage. 2009. www.integration.samhsa.gov/clinical-practice/safe-t_card.pdf
Interventions to reduce suicide in people with mental health problems
Mental healthcare can reduce suicide rates,86 yet most suicidal people receive no treatment.87 Studies from the UK and US show that only 60% of people presenting with self harm receive a mental health assessment in the emergency department.88 89 This section will highlight specific areas of clinical interventions for suicide. Although it is not a comprehensive review, it will complement other reviews that have examined interventions to prevent suicidal behavior.90 91 92 93
Evidence supports specific psychotropic drugs to reduce the risk of suicide.94 However, research in this area has not rigorously pursued RCTs that aim to prevent death by suicide associated with psychiatric illness. Psychological autopsy studies in high income countries suggest that more than 90% of people who have died by suicide had a mental disorder at the time of their death,95 implicating mental illness as an important and potentially modifiable risk factor. This implies that drug treatment of mental disorders is an important approach to preventing suicide.
Identification and treatment of depression is a key aspect of suicide prevention.10 Several studies have shown that antidepressants reduce suicidal thoughts and behavior, although evidence for their effect on suicide is controversial and may vary with age (see below). A non-systematic review described a series of clinical cohort studies showing risk reductions of between 40% and 81% in suicide attempts and suicide among depressed patients taking antidepressants.96 One meta-analysis of randomized placebo controlled trials of antidepressants found an increased risk of suicidal thoughts and behavior in participants aged less than 25 (odds ratio 1.62, 95% confidence interval 0.97 to 2.71), but a protective effect in those aged 25-64 years (0.79, 0.64 to 0.98) and those over 65 years (0.37, 0.18 to 0.76).97 An observational study from Finland showed that current use of antidepressants was associated with a 39% increased risk of suicide attempts but a 32% reduced rate of completed suicide compared with periods without antidepressant use.98 The risk of attempted suicide is 2.5 times higher in the month before starting an antidepressant compared with the period of antidepressant use.99 When interpreting these findings it is important to consider regression to the mean as an explanatory mechanism because the natural course of depression may result in its improvement (and consequent decrease in suicidal behavior), irrespective of the effects of treatment.
A challenge when using psychotropic drugs is that self poisoning is a common method of suicide and the agents used to self harm are often the same ones prescribed with therapeutic intent. Antidepressant drugs are involved in 20% of self poisoning suicides in the UK,100 and tricyclic antidepressants are the most lethal. A UK study found that tricyclic antidepressants have more than a 20 times greater case fatality in overdose than the selective serotonin reuptake inhibitors (SSRIs).101 In that study, 1088 suicide deaths in England and Wales involved tricyclic antidepressants, whereas 85 deaths involved an SSRI.
Venlafaxine has an intermediate risk in overdose, and among selective serotonin reuptake inhibitors the risk of poisoning death seems to be greatest with citalopram. Clinicians can try to mitigate this risk by choosing less toxic agents when treating patients potentially at higher risk (those with repeated self poisonings) and by prescribing limited supplies of the drug on a single prescription, at least in the early stages of treatment when risk is greatest.
Link with suicidal behavior
The problem is even more complex when considering the major controversy over the past 12 years regarding the Food and Drug Administration imposed black box warnings about suicidal behavior and antidepressants. The controversy persists today, with opponents highlighting reduced rates of antidepressant use, a reluctance of physicians in the US to diagnose depression, and increased rates of suicides in young people in regions with lower rates of antidepressant prescription.102 103 An analysis of four RCTs that examined the use of fluoxetine for major depression in young people (708 treated) found no increase in suicidal behavior,104 whereas a meta-analysis of antidepressant drug use for all categories of depression (4780 young people treated) showed a 2.3 increased odds of suicidal behaviour (1.04 to 5.09). This last finding corresponded to an absolute risk increase of 3.6 suicidal behaviour events per 1000 people when antidepressant treatment was compared with placebo.97 However, older adults who take antidepressants consistently have a reduced risk of suicidal behavior. Active monitoring of suicidal thoughts in people taking antidepressants, particularly young patients, and frequent follow-up in the short term are sound clinical decisions that may help to mitigate the risk of suicide during early phases of treatment.
Other psychotropic drugs
There is good evidence for the antisuicidal properties of lithium.96 Meta-analyses of RCTs in unipolar and bipolar disorders show that lithium significantly reduces suicide compared with placebo (odds ratio 0.13, 0.03 to 0.66).105 106 Interestingly, observational studies of naturally occurring lithium in drinking water show an inverse association with suicide; regions with higher concentrations (albeit subtherapeutic) have lower rates of suicide.107 108 109 In a Danish observational study, outpatients who continued to fill prescriptions for lithium had a lower risk of suicide than did those who filled only one script (relative risk 0.43, 0.2 to 0.69).110 Taken together, this consistent line of evidence underscores lithium as an important intervention for preventing suicide in patients with mood disorders.
Evidence also suggests that other mood stabilizers may reduce suicide risk, although most data come from observational studies.111 In a cohort study of patients with bipolar disorder, antiepileptic drugs as a class reduced suicide attempts by 81% when compared with patients with bipolar disorder not receiving psychotropic drugs.112 This protective effect was also seen when compared with the patient’s own history before starting the drug. An observational study in Denmark showed that, similar to lithium, continuous use of antiepileptic drugs correlated with lower suicide rates compared with a single prescription.113 However, a large US cohort study found that the risk of suicide was 2.7 times higher with divalproex treatment than with lithium.114
Clozapine has an approved indication for the reduction of suicidal behavior in patients with schizophrenia,115 but it tends to be reserved as a late option in the treatment algorithm.116 A meta-analysis of observational studies and randomized trials showed that compared with other antipsychotic drugs, people prescribed clozapine had a reduced incidence of suicidal behavior (risk ratio 3.3, 1.7 to 6.3).117
Ketamine: an emerging drug treatment
Early evidence suggests that ketamine may be an effective intervention for preventing suicide.118 Ketamine seems to possess unique antidepressant effects that are rapid in onset (hours) with consequent neurogenesis mediated through the NMDA (N-methyl-D-aspartate) neurotransmitter system.119 It also reduces suicidal thoughts within hours, mostly but not entirely due to reductions in depressive symptoms.120 One RCT using an active placebo (midazolam) showed that 53% of patients treated with ketamine scored 0 on three different measures of suicidal ideation 24 hours after infusion compared with 24% of those treated with midazolam (P=0.03).121 The reductions in conscious suicidal thoughts were also observed unconsciously through IAT testing. Interestingly, the anti-suicidal effect gained from a single dose of ketamine can be sustained for several days.122
Despite these compelling findings, data on ketamine in suicide are limited and no studies have examined its effect on suicidal behavior and suicide. Future research will hopefully answer important questions regarding optimal dosing, frequency of administration, safety, and long term effects. Until then, ketamine remains investigational (and unapproved for this indication) but is nevertheless a potential option for suicidal people.
Electroconvulsive therapy (ECT)
Electroconvulsive therapy has long been recognized as a highly efficacious treatment for people with depression. It is often used in severe and treatment resistant cases and is recommended for patients with intractable suicidal thoughts.123 ECT rapidly relieves suicidal thoughts in depressed patients, with one cohort study showing resolution of suicidal thoughts in 38% of patients after three treatments, and 62% after six treatments.124 125 126 A population based study of all suicides in Finland showed that the rate of ECT use within three months of suicide was extremely low, which could indirectly infer a protective effect or conversely low utilization of ECT.127 Although ECT seems to reduce suicidal thoughts and behavior in the short term, the limited number of available studies show that ECT treatment does not confer long term protection against suicide. However, it is important to consider the potentially higher risk of patients selected to receive ECT, and that most longer term follow-up studies have methodological limitations and were not designed to examine the effect of ECT on suicide.128 129
A non-systematic review published in 2014 summarized psychotherapies with evidence of efficacy in reducing suicidal behavior, focusing on those with evidence from RCTs.130 Participants in these studies were largely people who had self harmed. Meta-analysis of these trials yielded a pooled effect of 32% reduction in likelihood of future self harm when compared with usual care.81 However, this pooled effect was based on RCTs evaluating a range of different psychotherapies including cognitive behavioral therapy (CBT), dialectical behavioral therapy (DBT), and problem solving therapy in different patient populations; larger trials are needed to identify the most effective approaches to managing people who self harm.
Some individual RCTs have shown that CBT reduces future suicide attempts in people with recent self harm and in patients who present to the emergency department after a suicide attempt; this last study found that people in the CBT group were 50% less likely to reattempt compared with the usual care group.131 132 Patients with borderline personality disorder who were treated with DBT were half as likely as the control treatment group to attempt suicide,133 and the mean number of suicidal acts was reduced by 0.91 in those treated with CBT compared with those receiving usual treatment.134 In a large New Zealand follow-up study over one year, problem solving therapy reduced the risk of repeat episodes of self harm in people with a history of repeated self harm by 61%, but not in those who entered the study with only a single episode of self harm.135
The knowledge base for this intervention is limited. Specifically, no RCTs of psychotherapy have demonstrated a reduction in suicide, probably because of a lack of statistical power. In a large observational study in Denmark, psychosocial therapy after self harm reduced suicide over longer time frames, with an absolute risk reduction of 0.5%.136 In addition, the length of follow-up in psychotherapy trials is of short duration; a meta-analysis focusing on interventions in the emergency department found only two trials that examined repeat suicidal behavior over 12 months and the pooled results were non-significant.137
Online methods of delivering psychotherapy to suicidal people are an appealing option for several reasons, including access, healthcare system costs, and stigma. A Dutch RCT of an online suicide specific intervention that featured an unguided self help approach based in CBT, DBT, mindfulness, and problem solving therapy found a small but significant reduction in suicidal thinking (effect size d=0.28, 0.03 to 0.54).138 This finding was more pronounced in people with a history of suicidal behavior. However, among people calling a suicide helpline, augmentation of treatment with internet CBT directed at depression did not reduce suicidal ideation compared with usual treatment.139 This raises an interesting question of whether psychotherapy directed towards the underlying mental disorder is sufficient to reduce suicidal ideation in the individual. A meta-analysis showed that psychotherapy for depressive symptoms did not significantly reduce suicidal thoughts,140 suggesting that psychotherapeutic approaches specifically directed at suicidal symptoms may be needed. This analysis was limited to only three studies and it again noted that many trials of treatment for depression exclude suicidal participants.
Follow-up care for suicidal people includes a variety of approaches such as telephone calls, repeat assessments, case management, and caring letters or postcards.141 Patients themselves often request a need for care in the period after contact with mental health services, preferring telephone contact initially and letters later on.142 A non-systematic review found that five trials of follow-up care reduced future suicidal behavior, whereas four had mixed results and two did not show a reduction.92 However a meta-analysis of enhanced usual care (defined as augmenting treatment or adherence with little or no direct therapeutic contact) that included many of the same studies found no significant reduction in suicide attempts (adults: relative risk 0.90, 0.80 to 1.02; adolescents: 1.44, 0.36 to 5.76).81
Community mental healthcare has been shown in many countries to be important in preventing suicide. In a national study that examined suicide rates after guideline implementations in the UK, several forms of community care, including follow-up appointments within seven days of discharge, and assertive outreach for people who missed healthcare appointments, significantly reduced absolute suicide rates by 0.53/10 000 person years and 0.34/10 000 person years, respectively.86 National policy changes that have enhanced care during the 12 weeks after discharge have resulted in significant reductions in non-fatal self harm—14% within one week of discharge and 11% within one month of discharge.143 In Finland, an ecological study showed lower suicide rates in municipalities that had a higher outpatient to inpatient service type ratio (relative risk 0.93, 0.89 to 0.97).144
In another meta-analysis that pooled several methods of follow-up care (phone calls, postcards, letters, and green cards), the number of repeat episodes of self harm per person was reduced (incidence rate ratio 0.66, 0.54 to 0.80) but not the likelihood of any future self harm.145 The review also specifically examined suicide as an outcome, and although there was a trend towards a reduction in suicide, the finding was not significant. One study included in the meta-analysis was a multi-country RCT of almost 2000 people who had presented to the emergency department with a suicide attempt. It compared treatment as usual with an intervention that included a one hour information session and then nine follow-up contacts over 18 months (telephone calls or personal visits).146 The intervention group had significantly fewer suicides at the end of the study compared with the usual treatment group (0.2% v 2.2%; P<0.001), but there was no difference in the proportion of participants who had a repeat suicide attempt (7.6% v 7.5%; P=0.909).147 A Swedish RCT (included in the meta-analysis) of two telephone calls over one year to people after a suicide attempt showed no difference in repeat attempts compared with no intervention.148 In a controlled study in Spain of emergency department patients presenting with a suicide attempt, six calls over the next year reduced rates of a repeat attempt compared with controls (6% v 14%) and delayed the time to the next attempt.149
A Danish RCT examined the effect of assertive outreach for people discharged from hospital after a suicide attempt using a case management approach that featured eight to 20 follow-up meetings, telephone calls, and text messages, as well as meetings outside the hospital.150 The findings did not support the outreach approach over usual treatment—16% of patients in the intervention group were admitted to hospital for repeat suicide attempts compared with 11% in the control group.
It is cheap and intrinsically feasible to mail postcards or letters to people after they present to hospital with self harm, but the results of such an approach are mixed. In an Australian RCT, mailing postcards for a 12 month period after deliberate self poisoning did not reduce the likelihood of any repeat self harm, but it did reduce the number of self harm events (incidence risk ratio 0.55, 0.35 to 0.87; P=0.01).151 Findings were similar after 24 months and five years (at five years: 0.54, 0.37 to 0.81; P<0.01).152 153 Importantly, these results suggested that the postcards had a sustained effect beyond the end of the intervention. An RCT of postcards mailed over 12 months in Iran showed significant reductions in suicidal ideation, any suicide attempt, and number of attempts at one year.154 The findings were sustained at 24 months, with an 80% reduced risk of suicidal ideation (P<0.001) and 69% reduced risk of suicide attempt compared with controls (P=0.02).155 Another RCT showed that among people who initially refused psychiatric follow-up, those who received letters regularly over five years had lower suicide rates than those who did not (3.9% v 4.6%), although the difference was significant only during the first two years of the study (P=0.043).156 By contrast, postcard studies in New Zealand, Taiwan, and England showed no reduction in self harm or number of self harm events.157 158 159
The assessment and management of suicidal patients is challenging and further complicated by a limited number of efficacious approaches. It is hoped that management will improve as the science of suicide prevention progresses and the number and quality of studies increases. Nevertheless, current evidence shows that several clinical practices can improve the management of suicidal people. Given that risk assessment tools to date are limited in their predictive ability, it might be best to focus efforts on developing effective low resource intensity interventions that acknowledge a high false positive rate. An important first step is to appreciate the heightened risk in specific clinical populations and the temporal association between suicide and service discharge. It is also crucial to recognise the established risk factors such as a history of self harm. Finally, when determining a treatment approach, clinicians should consider suicidal thoughts and behavior as an important therapeutic target.
Future research questions
What constitutes a reasonable standard of care in suicide risk assessment given the challenges in behavior prediction?
Can large scale, longitudinal studies testing multiple assessment methods over short term intervals improve the prediction of suicide?
Which treatment approaches are most effective for which patients in reducing the risk of suicide during inpatient admissions and in the weeks and months after discharge from the emergency department and other psychiatric services?
How will neurocognitive tests be incorporated into clinical practice and will it be difficult to persuade patients who are not consciously aware of their intention to self harm that they need treatment?
Can effective assessment and treatment practices be adequately scaled up given the constraints on psychiatric resources?
Cite this as: BMJ 2015;351:h4978
The authors would like to acknowledge Hayley Chartrand, Yunqiao Wang, and Joanna Bhaskaran for their help with the literature search and manuscript preparation.
Competing interests: We have read and understood BMJ policy on declaration of interests and declare the following interests: none.
Contributors: All authors helped plan and prepare the manuscript and they meet all four ICMJE authorship criteria. Hayley Chartrand, Yunqiao Wang, and Joanna Bhaskaran contributed to the manuscript by performing literature searches and helping with manuscript preparation. JMB wrote the manuscript and is responsible for the overall content as guarantor. The guarantor accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish.
Funding: JMB is supported by a Canadian Institutes of Health Research new investigator award (113589) and a Brain and Behavior Research Foundation NARSAD young investigator grant. DG is a National Institute for Health Research senior investigator. GT is supported by grants from the Canadian Institute of Health Research MOP93775, MOP11260, MOP119429, and MOP119430; from the National Institutes of Health 1R01DA033684-01; and by the Fonds de Recherche du Québec-Santé through a Chercheur National salary award and through the Quebec Network on Suicide, Mood Disorders, and Related Disorders.