U.S. flag

An official website of the United States government

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Jones S, Riste L, Barrowclough C, et al. Reducing relapse and suicide in bipolar disorder: practical clinical approaches to identifying risk, reducing harm and engaging service users in planning and delivery of care – the PARADES (Psychoeducation, Anxiety, Relapse, Advance Directive Evaluation and Suicidality) programme. Southampton (UK): NIHR Journals Library; 2018 Sep. (Programme Grants for Applied Research, No. 6.6.)

Cover of Reducing relapse and suicide in bipolar disorder: practical clinical approaches to identifying risk, reducing harm and engaging service users in planning and delivery of care – the PARADES (Psychoeducation, Anxiety, Relapse, Advance Directive Evaluation and Suicidality) programme

Reducing relapse and suicide in bipolar disorder: practical clinical approaches to identifying risk, reducing harm and engaging service users in planning and delivery of care – the PARADES (Psychoeducation, Anxiety, Relapse, Advance Directive Evaluation and Suicidality) programme.

Show details

Chapter 5Suicidal behaviour in bipolar disorder


People with BD are at high risk of engaging in suicidal behaviours (i.e. suicide and self-harm). However, controversy remains over who is most at risk, which risk factors are specifically related to suicidal behaviour in BD rather than psychiatric illness in general, and how MHS can respond to reduce suicidal behaviours in this group. To address this, four discrete but related studies were conducted: (1) descriptive analyses of data from a large case series of people with BD who died by suicide; (2) case–control analyses of data on recent and current inpatients with BD who died by suicide; (3) case–control analyses and survival analyses of data on people with BD who self-harm; and (4) a thematic analysis of interviews with people who had direct experience of suicide and self-harm in BD. The results confirmed that the risk of suicide and self-harm is high in those with BD. Some characteristics and risk factors are shared with other psychiatric groups, but some may be unique to those with BD, such as longer duration of illness, high number of periods of inpatient care and problems establishing diagnosis. The interviews showed that rapid access to care was crucial in responding to suicidal behaviours in this group and that patients and carers needed to be made more aware of available services. Closer involvement of carers was seen as desirable, when appropriate, to help services respond faster to rapid changes in illness or behaviour and help prevent suicidal behaviours in people with BD.

Chapter overview

This chapter describes research designed to explore suicidal behaviour in people with BD. The introduction outlines what is already known about suicidal behaviour in those with BD from previous work, as well as priorities for future research. This leads to a description of the aims of the current work, which incorporates a series of discrete but related studies. For clarity, the methods and results for each study are described in turn. The discussion and clinical implications/conclusions sections then consider the results across studies and how the findings add to current knowledge about suicidal behaviour in those with BD. This section also reviews how the current findings might be useful from a clinical perspective.


Prevalence of suicidal behaviours

People with BD are at increased risk of engaging in suicidal behaviours (attempted suicide, self-harm, or completed suicide). Suicide is one of the most important causes of premature mortality in this group,240,241 up to 60% of whom will self-harm at least once during their lifetime.6,242 Figures for the prevalence of suicidal behaviour in those with BD vary depending on the methodology employed, but estimates generally agree that 5–20% will eventually die by suicide.6,243,244 Estimates from more recent studies are generally lower, which may be a consequence of changes in diagnostic criteria over time, the inclusion of BD 2 and less focus on inpatient populations.245,246 However, even the lowest estimates still exceed the figures for suicide in the general population (1.4% of all deaths worldwide in 2012247).

In terms of non-fatal self-harm, studies have found that up to 50% of people with BD report having made at least one suicide attempt.248,249 A number of studies also suggest that rates of suicide attempt are 1.5–2.5 times higher in those with BD than in those with unipolar depression.244,250,251

Risk factors for suicidal behaviour in bipolar disorders

Previous suicide attempt has repeatedly been identified as the most powerful predictor of future suicidal behaviour in those with BD.54,243,252,253 Men with BD tend to be at higher risk of completed suicide, whereas women with BD tend to be at higher risk of suicide attempt.54,254256 Risk is also thought to be higher at younger ages, especially in the years immediately after diagnosis;6,20,257259 thus, studies of outpatients have found that most serious suicidal behaviour happens within 10 years of diagnosis.260

There is ongoing debate over which phase of bipolar illness confers the highest risk of suicidal behaviour, although it is rare in mania.249,253 There is support for the view that depressive phases present the highest risk,249,256,261 but mixed states may combine low mood, irritability and the drive to act on suicidal thoughts.262 However, in a systematic review, Hawton et al.54 found no association between completed suicide and phase of illness.

Psychiatric comorbidities are often present in people with BD and may confer an increased risk of suicidal behaviour.6 There is evidence that the risk of suicidal behaviour in those with BD is increased by the presence of lifetime substance use disorders,257,263265 anxiety and panic disorders266,267 and cluster B personality disorders, which include borderline, narcissistic, histrionic and antisocial personality disorders.248,264,268

Other potential risk factors that may increase suicidal behaviours in those with BD include having a family history of affective disorder or suicide,248,264,269 undergoing recent negative life events,124,243,248,254 being single,248 having a serious physical illness,258 having had a higher number of previous episodes and cyclothymic temperament254 and having a comorbid eating disorder.248

Pharmacological treatment and the prevention of suicidal behaviour

Although antidepressants may be of value in the treatment of acute depressive episodes, their potential mood-destabilising effect may make them inappropriate for long-term treatment in people with BD. Over the long term, mood stabilisers are the preferred treatment.258,270 In particular, lithium has received increased attention owing to a growing body of data that suggests that it may have an antisuicidal effect in those with BD.53,242,253,271

Limitations of previous research

A lack of consensus remains on the prevalence of, and risk factors for, suicidal behaviour in people with BD. Methodological differences may account for some of this; for example, retrospective methods looking at antecedents of suicidal behaviour may be subject to recall or social desirability bias or be restricted by the accuracy of medical records, leading to an underestimate of suicidal behaviours. Prospective studies can be more accurate, but these are often limited by sample size and lack statistical power. Follow-up periods vary greatly, and, although longer follow-up may increase accuracy, there are many attendant difficulties, such as changes in diagnostic criteria and treatment interventions over time. Studies restricted to a single site or single patient type, such as recruiting only from among hospitalised patients, biases the results and limits the extent to which these can be generalised across the bipolar population.

Despite the limitations, there is general agreement that individuals with BD have a greatly increased risk of suicidal behaviour. Therefore, it is important that clinicians can reliably screen for characteristics that are indicative of high risk at different stages of the illness. There is a further need for clarity about how such characteristics might influence or precipitate suicidal acts, and how MHS can most effectively respond.


  • To understand the risk of suicide in people with BD in a NHS service setting.
  • To understand the demographic-, clinical- and management-related characteristics and risk factors for suicidal behaviour in people with BD.
  • To understand how these factors interact and influence suicidal behaviour in people with BD.

Methods and results

For ease of interpretation, this section will describe the methods and results used for each of the four studies included in this component of the programme, outlined below.

Study 1 used data from a large, national case series of individuals with BD who died by suicide. The overall characteristics of suicide in BD were explored and the risk of suicide in BD in the psychiatric population was examined. Demographic, treatment and clinical characteristics were then compared between people with bipolar disorder who had died by suicide, and people with other psychiatric diagnoses who had died by suicide.

Study 2 used a smaller case–control data set of current and recent inpatients to systematically compare characteristics identified in study 1 and establish which of them might be risk factors for suicide in people with BD, beyond those commonly found in psychiatric populations.

Study 3 used a cohort design to explore the characteristics of people with BD who self-harm, and the risk of repetition and subsequent suicide in those with BD compared with other people who self-harm.

Study 4 used qualitative methods to further explore factors found to be important in the previous studies, how these and other factors are experienced on an individual level, and the role of MHS in prevention of suicidal behaviour in BD. Rich, experiential data covering multiple perspectives were generated from a series of in-depth interviews with relatives of people with BD who died by suicide, and with people with BD who had self-harmed.

All statistical analyses were conducted using Stata/IC versions 11.0 to 12.1 and/or SPSS [versions 15.0 to 18.0 (SPSS Inc., Chicago, IL, USA) and version 19.0 (IBM Corporation, Armonk, NY, USA)] for Microsoft Windows® (Microsoft Corporation, Redmond, WA, USA).

Study 1: the National Confidential Inquiry (into Suicide and Homicide by People with Mental Illness) database

Clements C, Morriss R, Jones S, Peters S, Roberts C, Kapur N. Suicide in bipolar disorder in a national English sample, 1996–2009: frequency, trends and characteristics, Psychological Medicine, volume 43, issue 12, pp. 2593–602, reproduced with permission.272 © Cambridge University Press 2013.


Our analyses used data from the suicide database of the National Confidential Inquiry (into Suicide and Homicide by People with Mental Illness) (NCI).273,274 Data collection by the NCI involves three stages. In stage 1 the NCI collects basic information on a complete national sample of suicide deaths and deaths that have received an open verdict at coroner’s inquest for all individuals aged ≥ 10 years from the Office for National Statistics (www.ons.gov.uk, accessed 10 February 2012) for England and Wales. It is convention in the UK to include open verdicts in official statistics and research studies, as the majority of these are suicide deaths,275 and excluding cases given open verdicts may result in a 50% underestimate of the incidence of suicide.276 In stage 2 the NCI identifies people who died by suicide who were in contact with MHS in the 12 months before their death, and these become an ‘inquiry case’. In stage 3 the consultant psychiatrist responsible for the care of the patient is identified and sent a copy of the NCI suicide questionnaire to complete.

The suicide questionnaire collects information on social/demographic characteristics, clinical history, details of the suicide, details of care and final contact with services, and the respondents’ views on prevention. The NCI suicide database includes data on inquiry cases from 1996 onwards. The NCI suicide questionnaire has a response rate of > 97%277 and has been shown to be a reliable data collection tool.278

Ethics approval

The NCI has ethics approval from South Manchester Medical Research Ethics Committee, the North West Research Ethics Committee, the National Information Governance Board for Health and Social Care and the Patient Information Advisory Group and approval under Section 60 (now Section 251) of the Mental Health and Social Care Act.279 Individual research and development (R&D) approvals are also held for each trust where data were collected.

National Confidential Inquiry into Suicide and Homicide by People with Mental Illness psychiatric samples

Unless stated otherwise, the BD suicide sample used in the analyses included all cases from England on the NCI suicide database with a primary diagnosis of BD. This is also the case for other psychiatric samples used in the analyses (i.e. major depressive disorder, schizophrenia). Primary and secondary diagnoses are collected routinely, and primary diagnosis is determined by the responding clinician. Diagnostic categories are concordant with International Classification of Diseases, Tenth Edition (ICD-10) classifications. In reporting primary and secondary diagnoses, respondents may use a combination of case note review, consultation with other members of the mental health team and clinical judgement based on their own knowledge of the patient.


Poisson regression was used to carry out trend tests to identify changes over time in frequency and rate of suicide in BD. Tests were run for the whole BD sample, and for age and sex subgroups. Interactions between BD suicide deaths over time and general population suicide deaths over time were also examined, using data from the Office for National Statistics. Only suicide deaths occurring between 1997 and 2007 were included, as data completeness for these years is > 98%. Rates of suicide per year per 100,000 of the population were also calculated and examined for trends over time using Poisson regression.

Characteristics of suicide in bipolar disorder and comparisons with other groups

Descriptive statistics were used to illustrate the demographic, treatment and clinical characteristics of the bipolar suicides. Characteristics were compared between BD suicides and a general group of all other psychiatric suicides, and then with suicides in unipolar depressive disorder and schizophrenia. Any cases with unknown items were removed from the analysis of that item; therefore, the denominator in each analysis is the number of valid cases for each item. Pearson’s χ2 tests of association were used, and for two by two tables the continuity correction is given. Logistic regression was used for detailed between-diagnosis comparisons. ORs and associated 95% CIs are also presented. Because multiple testing was used, a p-value of 0.01 was selected to assess statistical significance. Only variables that reached statistical significance were retained in the model and reported. Age and sex were regarded as possible explanatory variables rather than confounders and so were not adjusted for in the primary analyses; however, in the secondary analyses carried out, all ORs were adjusted for age and sex. Rates of inpatient suicide were calculated per 1000 of the inpatient population for suicide deaths in BD, depression and schizophrenia. Inpatient population data by diagnosis, used as the denominator in these calculations, were obtained from Hospital Episode Statistics (www.hscic.gov.uk, accessed 26 April 2013), and included the overall number of new spells of inpatient care by year and by diagnostic group (ICD-10 codes: F30–31 and F34.1 for BD, F32–39 for depressive disorders, and F20–29 for schizophrenia and other delusion disorders).


Characteristics of suicide in bipolar disorder

There were 1489 BD suicides in our sample. This represented just under 10% of all suicide deaths on the NCI database (9.6% of 15,465 valid cases). The average number of BD suicides per year was 116. There were no significant trends over time in the frequency of BD suicides [incidence rate ratio (IR) 0.99, 95% CI 0.97 to 1.00; p = 0.15], whereas general population suicides over the same time period decreased significantly (IR 0.98, 95% CI 0.98 to 0.98; p < 0.01). There was a significant downwards trend in the frequency of suicide among those aged < 24 years in the BD group (IR 0.84, 95% CI 0.77 to 0.93; p < 0.001) and a similar but non-significant trend among those aged 25–44 years in the BD group (IR 0.97, 95% CI 0.94 to 0.99; p = 0.015), but no trend was apparent in the older groups.

Table 49 shows the demographic and diagnostic characteristics of BD suicides, compared with those of other psychiatric suicides. BD suicides were older, less likely to be male and less likely to have a formal secondary diagnosis. The most common secondary diagnoses in BD cases were personality disorder and alcohol dependence. Those with BD were more likely than those with other psychiatric illnesses to have had their primary diagnosis for > 5 years. There were statistically significant differences between BD suicides and other psychiatric suicides in marital status, employment status and living arrangements, with bipolar suicides less likely to be widowed, unemployed and homeless, but the differences between the two groups in terms of proportion across categories were minimal.



Demographic and clinical characteristics for suicides with a primary diagnosis of BD compared with suicides with any other primary diagnosis

Bipolar disorder suicides were more likely to be current or recent inpatients, to have more than five previous admissions to inpatient care, to be prescribed typical or atypical antipsychotic drugs and to be prescribed lithium or other mood stabilisers. They were less likely than other psychiatric suicides to be in contact with drug and alcohol services. These results are summarised in Table 50.



Care and treatment characteristics for suicides with a primary diagnosis of BD compared with suicides with any other primary diagnosis

Table 51 shows the clinical characteristics at last contact before suicide. BD suicides were more likely to have been seen by MHS in the 7 days before suicide and to have evidence of depressive symptoms at last contact. The majority of cases (> 85%) in both groups were considered to be at low or no immediate risk of suicide. Adjustments for age and sex did not make any significant changes to the results.



Clinical characteristics at last contact with MHS for suicides with a primary diagnosis of BD compared with suicides with any other primary diagnosis

Suicide in bipolar disorder compared with suicide in unipolar depressive disorder and schizophrenia

Rates of suicide for inpatients were highest among those with unipolar depressive disorder, with an average of 2.2 inpatient suicide deaths per 1000 inpatient admissions for depression. Rates of inpatient suicide were similar between schizophrenia and BD: 1.8 and 1.7, respectively.

In the multiple regression model comparing BD and unipolar depression (Table 52), there was little difference in demographic variables, although bipolar cases were slightly less likely to be male. Variables independently associated with BD included having a history of violence, using a suicide method that involved jumping, having had the primary diagnosis for > 5 years, having a secondary diagnosis of drug dependence, being aged ≤ 24 years at onset of illness, having a high number of admissions to inpatient care, being prescribed an atypical antipsychotic medication and being prescribed lithium (although this result is likely to be an artefact of the analysis, as lithium is unlikely to be prescribed to people with unipolar depression). People with BD were less likely than those with unipolar depression to have a secondary diagnosis of anxiety or to be prescribed antidepressants.



Multiple regression model of characteristics of suicide in BD, compared with suicide in unipolar depression

The multiple regression model comparing BD suicides with schizophrenia (Table 53) suggests that those with BD who died by suicide were more likely than those with schizophrenia to be older, employed and white, to have died by hanging, to have been prescribed lithium or another mood stabiliser, to have been prescribed antidepressants, to have experienced negative life events before the suicide, to have had evidence of depression at last contact with services and to have been assessed as at no immediate risk of suicide. People with BD who died by suicide were less likely than those with schizophrenia to be male, single or divorced, to have had a secondary diagnosis of depression and to have been prescribed antipsychotic medication.



Multiple regression model of characteristics of suicide in BD, compared with suicide in schizophrenia

Study 2: National Confidential Inquiry into Suicide and Homicide by People with Mental Illness case–control database (of current and recent inpatients)


Data were collected as part of the NCI’s work, using the same methods, and under the same ethics and research governance approvals as those detailed for study 1. The database contained detailed information on a sample of patients who were admitted to inpatient care, or were within 3 months post discharge from inpatient care, between 1999 and 2006. Cases were people who died by suicide during this time and were identified via the NCI suicide database. Controls were people who were alive on a specific ‘index date’ (i.e. the date of suicide of their respective case) and were identified via data provided by Hospital Episode Statistics (www.hscic.gov.uk/hes, accessed 8 October 2014). There were no other matching criteria used. All cases and controls were aged between 16 and 65 years (older ages were excluded as older people tend to receive care from different services, e.g. Rubenowitz et al.280), had a range of psychiatric diagnoses and demographic characteristics, and were not restricted by any other matching criteria.

The NCI suicide questionnaire was used to collect detailed information on cases, and an adapted version of the questionnaire was used to collect data on living controls. This adapted version made reference to an ‘index date’ instead of the date of suicide. Questionnaires were completed by clinicians involved in the patients’ care. These data were derived from four studies conducted by the NCI, and further details on each study and the data collection methods used can be found in Bickley et al.281 and Hunt et al.282284

Bipolar case–control sample

Sixty-three cases and 89 controls had a primary diagnosis of BD. To create the case–control pairs necessary for this study, we used the bipolar suicide cases to create two case–control data sets. First, bipolar suicide cases and living bipolar controls were matched as closely as possible on date of suicide/index date, the maximum difference being 121 days in either direction. The mean difference was 25 days and > 60% were matched to within 20 days. Second, we used the remaining non-bipolar suicide cases as a control group, matching to the bipolar suicide cases on date of suicide to within 7 days in either direction, with a mean difference of 1 day. Over 60% matched to within 1 day of the bipolar case’s suicide date. In order to explore the full range of variables available, we did not impose any other matching criteria on either data set.


Conditional logistic regression was used to compare cases and controls on key characteristics. Any cases with unknown items were removed from the analysis of that item; therefore, the denominator in each analysis is the number of valid cases for each item. ORs, 95% CIs and associated p-values were calculated. Tests were also run to adjust for age and sex. Characteristics that were significant at the p ≤ 0.05 level in the univariate analysis were entered into a multiple regression model. Multiple regression models were attempted, but, owing to the small number of cases included and high levels of multicollinearity (tested using the Stata collin command), the models were unreliable and are therefore not presented.


Twenty-eight (44%) bipolar cases were male. Age ranged from 23 to 65 years, with a mean of 45 years. The most common method of suicide was jumping from a height or jumping/lying before a moving object (n = 20, 32%), followed by hanging (n = 19, 30%) and self-poisoning (n = 8, 13%). Twenty-four (38%) cases had a secondary diagnosis; personality disorder was the most common (n = 10, 42%), followed by alcohol dependence (n = 7, 29%). Forty-three (68%) bipolar cases were inpatients at the time of the suicide.

Bipolar disorder suicide cases versus bipolar disorder living controls

Table 54 shows the key results of the univariate conditional logistic regression analyses.



Comparison of characteristics between cases of suicide in BD and living bipolar controls (matched by date of suicide/index date)

Odds ratios were wide for many variables, reflecting the relatively small sample size. Compared with living controls, cases who died by suicide were nearly 30 times more likely to have experienced recent adverse life events (such as recent unemployment, bereavement or relationship problems) (OR 29.99, 95% CI 3.95 to 212.89). Cases were over 20 times more likely to have a history of self-harm (OR 23.00, 95% CI 3.11 to 170.31), and three times more likely to have a history of alcohol use (OR 3.25, 95% CI 1.06 to 9.97). Those who died by suicide were also more likely to be prescribed a selective serotonin reuptake inhibitor or a serotonin–norepinephrine reuptake inhibitor (SNRI) antidepressant (OR 2.63, 95% CI 1.16 to 5.93), and to have evidence of depressive illness (OR 3.29, 95% CI 1.41 to 7.66) and evidence of self-harm (OR 5.00, 95% CI 1.10 to 22.82) at last contact with services. Cases were also more likely than living controls to be considered at moderate long-term, but not immediate, risk of suicide at last contact (OR 2.86, 95% CI 1.21 to 6.76). Living with family was less common in cases than in controls (OR 0.45, 95% CI 0.20 to 0.99), and controls were consistently rated as at low risk of suicide in evaluations of both immediate (OR 0.29, 95% CI 0.12 to 0.71) and long-term risk (OR 0.42, 95% CI 0.20 to 0.87). Adjustment for age and sex did not change the significance of any variables.

Bipolar suicide deaths versus non-bipolar suicide deaths

Cases with BD were more likely than those with other psychiatric diagnoses to have had their primary diagnosis for at least 6 years (OR 3.00, 95% CI 1.19 to 7.56) and were much more likely to have been prescribed lithium (OR 33.00, 95% CI 4.51 to 41.28). However, they were less likely to be male (OR 0.41, 95% CI 0.20 to 0.82), to have a history of drug use (OR 0.41, 95% CI 0.17 to 0.99) and to be assessed as at no long-term risk of suicide (OR 0.13, 95% CI 0.02 to 1.00). Adjusting for age and sex made history of drug use and assessment of no long-term risk non-significant, while the same adjustments made evidence of alcohol use at last contact significant (OR 0.17, 95% CI 0.03 to 0.97; p = 0.05). The results of this analysis are shown in Table 55.



Frequencies, ORs and associated p-values for univariate analysis of bipolar suicide cases and non-bipolar suicide controls, matched on suicide date

Study 3: Manchester Self-Harm Project database

Reprinted from Journal of Affective Disorders, Vol 173, Clements C, Jones S, Moriss R, Peters S, Cooper J, While D, Kapur N, Self-harm in bipolar disorder: findings from a prospective clinical database, 113–19, Copyright (2014), with permission from Elsevier. © 2014 Elsevier B.V. Published by Elsevier Inc. All rights reserved.285


The study used data from the Manchester Self-Harm Project (MaSH), which is a collaboration between the University of Manchester, local MHS providers and the general hospitals that serve the population of the City of Manchester (further information is available at www.manchester.ac.uk/mash, accessed 21 February 2014). Established monitoring methods identify episodes of self-harm presenting to study hospitals. The MaSH and, subsequently, the current study define self-harm as intentional self-poisoning or self-injury, irrespective of motivation.286 Data are collected on demographics and method of self-harm for all episodes, with more detailed clinical information obtained for those assessed by mental health specialists and/or by emergency department doctors (e.g. previous self-harm, drug or alcohol misuse, psychiatric treatment, mental state, risk assessment and clinical follow-up). Case ascertainment is essentially complete, and data cover all self-harm presentations between September 1997 and December 2010. Data completeness for individual variables is high.287

The nested case–control sample

The MaSH database is suited to a nested case–control design, as it represents a well-defined source population, of known size, permitting cases and controls to be selected from a matched and relevant risk set, making optimal use of the data available for what is a relative rarely recorded event (e.g. self-harm by people with a diagnosis of BD).

Bipolar disorder cases were any individual with a primary diagnosis of BD recorded in the diagnosis field for any episode of self-harm. Individuals with conflicting diagnosis in other self-harm presentations were excluded from the bipolar group (i.e. three individuals were excluded due to a conflicting diagnosis of schizophrenia recorded in another episode of self-harm). Cases and controls were matched 1 : 5 on date of self-harm. If fewer than five controls were available for a particular date, then controls were selected from the following day, and so on, until the five control condition was satisfied, in order to optimise statistical power.288 Power calculations indicated that five controls would give 99% power to detect a 30% difference at a 0.05 level of significance.


Descriptive statistics summarised the demographic and clinical characteristics of cases and controls, denominator was the number of valid responses for that item, and ‘unknown’ responses were removed from each analysis. All variables were examined in a series of univariate conditional logistic regressions, controlling for age and sex (an examination of the distribution of the age variable indicated it was not normally distributed; therefore, age groups were used instead of a continuous age variable), and significant variables (at a p-value of ≤ 0.05) were entered simultaneously into a multiple conditional logistic regression model. Summary statistics, Kaplan–Meier survival curves and a Cox regression model were used to examine repetition of self-harm in the case–control sample.


The study comprised 103 bipolar cases and 515 non-bipolar controls. Cases with a recorded diagnosis of BD made a total of 269 presentations over the study period, representing around 0.7% of all self-harm presentations. Nearly 60% (n = 60) of bipolar cases had at least one repeat presentation for self-harm recorded. The average period of follow-up after the initial self-harm episode was 7.7 years.

The key results of the univariate regression analyses are shown in Table 56. BD cases were more likely to be female, aged between 45 and 64 years, married or living with a partner, unemployed or registered as long-term sick and to have had sleep disturbance before the self-harm episode. They were nearly three times more likely to have had previous episodes of self-harm, and around seven times more likely to have current or previous psychiatric treatment. Self-harm was more than twice as likely to be considered a direct response to psychiatric symptoms in the bipolar group as in the control group, but less likely to be linked to either relationship problems or alcohol use.



Sociodemographic and clinical characteristics of bipolar cases and non-bipolar controls who had self-harmed

The multiple regression model in Table 57 shows that bipolar cases were three times more likely than non-bipolar controls to be unemployed, to have previous psychiatric treatment and to have sleep disturbance. Cases were less likely to have relationship problems, current alcohol use or suicide plans.



Multiple regression analysis of characteristics of people with BD who self-harmed compared with others without BD who self-harmed

The overall repetition rate (i.e. the proportion of the sample with at least one repeat episode at any time during follow-up) was 58% (n = 60) for cases and 25% (n = 129) for controls, and there was a significant association between diagnosis of BD and repeated self-harm (χ2 = 44.58; p < 0.001). Figure 27 shows time to first repeat in a Kaplan–Meier plot. A higher proportion of cases than controls had a re-presentation for self-harm during the study period (log-rank χ2 = 50.32; p < 0.001). However, there was little difference in the average time to first repeat (658 days for cases and 548 days for controls), and the difference was not significant [t(187) = –0.90; p = 0.37].

FIGURE 27. Kaplan–Meier survival curves of time to next presentation for self-harm.


Kaplan–Meier survival curves of time to next presentation for self-harm. Reprinted from Journal of Affective Disorders, Vol 173, Clements C, Jones S, Moriss R, Peters S, Cooper J, While D, Kapur N, Self-harm in bipolar disorder: findings from (more...)

Diagnosis of BD was significant in the Cox regression model (HR 3.08, 95% CI 2.20 to 4.18; p < 0.01) adjusted for age and sex. A post hoc sensitivity analysis, restricting the control sample to those with a formal psychiatric diagnosis (n = 95), attenuated the HR for BD, but it remained significant (HR 2.40, 95% CI 1.56 to 3.70; p < 0.01). In the multiple regression model, the inclusion of additional explanatory variables reduced the HR for BD, but it still maintained significance at the 0.05 level in the adjusted analysis (HR 1.68, 95% CI 1.10 to 2.56; p < 0.02). Previous self-harm (HR 15.28, 95% CI 6.17 to 37.84; p < 0.01) and current psychiatric treatment were also significant predictors (HR 1.80, 95% CI 1.20 to 2.70; p < 0.01).

Mortality outcome data were available for 79% (n = 491) of the case–control sample (data linkage was carried out from 2001 onwards). Four (3.9%) of the bipolar cases and seven (1.4%) controls died by suicide during the follow-up period.

Study 4: qualitative interview study – experiences of suicide and self-harm


The original recruitment target was 20 participants: 10 relatives of people with BD who died by suicide and 10 people with a diagnosis of BD who had self-harmed. Recruitment into both groups took longer than originally planned owing to difficulties identifying and approaching people who met the inclusion criteria. The original method of recruitment for the self-harm group (via the MaSH database) resulted in three completed interviews, and the original recruitment method for the relatives group (via the NCI database and GP referrals) resulted in four completed interviews. This method was time-consuming and required a large number of research governance and R&D/patient identification centre approvals to cover recruitment areas for a small return in numbers of participants. To achieve the recruitment targets, the methods of recruitment were expanded. Applications for additional approvals from the National Research Ethics Service added to delays in completing the qualitative study.

Additional recruitment methods included the following.

  • Recruitment via clinical studies officers (CSOs) at Manchester Mental Health and Social Care Trust (for self-harm participants only). CSOs sought patient referrals via care co-ordinators, clinicians, and patient groups within the trust.
  • Bipolar UK advertising (for both self-harm and relatives participants). Two adverts were placed in Pendulum, the magazine produced by Bipolar UK. The first, run in 2012, focused on the relatives group of participants and was successful in attracting a number of participants; therefore, a second advert was run in 2013 asking for participants for the self-harm group.
  • Bipolar UK support groups and events. Two support groups based in the north-west were visited in person and study information was handed out. Flyers were distributed nationally via Pendulum magazine and information was also given out at the 2012 Bipolar UK annual conference.
  • PARADES referrals. Information was given out to people attending events related to other PARADES WSs, and staff passed on information to suitable potential participants where appropriate.

Recruitment was completed in August 2013, with a total of 22 participants, 11 in each group, thereby slightly exceeding the original targets. As recruitment expanded to cover all of England, it was necessary to use telephone interviews for some participants instead of the face-to-face interviews originally intended. The resulting interview data are, therefore, a mixture of both methods. The transcription of the first interviews was initially carried out by the RA on the study, and later interviews were transcribed by a professional service (1st Class Secretarial Services, Gorebridge, UK; www.1stclass.uk.com, accessed 20 January).

Thematic analysis

A thematic analysis was used to examine the transcripts. The method has been described in detail by Braun and Clarke.91 Thematic analysis is unattached to any specific theoretical framework and, as such, offers the flexibility to explore and interpret detailed qualitative data with the goal of providing complete and useful accounts. The freedom and flexibility of thematic analysis was appropriate for these data, as the study was essentially exploratory in nature, allowing interviewees to guide the narratives towards topics that they felt were most important in relation to suicidal behaviour in people with BD.

An early review of the first six transcripts showed that experience of MHS and key issues around barriers and access to care were central in carer and SU accounts. Therefore, the analysis focused on these aspects of the narratives in the hope that it would give insight into how MHS respond to people with BD who are at high risk of suicidal behaviour, and if there might be missed opportunities for intervention and prevention. This focus is also consistent with studies 1–3 (clinician-generated data), and the most useful way to explore the data within the allotted time. These interviews are a rich and valuable source of data about a hard-to-reach population, and there is potential for further analysis at a later date.


The three previous studies have shown that people with BD who die by suicide are more likely than other psychiatric suicides to have high levels of contact, and recent contact, with MHS before engaging in suicidal behaviours (study 1); that people with BD who die by suicide are also more likely than living BD controls to have previously self-harmed, and to have had their primary diagnosis for longer than other psychiatric suicides (study 2); and that people with BD who self-harm are more likely to repeat self-harm and to repeat sooner than other people who self-harm (study 3). If these results are reliable, the question that arises is why people with BD who are at high risk are not being identified as such by MHS and being managed appropriately to prevent further suicidal behaviour.

This study uses the lived experience of SUs and carers/relatives (the term ‘carer’ will be used throughout the discussion) to explore issues that are important in their experiences of MHS. Carers formed a more coherent group in terms of shared experiences, as they were all bereaved by the suicide of someone with BD. Therefore, the analysis is focused primarily on these interviews. SUs were a more diverse group, and their experiences of suicidal behaviour were variable across the gamut of suicidal behaviours (e.g. from superficial self-injury to medically serious suicide attempt). SU interviews were used as a source of supporting information to test initial themes, and any novel items were incorporated into the existing thematic structure.

The two main themes identified were:

  1. experience of MHS at times of crisis
  2. routes to and roadblocks in communication.

Each theme will be outlined and, for brevity, illustrated only by key quotations. Figures 28 and 29 showing possible hierarchies and relationships of subthemes are included to aid interpretation. Themes and subthemes are broadly related, and this analysis presents just one possible interpretation of the data. For each participant quotation, anonymous identifiers are used (S for service user and R for relative), along with transcript line numbers.

FIGURE 28. Map of the theme ‘experience of MHS during crisis’.


Map of the theme ‘experience of MHS during crisis’.

FIGURE 29. Map of the theme ‘routes to and roadblocks in communication’.


Map of the theme ‘routes to and roadblocks in communication’.

Experience of Mental Health Services at times of crisis

Subthemes within experience of MHS are grouped under two headings: (1) accessing MHS and (2) quality of care.

Accessing Mental Health Services

See Figure 28.

Access to effective mental health care was considered to be of paramount importance during times of crisis and when there is imminent risk of suicidal behaviour. Not only were the majority of SUs (and carers) aware of when they needed help, they actively took steps to secure it. However, help-seeking was often blocked at an early stage, whether or not they had been in contact with MHS before:

I spent 3 days in and out of A&E [accident and emergency] begging them to help me.

S2 30–31

[I]t used to make him angry, cross, you know, and he couldn’t get people to sort of listen, and to help him.

R8 255–256

Lack of knowledge of services available, and/or knowledge of how to access them, seemed to compound the situation:

We didn’t realise that we had a pass, if you like, back into the mental health system so we went back to the GP.

R6 75–76

Services and organisations outside the NHS were seen as beneficial, but they were not signposted and so SUs and carers had to find them independently:

Then we came across someone called Crisis Point, I ended up there for 2 weeks, now they were good, they made a massive, massive, massive difference to me.

S1 510–511

I said can I short circuit it by paying for it and she sort of said ‘I’m not allowed to recommend anyone’. So it was this, I’m trying to do the best for my wife and you won’t help me.

R2 713–715

Similarly, SUs and carers felt that there were many barriers to accessing MHS. GPs were often the first point of contact, and the first obstacle, in accessing care, and acted as gatekeepers:

I have to say the GPs were very much, ‘oh pull yourself together’, you know, sort of attitude.

R6 33–36

I went to the GP, my friend, my best mate came with me, but I wish she’d sat in on the consultation with me, because you wouldn’t believe what this doctor said to me, he was horrible.

S2 147–149

However, this was not universally the case:

I cannot fault my GP, because they [ . . . ] could not have dealt with it any better. He was wonderful and got us into the psychiatric hospital as soon, the same day, which was just unbelievable.

R3 501–503

Other people experienced problems when they were put in contact with MHS:

Him and an assistant [ . . . ] I was as honest as I am being with you and he just turned to her and he went, ‘do you believe a word she’s saying’, and this woman just looked at him and went, ‘no’, and I felt about that big, and I’m sort of thinking I need help. Help.

R3 501–503

In some cases there was a sense of being seen as ‘not ill enough’ to need MHS. This may be particularly relevant to people with BD, who tend to be higher functioning than some other psychiatric groups:

I have so much insight when I’m ill, people don’t realise how ill I am, and that’s why I struggle to get help, every time. Because I know where I am and what’s going on, and I know I can express what I think’s going on in my head they don’t realise that I’m ill, as ill as I am. I get to a critical desperate point where I have to scream for help because people, professionals don’t realise how ill I am. It’s my downfall really.

S2 130–134

Diagnosis, or the lack of a correct diagnosis, can also be a barrier to accessing appropriate care and, therefore, the correct treatment. One participant described a cycle of suicide attempts, followed by psychiatric assessments that failed to identify a problem and then being discharged home without help, describing it as, ‘just like a merry-go-round’ (S1 311). There is a danger that people could be prescribed inappropriate medication when BD is not recognised or is overlooked:

He didn’t bother looking at my history, he should have known that I had bipolar, he just handed me antidepressants without a mood stabiliser, so if I’d just had antidepressants on their own I would have gone manic.

S2 149–152

Quality of care

Quality of care seemed to be judged not necessarily on outcomes, but on how sympathetic and understanding clinicians were perceived to be:

I got him to our doctor who was very sympathetic, who came, we went to see him, and then he came round here.

R3 107–108

This was related to two factors: continuity of care over time and trust. Continuity of care was seen as crucial in the care of SUs with BD, particularly when there were clear signs of suicidal thoughts or behaviour:

I think is one of the most significant problems in the system really is that at that point the psychiatrist [ . . . ] said well I can’t follow her [ . . . ], and there was no mechanism for transferring anything, alerting anybody.

R5 141–144

With continuity of care comes the ability to build supportive relationships with MHS staff, which feeds into developing trust. Trust was seen as vital, and clinicians who were open and inclusive towards carers were considered more reliable and trustworthy:

Most of them they excluded me [ . . . ] there was one guy, he was one of the few people that both [the patient] and I trusted because he was quite happy to have us both in at the same time.

R2 656–659

When there was neither continuity nor trust between SUs (and carers) and services, the care was seen as superficial:

Some were ok, some were terrible, so they’d come every day and drop my medication off, some wouldn’t even come in and talk to me.

S2 219–220

I mean I can’t remember how many different people, social workers have dealt with him. It would be sort of a few months and then they were having a baby, fair enough the woman would be having a baby, but you know it doesn’t help the person you are working with, and then it was somebody else who was a trainee.

R8 398–401

A better understanding of the individual patient (unique knowledge of that patient) and their needs was valued:

I said [the patient] liked you, and he always used to say oh [GP] is great. I said, because you understood him as much as anybody understand [him], you understood him.

R1 504–506

He had a CPN [community practice nurse] [ . . . ] who was extremely good from the point of view that he would speak to me and ask where [the patient] was up to with his behaviour.

R4 183–184

The converse was also true, whereby the patient and their typical behaviour were not known and so there was concern that red flag behaviour might be overlooked:

But there was quite a lot too much, and from the medics, never mind anyone else, of people just listening to what [the patient] had to say and believing it, and taking it at face value and she was not in a mental state to be telling anyone a straight tale.

R5 525–529

There was also concern about SUs not being treated as individuals with individual needs, and being reduced to nothing but their illness:

I think they look at them as if, well they just look at the illness and not the whole person maybe that needs to come in.

R7 786–787

Routes to and roadblocks in communication

See Figure 29.

Essentially, this theme is concerned with how information flows between the SU, MHS and carers. It also reveals how information regarding the mental health care of people with BD can become unreliable, and where problems may be encountered that contribute to risk.

How much and how explicitly the SU communicated with both MHS and carers was frequently mentioned as a source of concern. The non-disclosure of important information was not only an obstacle in accessing care, but also a cause of conflict between MHS and carers, as it precipitated an imbalance of available information when contributing to care decisions.

Failure to disclose information was not always intentional. In one interview it was clear that the SU had a disorganised lifestyle, which led to missed appointments that the carer mistakenly blamed on MHS:

I used to get quite cross because, you know, he would say he hadn’t seen this person, but when I inquired into it, the more likely thing was that they had gone to his house and he wasn’t there.

R8 357–360

When this failure to disclose was intentional, there were a number of motives proposed, including protecting family members from distress:

She had attempted it maybe about four times, but she didn’t want us to know, she didn’t want to worry us.

R7 194–195

Another motive was protecting SUs’ privacy (also linked to issues of confidentiality), and there was evidence of reluctance to open up to MHS or carers:

I think [the patient] was quite a private person as far as, as wanting to involve other people, he’d rather keep, keep it to himself. I think that maybe that’s why social, social services didn’t make more of an effort, because he seemed to be holding it together.

R3 685–687

In a few cases there was an element of denial:

She was putting up this front that she was absolutely fine, and that was the really big message that was coming across, that she thought she was fine, the world was against her.

R5 422–423

The impact on care was that SUs often presented as being less ill than they perhaps were, which is particularly important if suicidal behaviour or intent is being hidden:

[H]e certainly, sort of, was able to give the, the consultant the impression that, you know, I’m alright.

R4 599–600

Many carers raised concerns with MHS that were not followed up owing to the SUs’ denial of suicidal thoughts or plans. The knock-on effect is that risk is not accurately assessed and the opportunity for intervention is missed:

I came down and found him in the chair he took an overdose and he’d promised [the psychiatrist] that he wasn’t suicidal. He said that he was alright now things were getting done.

R1 78–79

Carers saw themselves as having an important role in the care of patients with BD, as they often possessed additional information on behaviour changes or life events that could help to inform the person’s treatment and care:

[W]e were aware that he was drinking. I brought all this, about a month before [the patient] died, I brought all this to the attention of his CPN [ . . . ] and he was quite shocked by that, and the living conditions, how [the patient] was living.

R4 225–230

Carers wanted to be seen as partners in the care of their loved ones:

[A]s a family we were their biggest resource and they never sort of came to us to help. Either to help us, or to use what we were saying to help him.

R4 276–277

There was support for the idea that mental health care should be collaborative between the SU, carers and MHS. The key point was that direct involvement, and being kept informed of care decisions, was valued:

I was involved, and I think that helped.

R3 703

However, SUs did not always want their friends or family involved, for a number of reasons, and this could create conflict on multiple fronts, such as between MHS and the SU:

As I say each time over the years, what to me had happened, and also [partner] has said this, I mean he’s very angry about it all, is that a lot of them have just seen that he’s been at the side of me, they’ve signed me out and just gone ‘you deal with it,’ and it’s been landed on [partner’s] shoulders, every single time and he’s not medically trained, and why should it be put on someone like that?

S1 760–765

Between the SU and their carer(s):

I was allowed to live with my best mate whilst I recovered rather than be put in hospital. I begged my best mates to put me in hospital because I know how difficult I am when I’m ill.

S2 214–216

And between MHS and carers:

[I]t was just extremely unhelpful, and inflamed things enormously, and what I was trying to do was to give the doctor information that [the patient] wouldn’t have given her about the way she was behaving and the way she was thinking.

R5 131–134

Mental Health Services staff are bound by strict rules regarding confidentiality of patient information, and this may exacerbate conflict between MHS and carers:

They then sort of diagnosed her as, she is bipolar and we’re going to treat her as bipolar, and I didn’t know anything about this because patient confidentiality was one of the big problems, you do not get any information out of them. They will tell you what they want to tell you and they won’t give you the full picture.

R2 220–224

You just got told on many, many occasions that, you know, its patient confidentiality. But, when you’re dealing with somebody who’s suffering from mental health, and if I turn round and said to them, I said ‘if my son had cancer, you would have involved me, to care for him quite well at home’.

R4 311–314

This can cause carers to feel frustrated, as they are unable to help, even when the information they have to offer is important:

[W]hen he died, you know, the psychiatrist had actually said that he didn’t think it was planned suicide, because he didn’t think [the patient] was in that frame of mind to make that decision at that time, even though, even though he had been telling me that he wanted, you know, he didn’t want to live. He wanted to die for probably months.

R6 457–461

What is very clear from the interviews is that information needs to flow between all parties for optimal care. When the lack of information was revealed, carers were often left feeling misled and powerless and, most importantly, that they might have been able to do more if they had known more about their relative’s illness and care:

He spoke to [the patient] and then he asked me to go in on my own and asked me what did I know of [the patient’s] condition, and I knew absolutely nothing. He was quite surprised really that I didn’t.

R4 115–117

It was in her records and of course once I’d seen her records I was sort of able to say, why didn’t you tell me all this stuff and I could have said look these are panic attacks and made a far better picture of what was going on, but as it was I was relying on them to put everything together and they just didn’t.

R2 322–325


Study 1

Summary of the characteristics of suicide in bipolar disorder

  • Ten per cent of psychiatric patients who died by suicide had a primary diagnosis of BD, a rate lower than for people with schizophrenia (18%) and people with unipolar depression (36%).
  • A higher proportion of female suicides.
  • Most common in those aged between 25 and 64 years.
  • More likely to have had their diagnosis for ≥ 5 years.
  • Common comorbidities were personality disorder and alcohol dependence.
  • More likely to have been recent/current inpatients and to have had five or more previous admissions.
  • Prescribed mood stabilisers.
  • Recent contact with MHS.
  • Assessed as at no or low immediate risk of suicide.

The results of this study are broadly consistent with previous research on factors found to be associated with completed suicide in people with BD.248,256,257 Our data confirm that suicide in people BD is associated with a high number of previous admissions to inpatient care,248 although the reasons for admission were not available. The rate of suicide by inpatients with BD (who might be considered the most severely ill of those in contact with psychiatric services) was similar to that by inpatients with schizophrenia, both of which were lower than that for inpatients with unipolar depressive disorder.

The number of bipolar cases with substance misuse256,258,264 and/or anxiety disorder146,248 was smaller than might have been expected given the prevalence of comorbid substance and anxiety disorders typically found in BD populations.6,146 It is possible that this was a recording bias; however, differences in diagnostic practices, in terms of clinical recognition of anxiety and/or substance use in BD in the USA and England, cannot be ruled out. However, our study was restricted to a clinical subgroup of people with BD who died by suicide, and previous work has suggested that substance/alcohol use is not strongly related to suicide in these individuals.54

Fewer than 40% of BD cases had evidence of depressive symptoms at their last contact with services before the suicide. The lower than expected proportion of cases with depressive symptoms could be a result of the particular characteristics of our clinical sample, a reflection of the time interval between being seen and death, or an indication of a crucial under-recognition of depression among this group.

Mood stabilisers were the most commonly prescribed pharmacological treatment for those with BD; however, given the possible antisuicidal properties of lithium,242 an important finding was that > 40% of cases were not prescribed mood stabilisers. Data for this variable were not complete (73% complete) and should be interpreted cautiously. It is likely that people not prescribed mood stabilisers would be over-represented in the missing data group and, therefore, the proportion of individuals not prescribed appropriate medication might be even higher.

Compared with unipolar depressive disorder, people with BD who died by suicide were more likely to have a history of violence, to die by jumping, to have their primary diagnosis for > 5 years, to have up to five previous admissions to inpatient care, to be prescribed atypical antipsychotics, to have a comorbid diagnosis of drug dependence, to have a young age at onset and to have five or more previous admissions to inpatient care.

Compared with suicide cases with a primary diagnosis of schizophrenia, those with a primary diagnosis of BD were more likely to be employed, to die by hanging, to be prescribed antidepressants, to have negative life events before the suicide, to be assessed as at no immediate risk of suicide at last contact with services, to be aged ≥ 65 years, to be white and to have evidence of depression at last contact with MHS.

Although the pattern of significant characteristics differed depending on the comparison diagnosis, factors that seem to be specifically associated with BD suicides, taken together, could broadly be classified into three areas: (1) markers of illness severity (e.g. recent or current hospital admissions, comorbidity), (2) factors related to duration or the chronic nature of the illness (e.g. repeat admission to inpatient care, > 5 years post diagnosis) and (3) markers of underlying impulsivity (e.g. history of violence, more violent methods of suicide).289

The majority of cases were seen within 7 days before the suicide and were considered to be at ‘low risk’ when assessed. This is consistent with the results of psychological autopsy studies (whereby detailed data are collected on the deceased from various sources, e.g. family members, treating clinicians and medical records), which show that suicidal intent is rarely communicated when the individual is seen by a health-care professional shortly before the suicidal act, although it may be communicated earlier to friends or family members.290

Study 2

Summary of inpatient suicide by people with bipolar disorder

Bipolar disorder suicide cases versus living bipolar controls
  • The presence of depression and treatment with antidepressant medication.
  • History of previous self-harm and recent episode of self-harm.
  • Alcohol use and/or comorbid personality disorder.
  • Adverse life events.
Bipolar disorder suicide cases versus other psychiatric suicide controls
  • More likely to have had their diagnosis for ≥ 5 years.
  • More likely to be female.
  • Less likely to have a history of drug use.

These analyses go beyond the previous descriptive work on the characteristics of people with BD who die by suicide in describing the possible risk factors for suicide within this particularly high-risk recent and current inpatient group. To our knowledge, inpatient suicide in BD has been examined only once before,291 even though this group may be best placed to receive preventative interventions. It is important that clinicians can accurately assess and monitor risk of suicide, perhaps especially during the transition from inpatient to outpatient care. With little work on suicide in inpatients with BD, it is difficult to say whether or not the risk factors for this group differ from general risk factors for suicide in BD. Although our study is small scale, it provides important clinical information about how these patients might differ and how those most at risk might be identified.

The main results of this study are also generally consistent with those of previous studies on factors related to suicide in people with BD.257,272 In particular, our study emphasises the high rate of previous self-harm found among those with BD who die by suicide.252,253 However, not all previous research is supported; for example, hopelessness was not found to be significant in either comparison, and being male, which has previously been cited as a risk, was found to be protective in the second set of analyses.54,256

Study 3

Summary of self-harm in people with bipolar disorder

  • More likely to repeat self-harm.
  • More likely to be female.
  • More likely aged between 45 and 64 years.
  • More likely unemployed/registered as long-term sick.
  • More likely history of previous self-harm.
  • More likely current/previous contact with MHS.
  • More likely sleep disturbance.

The results support work that has shown that women with BD have a higher rate of suicidal behaviour than men,256 and contrasts with the findings of studies that have shown a greater sex equality in self-harm.292 The results showed that women were over-represented among bipolar cases, even in comparison with ‘others’ who self-harm – a group already known to be disproportionately female.286 One possible explanation is that, although BD itself is unrelated to sex, women are more likely to seek help and, therefore, more likely to be in contact with services and given an appropriate diagnosis. Another explanation is that, although self-harm methods used by people with BD tend to be more potentially lethal in general,292 there is evidence that methods may be particularly dangerous in men with BD, with a higher case fatality.293

Although not significant in the multiple regression analysis (owing to the presence of ‘previous psychiatric treatment’ in the model, which may be acting as a confounder), previous self-harm was significant in the univariate analysis, and the difference in proportions between cases and controls was large, at 74% versus 49%. This is in line with a number of previous studies on both lethal and non-lethal self-harm in BD.54,253

It might also be of note that ‘direct response to psychiatric symptoms’ was the only significant predictor within the precipitants of self-harm variables, and it was more than twice as common in the bipolar cases. If this is considered, along with the tendency for the bipolar cases to be in current or previous contact with psychiatric services and to have previously self-harmed, it could be indicative of a group who are more seriously unwell and who might have had a longer duration of illness. There is some evidence that those with longer duration of untreated illness/incorrect diagnosis, greater illness severity and greater illness burden have a higher frequency of suicidal behaviour and poorer outcomes.242,244 By contrast, those without BD were more likely to report relationship problems as a precipitant.

Although depression and hopelessness were present, to some extent, across all cases and controls, as might be expected,54 these were not significant predictors of repetition. Of the clinical characteristics described in this data set, only sleep disturbance seemed to be a robust independent predictor in BD253 and, indeed, was almost three times more likely to be present in bipolar cases than in non-bipolar controls.

Previous work looking at the association of drug and/or alcohol use with suicidal behaviour in BD has produced contradictory results.292 We found that alcohol use was, in fact, recorded less frequently in cases than in controls. This is somewhat surprising given that alcohol and substance use is recognised as common in people with BD.6 However, there is evidence that alcohol/substance use may not be strongly related to suicidal behaviour in BD,54 and it may be that illness-related factors play a more important role in this group. For example, suicidal plans were much less common in bipolar cases, which supports the hypothesis that impulsivity, a characteristic trait in bipolar patients, may be of particular importance.294 Impulsivity, driven by positive or negative affect, in the presence of other factors recognised as associated with self-harm and suicide more generally, may be especially important in precipitating episodes of self-harm and may be a key area for future research.

The control sample contained data on 95 (18%) individuals who had a psychiatric diagnosis, including depressive illnesses (n = 43, 8%), anxiety and stress disorders (n = 20, 4%), alcohol misuse/dependence (n = 18, 3%) and personality disorder (n = 3, 0.6%). It was not our intention to compare self-harm between diagnostic groups in this study. Instead, we aimed to investigate factors that distinguish self-harm in BD from self-harm in the general population, which will include a proportion of people with psychiatric diagnoses. We believe that our study has potentially important implications for clinicians encountering a person with BD who has self-harmed. As the majority of controls (> 80%) had no recorded psychiatric diagnosis, it could be that some of the findings were not specific to those with BD but reflected wider risks for individuals with mental illness. However, a post hoc sensitivity analysis showed that the HR for repeat self-harm in those with BD compared with controls with a psychiatric diagnosis remained elevated, with a HR of 2.40 (compared with a HR of 3.08 for the whole sample).

Study 4

Summary of thematic analysis

  • Service users’ and carers’ primary concern is fast access to good-quality mental health care.
  • Help-seeking is common, but barriers, such as gate-keepers, are also common.
  • Correct diagnosis is essential to enable correct treatment, and SUs/carers should be given information about the diagnosis to help understand risks.
  • Diagnosis and treatment is facilitated by building trust and maintaining continuity of care.
  • Care should be collaborative, when possible, to minimise conflict and utilise the carer’s greater knowledge of the SU’s symptoms and circumstances.

Overall, SU and carer experiences of MHS were a mix of positive and negative; however, most participants described at least one occasion when they had had difficulty accessing care during times when the risk of suicidal behaviour was increased. This was generally the focus of the interviewees’ concerns, rather than any illness-related factors or behaviours of the SU. The importance of gaining a correct diagnosis was emphasised, as this seemed to facilitate access to care, and lack of diagnosis could contribute to a ‘merry-go-round’ of suicidal behaviour and missed opportunities to intervene. At least part of the cause of this pattern of behaviour identified by SUs was that being able to vocalise their symptoms was taken as proof of the lack of severity of those symptoms, leading to a situation in which the SU felt that they had to ‘scream’ for help.

Once the SU is in contact with MHS, the key issue, particularly for carers, is that care is as collaborative as possible – assuming that the SU gives permission for information about their care to be shared. There are multiple points in the flow of information at which details about care, treatment and SUs’ symptoms can become blocked, and vital information that could help prevent suicidal behaviour could be missed. Carers were keen to be included and viewed as collaborators in care, with many participants indicating that they were unable to pass on information to the care team when they became worried about changes in the SU’s symptoms. Keeping the same care team and building trust over time went a long way towards reducing these experiences. In general, mental health care was viewed more positively, regardless of outcome, when clinicians were seen as sympathetic and interested in involving carers.

Most of the factors identified as important to SUs and relatives in the narratives are experiences that may be more generally related to problems with mental health care, such as the importance of working collaboratively and maintaining continuity of care over time, and could apply equally to other high-risk psychiatric groups as to people with BD. However, this does not diminish the importance of these issues or their impact on the SU or carer, and it does not rule out the possibility that these problems are over-represented among people with BD. Some themes identified are likely to be more specifically related to a diagnosis of BD; for example, difficulties in establishing a diagnosis may be particularly relevant to this group, as BD is often initially misdiagnosed as unipolar depression, or symptoms are confused with cluster B personality disorders. Likewise, some of the barriers to accessing MHS raised by participants might be encountered more often by people with BD. A SU’s ability to express distress and describe their symptoms was taken to mean they were not ill enough to require access to MHS. There is some evidence that people with BD are higher functioning than those in other high-risk psychiatric groups (such as those with schizophrenia295,296), and do not differ from general population samples in terms of educational attainment, which might contribute to their presentation being misinterpreted by services.297

Methodological limitations

Not all of those with BD who die by suicide will be in contact with services in the 12 months before death,298 and not all people who self-harm present to hospital following the act; therefore, the BD cases included in these studies may be a systematically different group from people who engage in suicidal acts but do not have contact with services. However, the quantitative studies compare BD cases only with others who also had contact with services. The differences identified are therefore reliable within those populations, and important, as these are the people who MHS are best positioned to identify and help. The qualitative work, however, was not restricted to people who were in contact with services at the time of suicide or self-harm and, therefore, the results may be more broadly generalisable.

A study from the Netherlands by ten Have et al.299 found the most common reason for lack of contact with MHS was lower severity of BD symptoms; therefore, our sample may be biased towards the more seriously ill. It would have not been feasible to collect detailed data on all individuals with BD who had not been in contact with MHS or medical services. The lack of generalisability is particularly relevant in regard to study 2, which specifically examined the most severely ill patients. However, it is encouraging that the findings in study 2 are consistent with the results of study 1, which included outpatients, and there is support in the study for the idea that those who are more severely ill are at higher risk for suicidal behaviour.

The data used in studies 1 and 2 were collected from clinicians after the suicide had occurred. Completion of the NCI suicide questionnaire relies on the accuracy of case notes and the clinician’s knowledge of the patient. The clinicians were therefore not blinded to patient outcome when completing the questionnaire, and it is possible that this could have influenced the results, particularly those relating to care or treatment before death. In study 3 the self-harm data were collected at presentation, or very soon thereafter, and so recall bias was less of a problem, although accuracy was still dependent on clinicians.

Allocation to the BD case-group was dependent on the recording of primary diagnosis by clinicians, and it is possible that, when there were multiple comorbidities, primacy was attributed incorrectly. In study 1, there were 89 cases with a secondary diagnosis of BD, all of whom had a primary diagnosis of schizophrenia. We adopted a conservative approach and included these in the ‘other primary diagnosis’ comparison group, which might have reduced the between-group differences. A post hoc sensitivity analysis showed that the inclusion of the 89 individuals in our BD suicide group had little effect on the results overall. Comorbidity might also be a concern in the self-harm study (study 3); however, only 11 (11%) cases had a comorbid psychiatric diagnosis recorded; depressive illnesses were most common (n = 5, 5%), followed by anxiety and stress disorders (n = 3, 3%), alcohol misuse (n = 2, 2%) and learning difficulties (n = 1, 1%). Nineteen controls (20% of the controls with a psychiatric diagnosis, 4% of all controls) were also recorded as having comorbid psychiatric diagnoses.

Diagnosis in the quantitative studies was assigned by clinicians according to ICD-10 categories, rather than standard interview measures, and misdiagnosis is possible. In study 3, the diagnosis was given by the clinician at the time of the assessment and reflects the patient’s mental state at the time in addition to historical information taken from the patient and medical records. When a specific diagnosis is unclear, the field will remain empty, and any diagnosis given is, therefore, more likely to be based on clear evidence, which may favour those with more pronounced symptoms and more severe illness.

In study 3, the number of cases with a diagnosis of BD was small in comparison with the size of the overall data set (< 1% of total presentations). This suggests that psychiatric diagnoses in general may be under-reported on the MaSH database. This is in line with previous reviews, which suggest that only a small proportion of self-harm patients receive a psychiatric diagnosis when they present to hospitals.300 This may be due to time pressures within the emergency department, the need for rapid assessment and turnover of self-harm patients, or a reluctance to assign a diagnosis on the basis of one assessment.

Study 4 took a qualitative approach, and the authors acknowledge that narratives were subjective and retrospective and might have been influenced by recall bias. Participants were self-selected volunteers, and this may have created selection bias within the sample, in that people who had very positive or very negative experiences with MHS might have been more motivated to participate. However, participants’ experiences of MHS were mixed overall, with participants reporting both good and bad aspects of care. Time constraints made it necessary to focus the analysis on the themes most pertinent to suicidal behaviour in BD, and other themes may be explored within this data set at a later time. Despite these limitations, these data represent a body of valuable information on a difficult-to-reach group (particularly in regard to the carers who had been bereaved by the suicide of someone with BD), and the lived experience of suicidal behaviour in BD, and may be important in illustrating where services might make improvements tailored towards this patient group.

Misdiagnosis could be a problem within BD cases, as it is known to be difficult to distinguish BD from diagnoses such as borderline personality disorder.261,301 However, this is true of all studies of BD that rely on past records and clinical judgement, and verifying diagnoses is beyond the scope of this work. The effect of bias from such missed or misallocated BD would be to minimise between-group differences, and it is reassuring that the differences seen in the findings across the quantitative studies are consistent with those of previous work.

It is possible that the clinicians completing the questionnaire in studies 1 and 2 were not the professionals with whom the last contact was made and, consequently, variables relating to clinical state might have been under-reported (e.g. evidence of depressive symptoms). However, when filling in the NCI questionnaire clinicians generally use patient records in addition to their personal knowledge and recollection of the patient, and it would be unusual for significant changes in mood state not to be recorded as part of the clinical record.

There are some potential issues with missing data on the NCI and MaSH databases and, therefore, the results presented should be interpreted with caution. The majority of variables were > 90% complete on the NCI suicide database and > 75% complete on the MaSH self-harm database. Patient treatment variables had greater levels of missing data than demographics. In most cases this would have been because the clinician left some fields blank, but some might be historical artefacts due to the evolving nature of the suicide questionnaires/self-harm collection forms, and the addition of new questions over time that cannot be back-coded for older cases.

Variables entered into the multiple regression in study 1 were not screened for correlations or specified a priori. Given the exploratory nature of this work and the expectation that correlations would fall below the commonly recommended 0.8 level, including all relevant variables in the analysis was intended to give as broad a picture as possible of the key differences between suicides by different diagnostic groups.

There are a number of psychological characteristics that are of current research interest in relation to self-harm and suicide in BD, such as temperament302,303 and impulsivity,294,304 that may be important in suicidal behaviour in BD, but they were unable to be explored as there was not sufficient detail within the quantitative data and so they were not directly addressed in the qualitative work.

Clinical and research implications

Suicide prevention is a priority for MHS and it is, therefore, important to be able to identify those at risk of suicidal behaviour.277 Knowledge of what characterises increased risk for specific diagnoses, rather than for psychiatric illness as a whole, might allow for improved identification of the most at-risk patients. The results of this study show that, although risk of suicide in BD is high compared with general population studies, it may be that some of the societal determinants of suicide have less of an impact in this clinical population, in whom illness-related factors may be more relevant. The possibility that individual clinical groups are not benefiting from an overall decline in suicide raises issues of equity, which warrant specific exploration in future studies. However, we also found that suicide is more common in unipolar depressive disorder and schizophrenia, which is surprising, and our results are likely to have underestimated the magnitude of this risk. For inpatients, among whom illness severity is commonly high, suicide rates for BD are similar to those in schizophrenia (1.7 vs. 1.8 per 1000 inpatient admissions by diagnosis).

In contrast to general population suicides in England, the number of bipolar suicides remained relatively stable over time. This could also be a consequence of the improved diagnosis of BD, and perhaps of BD 2 in particular, which might have caused an increase in BD suicides, as people who would previously have been diagnosed with unipolar depressive disorder are moved into the BD category.

Cohort studies of those with mental illness often describe the risk of suicide as highest at the time of diagnosis, with the risk subsequently diminishing over time.6,299 We found consistently that in these studies suicides took place at least 5 years after diagnosis. In addition, those with BD were more likely than those in the comparison groups to self-harm or die by suicide in middle age. These findings suggest that risk of suicide in BD may be relatively persistent. Study 3 confirmed that risk of suicidal behaviour is not restricted to the short term following self-harm: the average time to repetition approached 2 years in this study. Sixteen per cent (n = 13) of bipolar cases who were in previous contact with psychiatric services were recorded as not in current contact with services. As risk appears to be enduring, continued contact with services may be appropriate.

Although there were similar levels of comorbid alcohol dependence in the BD group and in those with other primary diagnoses, the BD group were less likely to be in contact with alcohol services, highlighting a possible deficiency in the provision of dual diagnosis services despite a current policy focus on improving services for people with psychiatric disorder and drug or alcohol misuse.305,306 Despite NICE guideline recommendations,1,58 over one-third of people with BD who died by suicide in this study were not prescribed lithium or another mood stabiliser, although there were some missing data for this variable. Given the increased risk of suicide that seems to accompany a diagnosis of BD, and the strong evidence for the protective effect of lithium,53,253,256 this might be considered a key treatment in long-term management.

A clinically important result from study 1 was that the majority of patients with BD were seen by MHS during the week before the suicide, and nearly one-quarter within the 24 hours immediately preceding it, providing a potential opportunity for services to intervene. However, 9 out of 10 people with BD who died by suicide were assessed as being at no or low risk at last contact. These data, along with those from the case–control study of recognised high-risk patients, suggest that clinical assessments may not be accurately identifying the BD patients who are most at risk. An alternative explanation is that patients with BD might be subject to rapid changes in mental state, consistent with the hypothesis that suicidal behaviour in BD is associated with impulsivity.253,289 Impulsivity might also explain why BD cases in study 3 were less likely to have ‘suicidal plans’, even though the proportion of those with suicidal thoughts was similar in both groups of self-harmers. The impulsivity of suicidal behaviour in BD may make it more difficult to predict when patients are most at risk. However, good-quality monitoring and management of symptoms and improved approaches to detection of risk – including assessing levels of impulsivity, which is variable within the BD population – could help.

It is important that health professionals are able to accurately identify those patients who might be most at risk of suicide. Factors such as alcohol dependence/misuse, personality disorder, depressive episode, multiple psychiatric admissions and current or recent admission could characterise such a high-risk group. In our sample of BD suicide deaths, 75% had one or more of these factors, 34% had two or more, 8% had three or more and < 1% had four. In terms of risk management, a reduction in the emphasis placed on traditional indicators of risk, such as social circumstances and recent diagnosis, and an increase in the emphasis on diagnosis-specific factors may increase the utility of assessments. Ideally, clinically useful indicators of suicide risk that can be operationalised need to be identified in future studies. However, the identification of a high-risk group with BD on the basis of factors associated with suicide in this sample would be premature, as two-thirds of completed suicide had fewer than two of these risk factors. More work is required to determine causes of suicide in BD and to explore how risk factors interact to precipitate suicidal behaviour.

The results of the case–control analysis confirmed the importance of so-called ‘conventional’ risk factors for suicide, such as self-harm, markers of depressive illness, social isolation, life events and alcohol misuse. Our secondary analysis identified factors that may have been more specific to the BD group, such as long duration of illness, and poor recognition of risk despite recent contact with services.

Previous self-harm is a strong predictor of future suicidal behaviour, and in the self-harm study (study 3) we found that repetition of self-harm was particularly common in individuals with BD: nearly 6 out of 10 repeated during the follow-up period. People with BD who self-harmed typically had a number of high-risk characteristics (e.g. previous self-harm, history of psychiatric care). However, these characteristics, recognised as risk factors in general and psychiatric self-harm populations, did not comprehensively explain the increased risk of repetition associated with a diagnosis of BD.

Sleep disturbance was a common and independent predictor of self-harm in the bipolar group. This may be a marker of changes in illness, agitation or mood-state, as found in mixed depressive episodes and known to be associated with self-harm in BD.262 Depressive symptoms were also evident in nearly 65% of the bipolar cases. We were not able to assess psychiatric comorbidity, severity or duration of illness in this study.

Given the frequency of previous self-harm and current/previous psychiatric care in the BD cases across studies, it is likely that those with BD who engage in suicidal behaviours could be distinguished from other disorders on markers of severity and chronicity (e.g. long duration of illness, multiple inpatient admissions, previous self-harm). The key precipitant that distinguished the BD group in the self-harm study (study 3) was ‘a direct response to psychiatric symptoms’, which indicates that illness-related factors may play a larger role in acts of suicidal behaviour in BD than any specific negative life events.

Many factors associated with suicidal behaviour in BD are known risk factors for suicide and self-harm in general and within the psychiatric population. However, those with BD seem to carry a particularly heavy burden of such risk factors – with very high levels of previous self-harm, multiple inpatient admissions, a long duration of illness, comorbid personality disorder and/or alcohol use – that goes beyond that of other groups. Identification of risk factors alone may not contribute much to clinical practice; however, in combination with results from across the other studies within the suicidality WS, this work provides new and valuable information about who might be most at risk and how services might better respond to suicidal behaviour in BD.

Clinicians need to be aware of the higher risk of self-harm in BD and the link between previous self-harm and suicide. Of course, these risks might not be restricted to those with BD, and those with other psychiatric disorders may also be at higher risk.

General preventative measures, such as good-quality assessment and interventions for people who self-harm, the treatment of depressive symptoms and prompt access to services after significant life events, are likely to help prevent suicidal behaviour in people with BD. However, more tailored approaches, perhaps focused on recognising the long-term risk that BD may confer, are also likely to be of benefit. The use of lithium, which has been shown to have specific benefits for prevention of suicide,256,307 and other mood stabilisers may be beneficial. Psychological approaches, such as CBTs, which have been shown to have a positive impact on suicidal behaviour in general, might also have an important role in prevention.308

Copyright © Queen’s Printer and Controller of HMSO 2018. This work was produced by Jones et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.
Bookshelf ID: NBK525333


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (240M)

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...