Aims: The aim of this study was to identify predictors of completed suicide in a wide sample of psychiatric inpatients receiving retrospective and prospective DSM-IV diagnoses.
Methods: We followed up 4441 severe psychiatric patients who were hospitalized for some time during a 35-year period in a private hospital setting. We collected sociodemographic, clinical and temperamental data.
Results: Ninety-six patients from the sample committed suicide. There were no sex differences in suicide completion and no differences between major psychiatric disorders, but people who had been hospitalized for anxiety disorders did not commit suicide and people with bipolar disorders were more likely to commit suicide than people with unipolar major depression. Shorter-term treatment with lithium and anticonvulsants, longer-term treatment with antidepressants, history of suicide attempts, suicidal thinking, and single status positively predicted completed suicide. Suicide tended to occur after a mean period of about 14 years of duration of disease. Patients' symptoms during the period preceding suicide were assessed through interviewing patients' physicians or family members. Symptoms occurring in >10% of cases were, in decreasing order, inner tension, racing/crowded thoughts, aggressive behavior, guilt, psychomotor agitation, persecutory ideation, anxiety, and hallucinations. Surprisingly, cyclothymic temperament was less associated with completed suicide as compared to other temperaments.
Conclusions: Suicide is likely to occur in a milieu of agitation, mixed anxiety and depression, and psychosis. Longer-term mood stabilizer treatment may reduce the rate of completed suicide.
SUICIDE IS A significant public health issue, accounting for about 1.5% of the global health-care burden and more than 20 million disability-adjusted life-years.1 Annual suicide deaths increased across years and are estimated to reach 1.2 million people by 2020.2 Annual rates vary across countries.2,3 In Italy (about 60 million residents), there were 2867 suicide cases in 2007, a 4.77 per 100 000 annual rate.4 More than 90% of suicides in various studies had a DSM-IV diagnosis, but lesser proportions are found in some countries.5 Mood disorders, witnessed by the presence of major depressive episodes, are associated with 56–87% of suicides.6
Given the major contribution of mental illness in suicide, investigations focused on the possible preventive or inductive role of psychiatric drug treatment. Generally, it is held that stabilizing treatment is protective, evidence for antipsychotic treatment is weak, while the role of antidepressant treatment is controversial. However, so far, only the mood stabilizer, lithium,7–12 and the atypical antipsychotic, clozapine,13 were found to be reliably associated with reduction in suicide rates. The beneficial effect of lithium appears to be independent from diagnosis and clinical effectiveness,14,15 and occurs also in recurrent major depressive disorder.16 The evidence for a specific antisuicidal effect of clozapine comes mainly from patients with schizophrenia or schizoaffective disorder,13,17,18 and from a large British register,19 but cannot be considered conclusive.20
Antidepressants may reduce suicide rates by reducing depression, a known risk factor for suicide, but they have also been reported to be associated with increased suicidality; tricyclic antidepressants were linked to suicidality since the early imipramine trials,21 and later this was extended to the selective serotonin reuptake inhibitors (SSRI) and other antidepressants, which do not differ in their suicide hazard ratio.22 The American Food and Drug Administration issued a black-box warning in 2004 for all antidepressants, concerning suicide ideation and treatment of pediatric mood disorders. Data in adults are not clear, but there is evidence for a higher suicidality risk in young adults and a lower one in aged people with antidepressants.23 Despite authoritative reviews of the issue24 and considerable debate,25 positions are quite distant, supporting, respectively, that increased use of antidepressants in entire populations within specific geographic areas is associated with reduced suicidality, and that antidepressants may increase suicidality in specific age ranges compared to placebo.
It is long held that antidepressants could increase suicidality by increasing agitation, although there is no such evidence from clinical trials of depression.26 Interestingly, Kenchadze et al.27 found an increased suicide risk for patients with agitated depression using the Mind Over Mood Anxiety Inventory.28 However, all evidence connecting antidepressant use to agitation and the latter to increased suicidality is largely anecdotal.
Our hypothesis was that psychic/motor agitation and psychotic symptoms would bear a relationship with suicide risk and that treatment patterns and psychopathology could modulate this relationship. To test this hypothesis, we collected and analyzed the longitudinal course, drug treatment, and clinical characteristics of a sample of psychiatric inpatients who committed suicide.
Our sample included a large number of inpatients from the Belvedere Montello Hospital, Rome, Italy, admitted during the years 1964–1998. The hospital, which is run by one of the authors (At.K.) and his medical team, is a private psychiatric institution specializing in the treatment of mood disorders. All clinical data were collected directly from patients, accompanying family members, and past available medical records at admission, and were conveyed in new, structured, medical records. Free, informed consent was obtained from patients as soon as the relative legislation was enforced. At.K and D.R. followed up all patients.
A total of 4441 patients had been hospitalized at Belvedere Montello, of whom 2390 (53.8%) were women and 2051 men (46.2%; male-to-female ratio [M/FR] 0.86). Among them, 96 (2.2%) committed suicide; of these, 43 were women and 53 were men, with M/FR 1.44. Patient diagnoses were: bipolar I disorder 1163 (26.2%, M/FR 0.7; of these, 32 committed suicide [2.8%], M/FR 2.09); bipolar II disorder 602 (13.5%, M/FR 0.73; 25 suicides [4.2%], M/FR 1.27); major depressive disorder 1142 (25.8%, M/FR 0.45; 22 suicides [1.9%], M/FR 1.54); major depressive episode 153 (3.5%, M/FR 0.49; one suicide [0.75%], a man); schizophrenia 143 (3.3%, M/FR 0.99; five suicides [3.5%], M/FR 4.06); schizoaffective disorder 122 (2.7%, M/FR 3.06; seven suicides [5.7%], M/FR 0.82); acute psychotic episode 132 (3%, M/FR 2.3; one suicide [0.75%], a man); substance abuse 639 (14.5%, M/FR 3.18; three suicides [0.5%], M/FR 0.66); and major depressive disorder 1142 (25.8%, M/FR 0.45; 22 suicides [1.9%], M/FR 1.54). For obsessive–compulsive disorder (n = 111, 2.5%; M/FR 0.63), panic disorder (n = 71, 1.6%; M/FR 3.44), generalized anxiety disorder (n = 61, 1.38%; M/FR 0.63), and other diagnoses groups (n = 102, 2.3%; M/FR 0.67) there were no patients who committed suicide.
During the 35-year activity of the Belvedere Montello Hospital, there were 9041 admissions for 4441 patients (2390 women, 54%). Diagnoses are shown in Table 1. Suicide occurred in 2.2% of patients during an average post-discharge period of 2.9 ± 4.2 (0.1–27) years.
Table 1. Diagnoses of 4441 patients and 96 suicide patients
|BPI||1163 [26.2]||58.8||32 [2.8]||13 [1.9]||19 [4.0]||2.09|
|BPII||602 [13.5]||57.8||25 [4.2]||13 [3.7]||12 [4.7]||1.27|
|MDD||1142 [25.8]||68.9||22 [1.9]||13 [1.7]||9 [2.5]||1.54|
|MDE||153 [3.5]||67.3||1 [0.7]||0 [0.0]||1 [2.0]||–|
|Schi||143 [3.3]||50.3||5 [3.5]||1 [1.4]||4 [5.6]||4.06|
|SA||122 [2.7]||24.6||7 [5.7]||2 [6.7]||5 [5.4]||0.82|
|APE||132 [3.0]||30.3||1 [0.8]||0 [0.0]||1 [1.1]||–|
|SUD||639 [14.5]||23.9||3 [0.5]||1 [0.7]||2 [0.4]||0.66|
|OCD||111 [2.5]||61.3||0 [0.0]||0 [0.0]||0 [0.0]||–|
|PD||71 [1.6]||22.5||0 [0.0]||0 [0.0]||0 [0.0]||–|
|GAD||61 [1.4]||45.9||0 [0.0]||0 [0.0]||0 [0.0]||–|
|Other||102 [2.3]||59.8||0 [0.0]||0 [0.0]||0 [0.0]||–|
|Total||4441 ||53.8||96 [2.2]||43 [1.8]||53 [2.6]||1.44|
The patients were diagnosed according to the Research Diagnostic Criteria (RDC)29 during the first period, and their diagnoses were progressively updated to the DSM-IV-TR criteria,30 except for agitated depression, for which we applied RDC. We felt compelled to do so because the DSM-IV has no specific criteria for agitated depression, and psychomotor agitation plus major depression are not sufficient to characterize agitated depression as a subtype.31 Premorbid temperament was assessed using Akiskal and Mallya's criteria for mood disorder patients.32 For patients admitted before 1987, these criteria were used retrospectively.
Medical records contained extensive medical and psychiatric history of patients prior to hospitalization, daily clinical notes regarding treatment, medical and psychopathological assessments with particular focus on mood, and outcome at discharge. We collected data for patient status at the time about which they committed suicide through either psychological autopsy, interviewing survivors, and/or through telephone interviews of their doctors. Autopsies were carried out through face-to-face interviewing significant others and/or treating physicians. Interviews lasted about 4 h each; narratives were averaged when informants were more than one.
A panel of expert psychiatrists (A.K., L.T., P.G., R.T.) analyzed data from those patients who committed suicide at any time after hospitalization at the Belvedere Montello Hospital and compared them with a group from the same population, consisting of patients who were hospitalized at the Belvedere Montello Hospital, but did not commit suicide. We matched the two groups by sex, diagnosis, and year at first hospitalization, at the ratio of 1:2.
The study received the approval of the ethical committee of the Sant'Andrea Hospital–Sapienza University, Rome, Prot. C.E. 960/09.
We compared the effect of treatment duration on suicidality by post-hoc subdividing the sample into a longer-term (defined as lasting at least half of the total time since illness onset) and a shorter-term (lasting less than half of the total time from illness onset) treatment groups.
We employed contingency tables for categorical comparisons (χ2 or Fisher's exact), odds ratios (OR) for proportions, and the Student's t-test for continuous variables. Data are shown as means ± SD or as frequencies (percentage [%]). We set statistical significance at P < 0.05, two-tailed. Analyses employed commercial computer programs: stata 8.0 (stata Corp., College Station, TX, USA), and Statview 5 (sas Corp., Cary, NC, USA).
Suicide rates and diagnosis
Of the entire sample of 4441, 83.3% of patients had been diagnosed with a mood disorder, specifically, 40% with bipolar disorder type I (BPI) of whom 59.4% were women; 31.2% with bipolar disorder I (BPII), 52% women; and 28.5% with major depressive disorder (MDD), recurrent or single episode (MDE), 56.5% women. Among patients with no mood disorders, 13.5% had a psychotic disorder: schizoaffective disorder (SA) was found in 53.8%; schizophrenia (SZ) in 38.5%; and acute psychotic episode in 7.7%. Substance use disorder (SUD) was diagnosed in 3.1%. No suicide occurred in patients admitted for any anxiety disorder.
Regarding the entire sample, comparisons between BPI and BPII showed no differences in risk for suicide (χ2 = 2.49; P = 0.114). Patients with BPI and BPII showed a 1.94-fold higher risk than patients with MDD and MDE (χ2 = 6.58; P = 0.01). Suicide rate in patients with schizophrenia did not differ from that of patients with SA (Fisher's exact test: P = 0.395). Suicide rates in patients with BP were not different from those of patients with psychotic disorders, including SZ, SA, and acute psychotic disorder (Fisher's exact test, P = 0.395). The rank by diagnosis of male-to-female ratio was schizophrenia > BPI > MDD > BPII > SA > SUD (Table 1). Among the 16 patients with substance-related disorder (cocaine, hashish, heroin or alcohol abuse), 13 had one comorbid psychiatric disorder; in particular, six had MDD, four had BPII, two had BPI, and one had SA.
Characteristics and clinical course of suicidal versus non-suicidal patients
Among the 96 suicidal patients, 52% were women. Patients who committed suicide versus non-suicidal controls were more likely to be unmarried (single, divorced or widowed 61.4% vs 24.2%; χ2 = 24.2; P < 0.0001), more likely to have a longer illness duration (78.1% vs 55.7; χ2 = 13.8; P < 0.001), to have more lifetime suicidal ideation (75.0% vs 58.3%; χ2 = 7.71; P = 0.006), and, as expected, to have more frequent past suicide attempts (46.8% vs 20.3%; χ2 = 21.9; P < 0.0001). However, suicidal patients did not differ from non-suicidal patients in educational level, family history of psychiatric disorder or suicidal behavior, type of first lifetime episode, age at onset, and number of hospitalizations (Table 2). Moreover, while dysthymic temperament was associated with a higher suicide risk (OR = 1.35, 95% confidence interval [CI][0.72–2.53]), anxious (OR = 0.50; 95%CI [0.24–1.03]), hyperthymic (OR = 0.62; 95%CI [0.38–1.01]), and cyclothymic (OR = 0.51; 95%CI [0.30–0.88]; P < 0.05) temperaments were associated with significantly less likelihood to commit suicide compared to the former (Table 2).
Table 2. Characteristics of patients who committed suicide versus those who did not
|n||96||192|| || |
|Female sex||43||86||[Matched]|| |
|Single (n[%])†||59 [61.4]||59 [30.7]||24.2||P < 0.0001|
|Education >8 years (n[%])‡||80 [83.3]||162 [84.4]||0.05||P = 0.82|
|Family history of psychiatric disorder (n[%])||53 [55.2]||115 [59.3]||0.58||P = 0.45|
|Family history of suicide/SA (n[%])||9 [9.3]||10 [5.2]||1.80||P = 0.18|
|First episode (n[%])|| || || || |
| Depressive||62 [64.5]||120 [62.5]||0.18||P = 0.70|
| Manic§||18 [18.7]||40 [20.8]|| || |
|Age at onset (n[±SD])||27 [14.9]||26.4 [11.6]||0.38||P = 0.71|
|Morbidity indexes (n[%])|| || || || |
| >50% duration of illness||75 [78.1]||107 [55.7]||13.8||P < 0.001|
| >5 hospitalizations||23 [23.9]||65 [33.8]||1.39||P = 0.24|
|Suicidal ideation (n[%])|| || || || |
| Present||72 [75.0]||112 [58.3]||7.71||P = 0.006|
| Absent||24 [25.0]||80 [41.7]|| || |
|Previous SA (n[%])||45 [46.8]||39 [20.3]||21.9||P < 0.0001|
|Temperament (n[%])¶|| || ||OR||95% CI|
| Anxious||11 [13.7]||22 [18.3]||0.50||[0.24–1.03]|
| Cyclothymic||20 [25.0]||39 [32.5]||0.51||[0.30–0.88]|
| Dysthymic||23 [28.7]||17 [14.2]||1.35||[0.72–2.53]|
| Hyperthymic||26 [32.5]||42 [35.0]||0.62||[0.38–1.01]|
Effects of treatments
All patients were receiving psychiatric care when they committed suicide. Longer-term treatments (defined as lasting at least half of the total time since illness onset) with lithium (χ2 = 34.5; P < 0.0001) and anticonvulsants (χ2 = 4.33; P = 0.037) were significantly more likely to be associated with lower suicide risk than shorter-term (lasting less than half of the total time from illness onset). In contrast, longer-term treatment with antidepressants was associated with significantly higher suicide risk than shorter-term (χ2 = 8.22; P = 0.004). No significant differences were found for longer- or shorter-term treatments with antipsychotics. Longer- or shorter-term treatments with anticonvulsants, antidepressants, or antipsychotics affected the composition of the two groups (suicide completers vs others) not differently from lithium (Table 3).
Table 3. Shorter- and longer-term treatments in patients dying by suicide vs others†
|Shorter- vs Longer-term|| || || || |
| Shorter-term lithium||41 [70.7%]||38 [26.2%]||34.5||P < 0.0001|
| Longer-term lithium||17 [29.3%]||107 [73.8%]|| || |
| Shorter-term AC||20 [80%]||39 [56.5%]||4.33||P = 0.037|
| Longer-term AC||5 [20%]||30 [43.5%]|| || |
| Shorter-term AD||11 [42.3%]||35 [77.7%]||8.22||P = 0.004|
| Longer-term AD||15 [57.7%]||11 [22.3%]|| || |
| Shorter-term AP||10 [100%]||28 [71.7%]||–||P = 0.090‡|
| Longer-term AP||0||11 [28.3%]|| || |
|Treatment§|| || || || |
| Lithium||58 [49.6%]||145 [48.5%]||–||–|
| AC§||25 [21.7%]||69 [23.1%]||0.12||P = 0.72|
| AD§||26 [22.2%]||46 [15.4%]||1.42||P = 0.23|
| AP§||10 [8.54%]||39 [13.0%]||1.34||P = 0.25|
Characteristics of suicide
In the suicide group, people had been ill for 14.5 years (mean; range 1–52, SD = 10.2). Five patients killed themselves during hospitalization, two women the same day of their discharge, and one woman when she was in jail after discharge. The first three most common methods to commit suicide were defenestration in 37.5% of the cases, medication overdose in 19.7%, and hanging in 12.5%. Burning, acid ingestion, gas inhalation, poisoning, and self-suffocation contributed one case each (1%).
When they committed suicide, 31.2% of patients had an agitated depressive episode, 23.9% a major depressive episode, 19.7% a depressive episode with psychotic features, 15.6% a psychotic episode, 6.2% a dysphoric manic episode and 3.1% a manic episode with psychotic features.
The week preceding their death, 74% of patients had manifested suicidal thinking. Psychopathological features of the patients at the time of suicide included inner tension in 54.2%, guilt in 23.4%, persecutory ideation in 12.5%, feelings of worthlessness in 6.2%, grandiose ideation in 2.1%, aggressive behavior in 31.3%, psychomotor agitation in 30.2%, racing/crowded thoughts in 53.1%, anxiety in 12.5%, and hallucinations in 10.4% (Table 4).
Table 4. Psychopathology in the suicidal sample
This naturalistic retrospective study provided data on suicide from a cohort of 4441 inpatients followed up for up to 35 years by the same attending psychiatrists (At.K and D.R.).
The estimated annual global suicide rate is 17.7 per 100 000 for men and 10.7 per 100 000 for women;33 this is about one-tenth the figure we obtained in our psychiatric patient population. The male-to-female ratio in the general population is 1.65, while in our sample, the male over female preponderance was somewhat weaker (1.44). The fact that our sample mainly consisted of patients with mood disorders may have dampened the male preponderance seen in the general population, in agreement with other reports.34,35
A striking finding was that no patient with any anxiety disorder without comorbidity for other conditions committed suicide in our cohort, despite the claim of an increased suicide risk in anxiety disorders.36,37 People with anxiety disorders in very large cohorts, which are five times the size of ours, committed suicide 6–7 times less than patients with severe psychotic, depressive or comorbid disorders.36 Patients with anxiety disorders were few in our sample, compared to patients with BP and schizophrenia, hence, the proportions did not reflect those of the general population. Furthermore, patients with anxiety disorders are seldom hospitalized compared to other diagnoses, hence we might conclude that they were more severe cases. However, it is possible that people with anxiety disorders in our inpatient sample were insufficient to detect even the lowest suicide rates.
We found that 78.1% of the patients committed suicide during an episode of depression with or without agitation, similar to the proportion reported by others.11,38 However, we should remark also that three-quarters of all suicides were associated with psychotic/agitated states. It is tempting to hypothesize that mounting energy and impulsiveness, typical of similar states, facilitated suicide in these patients,39 but our data do not allow adequate testing of this speculation. This finding parallels some other evidence regarding the relationship between mixed affective states and, in particular, agitated depression, and suicide.39–45 Interestingly, suicide occurs most frequently during the mixed phase of BPI and BPII,46 when agitation is often present. Symptoms of people who died by suicide in our sample and occurred in more than 10% of the sample comprise all characteristics of mixed states, as well as positive symptoms like delusions and hallucinations (Table 4). It is interesting, however, that delusions were most frequently related to affective issues, with the combination of guilt, worthlessness, and grandiosity largely exceeding persecutory ideation.
Our data are in line with the finding of a reduction of suicide risk associated with longer-term lithium treatment,7–9,11,12,16 and anticonvulsants, whereas longer-term antidepressants were associated with increased suicide rates, compared to shorter-term use. We have no reason to explain the latter, but many factors are likely to participate, such as tachyphylaxis and non-adherence. Antipsychotic intake pattern did not appear to influence suicide rates; however, few patients were longer-term users and all antipsychotic prescriptions in our population regarded classical neuroleptics and not atypical antipsychotics. We may not infer on the characteristics of patients in our population who are still alive and did not attempt suicide, but it is interesting that about 50% of the people discharged from the Belvedere Montello were receiving lithium. A protective effect for lithium has been advanced with regard to suicide7–9,11,12,16 and there is evidence that lithium has a general protective effect from mortality;47 an independent effect from its stabilizing activity is also suggested from a recent study showing that the presence of lithium in drinking water, even in trace amounts, is associated with reduced suicide rates,48 hence prompting for the addition of lithium to drinking water to improve public mental health.49
As far as clinical course is concerned, there are some findings to underline. Mean time from onset to suicide was 14.5 years, which is comparable to that of patients with schizophrenia.50 Other investigators11,34 found that about 25% of suicides occur during the first years of illness and about 50% in the first five years, but a recent paper reported latencies similar to ours.51 It is possible that suicide is a complex behavior in which impulsiveness may sometimes play an important role. Suicidal ideation may be present right from the start of a psychiatric disorder or appear at any moment with no immediately apparent consequences but may recur throughout the illness. Suicidal ideation should never be neglected as many patients commit suicide after life-long suicidal thinking.51 Although some people who have never manifested suicidal thinking eventually kill themselves (24 in our series, 25% of all suicides), most suicide victims (72 people, 75%) had manifested suicidal thinking. Many who did manifest it still did not (112, 58.3% of those who did not commit suicide), whereas most people who commit suicide manifest suicidal thinking (72 in our case series, 75% of all suicides) and those who do not, usually do not commit suicide (80, 41.7% of those who did not commit suicide, but 76.92% of those with no suicidal thinking). Almost half (n = 45, 46.8%) of suicidal patients had attempted suicide at least once, whereas among patients who did not commit suicide, only one-fifth (n = 39, 20.3%) had attempted it.
Cyclothymic temperament was significantly associated with lower suicide rate than anxious, dysthymic and hyperthymic temperaments, as assessed through Akiskal and Mallya's criteria.32 We found no ‘protective’ effect for the hyperthymic temperament, as found by others in suicidal psychiatric inpatients46,50 or non-violent suicide attempters,51 who used the Temperament Evaluation of Memphis, Pisa, Paris and San Diego autoquestionnaire.52 It is surprising that we found only the cyclothymic temperament to be associated with decreased likelihood for completed suicide, as this temperament was found to be a predictor of suicidality in a population of children and adolescents with a major depressive episode.53 Methodological differences could account for the discrepancies between our results and those of the above-cited investigators and it is to be underlined that our sample compared people who completed suicide versus those who did not, whereas all other studies mainly assessed suicidality.
Our patients who died by suicide were followed up closely, either by the first four authors or by other psychiatrists and primary care physicians, and were in frequent contact with the health-care system services. In contrast, a US population of veterans with depression was mainly seen in mental health settings.54
Our findings may enhance the ability of clinicians to detect individuals at risk of suicide and institute appropriate preventive therapeutic interventions.
The main limitations of this study are its naturalistic retrospective design, and the possible underestimation of suicide rates due to patient attrition, which might have been influenced by the long period taken into consideration. Another major limitation is that we did not use standardized assessment of clinical status and temperament. In fact, we mainly relied on psychological autopsy methods; however, these were carefully conducted by At.K and D.R., who also carefully followed up patients from their first hospitalization at the Belvedere Montello Hospital.
In conclusion, despite the limitations inherent in the study design, our results show that suicide in psychiatric patients is likely to occur in periods of agitation and inappropriate drug treatment. Although definite conclusions may not be drawn from our data, there is only one previous report that agitation is unrelated to suicidality,26 whereas most evidence is in line with our findings.55,56 Whatever the case might be, psychiatric patients should be closely followed up, paying particular attention to their mood, but also to psychomotor aspects of behavior. The current DSM-IV formulation of mood disorders does not allow for agitated depression to be adequately investigated, while the more restrictive RDC criteria do allow for better framing of this entity. The incorporation of some RDC-based criteria for agitated depression in the DSM-V could possibly overcome our prediction of suicidal events and promote better suicide prevention.
This paper has not received funds from any agency. The authors do not have an affiliation with or financial interest that might pose a conflict of interest for this paper. However, Leonardo Tondo has received sources of funding from clinical grants from NARSAD and the Stanley Foundation; in the past, Paolo Girardi has received research support from Lilly and Janssen and honoraria from Lilly and Organon, and has also participated on advisory boards for Lilly, Organon, Pfizer, and Schering; and Roberto Tatarelli has participated on advisory boards for Schering, Servier, and Pfizer and received honoraria from Schering, Servier, and Pfizer. Other authors declare no conflict of interest whatsoever.