In a large clinical case register sourced from electronic mental health records linked to national mortality data, we identified a population of mental health service users with SSD who ended their lives and we investigated risk assessment differences between these individuals (cases) and controls (those who did not take their lives), while adjusting the analyses for potential confounders, including sociodemographic and clinical variables and service use-related factors. In line with our hypotheses, young age at first contact with mental health services, previous suicide attempts and suicidal ideation were associated with suicide. Owing to the rarity of the outcome and limited statistical power, suicide in SSD is not a predictable occurrence, with only 21.5% of the variance explained by the final regression model, yet older age at first contact with mental health services and lack of suicidal history and suicidal ideation are useful protective markers indicative of those less likely to end their own lives.
Comparison with previous research
Interestingly, suicidal cases were less likely to have a full risk assessment documented, which was somewhat contradictory to previous literature.29 However, we found a significant time trend in an increase of suicide risk assessment completion rates over the study years, which suggests that this difference may have been due to the longer care received by controls. Indeed, those patients with a risk assessment completed had been under the Trust teams for a longer time than those participants without risk assessment. Also, a full risk assessment may have been completed due to concerns raised regarding other risks such as violence and/or self-neglect.36 Moreover, recording of risk assessment has been reported to be, to some extent, circular. Specifically, some data from risk assessments following self-harm are more likely to be recorded if episodes result in a specialist assessment.39
In line with our hypotheses we replicated the role of suicidal history, suicidal ideation, previous use of a lethal method5,12,16,40,41 and recent hospital discharge2,40 in suicide risk. However, no associations of recorded hopelessness, impulsivity, alcohol/drugs misuse, living alone or significant losses with suicide were found in line with some previous literature on suicide and psychosis.42,43 Moreover, most participants with SSD who ended their lives (cases) did not have the factors evaluated by the risk assessment with the exception of ‘suicidal history’. Therefore, suicide in SSD may represent a challenge to the classic suicide model,44 particularly regarding the role of hopelessness44,45 and impulsivity44,46 in suicide, although this finding may be due to the relatively high rates of such factors in SSD per se therefore making it more difficult to ‘pick them out’ against the background. Furthermore, the classic psychosocial factors evaluated by suicide risk assessment instruments were found to be more relevant for patients without SSD who died from suicide.21 This finding is consistent with a previous comparison study between suicide attempters with schizophrenia and depression which showed that attempters with schizophrenia had a lower number of life events which were also less influential on the attempt.47
Of note, a number of suicides occurred shortly after having been seen by a member of staff, which is in keeping with previous studies showing the relative inability of clinicians to predict and/or prevent imminent suicide risk in individuals with SSD under their care40,48,49 despite risk assessment.
In addition, the ROC curves showed that overall risk assessment total scores performed poorly in terms of sensitivity, specificity and positive predictive value, while the test had a very high negative predictive value, which is in full agreement with a recent systematic review of risk assessment scales for predicting repeat self-harms in suicide attempters.28 However, at a service level the use of risk assessment tools in NHS-funded hospitals in England was associated with a lower incidence of repeat self-harm at 6 months.50
In summary, although suicidal history, suicidal ideation, previous use of a violent suicide method and recent hospital discharge were significantly associated with risk of suicide, the regression models correctly classified most of controls and just a very small proportion of cases, which is in line with our hypotheses and a recent editorial.30
Those participants with SSD who died by suicide (cases) had their first contact with mental health services at a younger age than controls, which also remained significant in the multivariable regression models, in line with previous studies showing an increased suicide risk in early psychosis,4 although late illness onset reduced suicide risk in an epidemiological study from Taiwan.50 Regarding gender, we replicated previous findings from samples of patients with schizophrenia of higher risk of suicide in men,12,16,40,41,51 which is also consistent with previous psychological autopsy studies in both psychotic and some, but not all, non-psychotic psychiatric populations.52
Interestingly, neither being unemployed nor unmarried was associated with increased suicide risk, consistent with a previous meta-analysis in this diagnostic group.12 Also, despite previous literature showing a relationship between ethnicity and suicide in patients with schizophrenia12 and in the general population,53 we found no evidence of this. This lack of differences in sociodemographic characteristics between cases and controls may have been due to insufficient statistical power. However, it should be noted that the vast majority of patients with psychosis living in South-East London tend to be single, unemployed, socially deprived and of black ethnicities.54,55 Moreover, a previous population-level community-based survey conducted in South-East London found poor mental health, low socioeconomic status and being unmarried to be associated with a history of previous suicide attempts, while black African ethnicity was protective.56 Although we failed to replicate these findings in our sample of patients with schizophrenia under secondary mental healthcare who took their lives, our previous comparison study of participants under secondary mental healthcare who died by suicide had reported significantly higher levels of social deprivation in patients with SSD than in those with non-SSD diagnoses.21 Therefore, there are grounds to consider that low socioeconomic status, which is strongly linked with psychosis in our catchment area,54,55 may increase suicide risk in these patients.
Service use-related factors
Of note, cases tended to have a shorter interval than controls between the referral and first contact with services. This finding is in full agreement with a previous first-episode psychosis study which showed that suicidal behaviour preceding first contact with services may shorten the duration of untreated psychosis via leading the patient to receive psychiatric attention earlier.57 Also, the duration of care was significantly shorter in those who died from suicide (cases), which is likely to have been an artefact due to the survival effect among controls. We replicated the high risk of suicide in patients with schizophrenia during the immediate period after hospitalisation.2,16,40,41,48 A third of suicides occurred in the 6-month period after being discharged from a psychiatric ward, which is of particular concern in patients with a first-episode psychosis.58 Hence, close monitoring over that period of time should be strongly recommended.2,14
Despite the well-known antisuicidal properties of clozapine59,60 and some previous recommendations promoting the use of antidepressants41,61 and depot antipsychotics41 for suicide prevention in schizophrenia, we found no significant differences between cases and controls in the use of these pharmacological treatments. However, our negative results were purely observational and consistent with previous case–control studies in schizophrenia,16,51 which all may have had insufficient statistical power.
Receiving community mental healthcare under restriction in accordance with the UK Mental Health Act 1983 (Amended 2007),37 which is known as a CTO, showed no association with suicide. These findings seem to be consistent with the UK National Confidential Inquiry into Suicide and Homicide report,2 according to which there were 42 suicides in patients subject to a CTO between 2009 and 2013 in England. The suicide rate in CTO patients (2.0 per 1000 CTOs in 2009–2012) was higher than the suicide rate for all patients (0.09/1000-year), and this is not unexpected, since one criterion for selection of patients for CTO is risk on discharge.
Regarding suicide method, we replicated that hanging and jumping (from a height or in front of a vehicle) were the most common suicide methods5,16,20,24,40,42,62 in SSD. However, patients with SSD were reported to kill themselves by taking overdoses in Finland.48 Of note, no suicides by firearms were identified in our study, which is in line with previous reports from Europe,4,15 reflecting the restrictions to firearms compared with the USA.
Limiting availability of lethal methods has been demonstrated to reduce suicide rates at a population level.22,24 Also, restricting access to suicide hotspots such as heights through safety barriers23 and railway lines by installing platform edge doors63 has been reported to reduce overall suicide rates at such places.64 Hence, installation of physical barriers on bridges, tall buildings and railway stations, particularly near psychiatric hospitals given our replication finding concerning the increased suicide risk after hospitalisation2,16,40,41,48,58 may prevent patients with SSD from suicide.21
Strengths and weaknesses
This study focused on the rare outcome of suicide. By using a large case register linked to national mortality data, all those patients with a diagnosis of SSD who were receiving secondary mental healthcare in our catchment area and died by suicide over 2007–2013 were included in the study with the only exception of those who ended their lives outside the UK. Most patients were followed up over a prolonged period (median=6.17 years). As only a tiny proportion of patients living in South-East London receive private mental healthcare, our sample is likely to be representative. In addition to risk assessment, a wide range of demographic and clinical variables, including service use-related factors, were analysed.
However, our results should be considered in the light of several limitations. First, the sample was formed of secondary mental health services users living in South-East London, an inner urban area, and results may not generalise to people receiving mental health input from primary care or those in rural areas. Second, risk assessment ratings were unavailable for a number of patients. Also, we can speculate that those patients who had a risk assessment completed were deemed ‘at high risk’ by their clinical teams. Hence, the likelihood that these measures and findings relating to speed of assessment reflect the clinical response to perceived suicidal risk rather than potential predictors. In addition, although just the last risk assessment was considered, risk factors evaluated by this instrument may have changed from that point to death. Also, it should be noted that a wide range of variables have been taken over a prolonged period of time, which also varies across the study patients, who ranged from having one single assessment to several years under secondary mental healthcare, thus reflecting the real-world nature of our data. In addition, other non-tested variables such as premorbid personality and premorbid adjustment may have contributed to suicide in our sample. Finally, our control recruitment method explained above did not include an algorithm for randomisation of cases and controls.