Suicide risk scales: do they help to predict suicidal behaviour?

A Machine Learning Approach to Identifying Changes in Suicidal Language

OBJECTIVE

With early identification and intervention, many suicidal deaths are preventable. Tools that include machine learning methods have been able to identify suicidal language. This paper examines the persistence of this suicidal language up to 30 days after discharge from care.

METHOD

In a multi-center study, 253 subjects were enrolled into either suicidal or control cohorts. Their responses to standardized instruments and interviews were analyzed using machine learning algorithms. Subjects were re-interviewed approximately 30 days later, and their language was compared to the original language to determine the presence of suicidal ideation.

RESULTS

The results show that language characteristics used to classify suicidality at the initial encounter are still present in the speech 30 days later (AUC = 89% (95% CI: 85-95%), p < .0001) and that algorithms trained on the second interviews could also identify the subjects that produced the first interviews (AUC = 85% (95% CI: 81-90%), p < .0001).

CONCLUSIONS

This approach explores the stability of suicidal language. When using advanced computational methods, the results show that a patient's language is similar 30 days after first captured, while responses to standard measures change. This can be useful when developing methods that identify the data-based phenotype of a subject.

A Machine Learning Approach to Identifying the Thought Markers of Suicidal Subjects: A Prospective Multicenter Trial

Death by suicide demonstrates profound personal suffering and societal failure. While basic sciences provide the opportunity to understand biological markers related to suicide, computer science provides opportunities to understand suicide thought markers. In this novel prospective, multimodal, multicenter, mixed demographic study, we used machine learning to measure and fuse two classes of suicidal thought markers: verbal and nonverbal. Machine learning algorithms were used with the subjects' words and vocal characteristics to classify 379 subjects recruited from two academic medical centers and a rural community hospital into one of three groups: suicidal, mentally ill but not suicidal, or controls. By combining linguistic and acoustic characteristics, subjects could be classified into one of the three groups with up to 85% accuracy. The results provide insight into how advanced technology can be used for suicide assessment and prevention.

Estimating suicide occurrence statistics using Google Trends

Data on the number of people who have committed suicide tends to be reported with a substantial time lag of around two years. We examine whether online activity measured by Google searches can help us improve estimates of the number of suicide occurrences in England before official figures are released. Specifically, we analyse how data on the number of Google searches for the terms 'depression' and 'suicide' relate to the number of suicides between 2004 and 2013. We find that estimates drawing on Google data are significantly better than estimates using previous suicide data alone. We show that a greater number of searches for the term 'depression' is related to fewer suicides, whereas a greater number of searches for the term 'suicide' is related to more suicides. Data on suicide related search behaviour can be used to improve current estimates of the number of suicide occurrences.

ELECTRONIC SUPPLEMENTARY MATERIAL

The online version of this article (doi:10.1140/epjds/s13688-016-0094-0) contains supplementary material.

Royal Australian and New Zealand College of Psychiatrists clinical practice guideline for the management of deliberate self-harm

OBJECTIVE

To provide guidance for the organisation and delivery of clinical services and the clinical management of patients who deliberately self-harm, based on scientific evidence supplemented by expert clinical consensus and expressed as recommendations.

METHOD

Articles and information were sourced from search engines including PubMed, EMBASE, MEDLINE and PsycINFO for several systematic reviews, which were supplemented by literature known to the deliberate self-harm working group, and from published systematic reviews and guidelines for deliberate self-harm. Information was reviewed by members of the deliberate self-harm working group, and findings were then formulated into consensus-based recommendations and clinical guidance. The guidelines were subjected to successive consultation and external review involving expert and clinical advisors, the public, key stakeholders, professional bodies and specialist groups with interest and expertise in deliberate self-harm.

RESULTS

The Royal Australian and New Zealand College of Psychiatrists clinical practice guidelines for deliberate self-harm provide up-to-date guidance and advice regarding the management of deliberate self-harm patients, which is informed by evidence and clinical experience. The clinical practice guidelines for deliberate self-harm is intended for clinical use and service development by psychiatrists, psychologists, physicians and others with an interest in mental health care.

CONCLUSION

The clinical practice guidelines for deliberate self-harm address self-harm within specific population sub-groups and provide up-to-date recommendations and guidance within an evidence-based framework, supplemented by expert clinical consensus.

Development of the Risk Assessment Suicidality Scale (RASS): a population-based study

BACKGROUND

Suicide and suicide attempts are significant and costly public health problems. In order to prevent suicidal and other self-injurious behaviors, research on the multiple factors involved in these behaviors with comprehensive and user-friendly instruments is necessary. The aim of the current study was to construct a self-report instrument with emphasis on items describing suicide-related behavior itself rather than strongly related clinical features on the basis of a general population study.

METHODS

Twelve items comprising a new scale were applied to 734 subjects from the general population (40.6% males and 59.4% females) aged 40.8±11.5 along with the STAI and the CES-D.

RESULTS

The scoring method was developed on the basis of frequency table of responses to the individual scale items. The factor analysis returned 3 factors explaining 59.19% of total variance (Intention, Life, and History). The Cronbach's alpha was 0.85 for the Intention, 0.69 for the Life and 0.52 for the History subscale.

LIMITATION

The findings need replication in clinical and epidemiologic studies.

CONCLUSION

The RASS is a reliable and valid instrument which might prove valuable in the assessment of suicidal risk in the general population as well as in mental patients.

The Schizophrenia Suicide Risk Scale (SSRS): development and initial validation

BACKGROUND

Estimations about the lifetime risk of suicide in schizophrenia vary between 4 and 10%. At present, there does not exist a suicide risk scale developed particularly for schizophrenic patients. The aims of the present study were to: (1) develop a clinically useful semi-structured scale for the estimation of short-term suicide risk among schizophrenic patients, and (2) to carry out an initial validation of the scale.

METHODS

A 25-item Schizophrenia Suicide Risk Scale (SSRS) was constructed on the base of the literature. The SSRS scores of 69 living schizophrenic patients (LS group) were compared with the scores of 69 schizophrenic suicides (SS group) whose data had been collected previously from The Finnish nationwide and representative psychological autopsy study. Internal consistency of the SSRS was evaluated with Cronbach alpha. The most important SSRS items predicting suicide were identified with a logistic regression analysis. Sensitivity, specificity, positive predictive value, and negative predictive value of the SSRS in predicting suicide with various cut-off scores were calculated.

RESULTS

In the final logistic regression model, the following SSRS items significantly predicted suicide: suicide plans communicated to someone during the past 3 months; one or more previous suicide attempts; loss of professional skills demanding job; depression observed during an interview; and suicide plans communicated during an interview. With high cut-off scores the specificity of the SSRS became satisfactory, but the sensitivity dropped below 32%. Internal consistency of the anamnestic history of the SSRS was low, which suggests that anamnestic risk factors for suicide in schizophrenia are multifactorial. Internal consistency of the interview-based items was high, and present state risk factors seemed to consist of two separate factors, depression-anxiety and irritability.

CONCLUSIONS

The SSRS may be clinically useful in identifying schizophrenic patients with a particularly high risk for suicide. However, the SSRS seems not to be a practical screening instrument for suicide risk in schizophrenia, and it is probably impossible to construct a suicide risk scale with both high sensitivity and high specificity in this disorder.

Hopelessness and suicidal behavior

The main purpose of this study was to investigate the relationship of the Beck Hopelessness Scale (BHS) with suicidal behavior and other clinical variables in 212 hospitalized suicide attempters. Another purpose was to analyze the usefulness of BHS in this particular context. Those who later committed suicide (N = 13) had significantly higher median scores than those who did not. Even though the median BHS score for the total population was above the cut-off score for future suicide risk, suggested by Beck et al., the relative ratio of correctly positive/false positive was low. Therefore the BHS does not seem to be a satisfactory instrument for predicting future suicide in hospitalized suicide attempters. Our findings rather suggest that the BHS is related to depression ratings, mood disorders and/or personality disorders.

Attempted suicide: efficacy of treatment programs

Persons with a history of attempted suicide have an increased risk of eventually committing suicide. Eleven controlled perspective studies are reviewed on treatment programs aimed at reducing the rates of suicide and of repeated suicidal attempt in these considered at risk. None of these aftercare programs had a demonstrable impact on the suicide rate, and only in one investigation was a significant reduction of repeated attempts observed. Lack of statistical power and inadequate treatment strategies are discussed as major contributors to this failure. Recent evidence of a reduction of suicide rates and mortality by consequent and long-term treatment of patients with affective disorders, as well as reports of undertreatment of depression prior to completed suicide, might provide a new stimulus for the development and evaluation of aftercare programs for those who attempt suicide. The available evidence advocates a shift of emphasis from crisis management to the identification and treatment of psychiatric illness in this risk group.

Suicide and violence risk. I. Relationship to coping styles

Thirty suicidal psychiatric inpatients were compared with 30 nonsuicidal psychiatric inpatients. A battery of self-report tests was administered measuring suicide risk, violence risk and coping styles. The suicidal patients scored higher on both the suicide risk and violence risk. They also used almost all coping styles less frequently than the nonsuicidal patients. Among suicidal patients, suicide risk was negatively correlated with the coping styles of minimization, replacement and blame. Further, suicide risk and violence risk in suicidal patients were shown to be predicted by coping styles. These data indicate that suicidal patients have inadequate mental resources to deal with life problems. Besides their significance for assessing suicide risk, the findings may have possible clinical implications for case detection and early intervention with potential preventive value.

Correlates of suicide and violence risk in an inpatient population: coping styles and social support

Forty-six suicidal psychiatric inpatients were compared with 44 nonsuicidal psychiatric inpatients on measures of suicide risk, violence risk, impulsivity, feelings of anger, social support, and eight coping styles. The two groups were similar on demographic variables, but the suicidal patients were higher on the suicide risk scale, the violence risk scale, the impulsivity scale, and feelings of anger. Suicidal patients were significantly less likely to use the coping style of minimization to deal with life problems. A negative correlation was found between the social support measure and both suicide risk and violence risk. Three coping styles were found to correlate negatively with violence risk, but none with suicide risk. Coping styles were found to be a better predictor of violence risk than of suicide risk. It was also found that the more variables included in the predictor equation, the higher the correlation with the risk measures.

The Meta-Contrast Technique--a projective test predicting suicide

Sixty-nine inpatients who had attempted suicide were studied by means of the Meta-Contrast Technique (MCT), a projective test measuring personality factors, especially defensive strategies. The patients were divided into 3 subgroups, one of which was defined as stereotypy only, which denotes stereotypy (perceptual retardation) without any other coded defenses. At follow-up 7 of 8 completed suicides belonged to this subgroup and they matched various main diagnoses according to DSM-III-R. When the MCT findings of all patients were compared with 99 depressed inpatients from a previous study, the latter group more often had mature defensive strategies. In both investigations most completed suicides were found in the stereotypy only group. Our findings indicate that, regardless of psychiatric diagnosis, stereotypy without other defenses in MCT predicts suicide.

The Suicide and Aggression Survey: a semistructured instrument for the measurement of suicidality and aggression

Despite the rising concern with the increasing rate of violent and suicidal behaviors in this country and others, we remain relatively limited in our ability to rigorously define, classify, and measure these behaviors. In addition to our previous work in developing self-rating scales assessing aspects of aggressive behaviors, we have developed the Suicide and Aggression Survey (SAS), a new, comprehensive, semistructured interview and research tool, for the purpose of comprehensively evaluating and understanding the multiple constituents of these behaviors, and for assisting in predicting which individuals might be at high risk for suicide or violence. The present report describes the need for such an instrument and the theoretical models that have guided us in constructing it; one of these is a sequential description of the major classes of variables related to aggression and the other is a two-stage model of countervailing forces. Preliminary reliability data and a description of the structure of the interview are included.

Suicides and parasuicides in a high-risk patient group on and off lithium long-term medication

68 patients with affective disorders, and receiving lithium prophylaxis in a specialized lithium clinic were followed up for 8 years on average. Patients were selected for this study according to 2 criteria: They had been given lithium for at least 12 months, and they had attempted suicide at least once before onset of lithium prophylaxis. Outcome was analysed in terms of suicidal and parasuicidal behaviour. 55 patients took their lithium regularly, 13 discontinued or dropped the medication. One third of those patients having discontinued the medication died from suicide. Only one suicide occurred in patients with regular lithium intake and proven compliance during the last check before death. An impressive drop of parasuicides was observed in responders as well as in apparent non-responders. In total, 11 of 13 patients showed suicidal or parasuicidal behaviour 2 weeks-44 months after lithium discontinuation, which in about half of these cases took place on advice or with consent of the treating psychiatrist. It is concluded that lithium may have specific anti-suicidal properties, possibly related to its anti-aggressive effect, and that patients apparently not responding satisfactorily in terms of reduced number of episodes may still be protected against suicide or parasuicide.

The construction and selection of predictive scales, with special reference to parasuicide

Issues concerning the use of predictive and screening instruments include the difficulties posed by temporal changes in the base rates for the behaviour of interest, the possible advantages of using three rather than the customary two risk categories, and the desirability of specifying in advance the proportional sizes of the risk subgroups. A method is proposed for constructing and assessing tripartite risk scales, and a new scale for predicting the repetition of parasuicide, with the results of a prospective validation study, is reported.

Assessing the validity of the Plutchik Suicide Risk Scale

One of the newer scales of suicide risk devised by Plutchik and colleagues was tested to see if it could discriminate between levels of suicide ideation and suicide attempts. A sample of 80 subjects were placed into various categories of suicide risk on the basis of a psychiatric examination. Results showed that the full scale did not adequately discriminate among groups. An explantation of the results has been proposed and a design to test this suggestion described.

Clinical predictors of completed suicide and repeated self-poisoning in 8895 self-poisoning patients

The diagnoses of 8895 patients who were admitted for intentional self-poisoning with psycho-active drugs were studied in order to find predictors for subsequent completed suicide and repeated self-poisoning. Automated record linkage by means of the Swedish personal identification numbers was performed between the Stockholm County inpatient registry and the cause-of-death registry. With Cox regression models, several diagnostic predictors were identified although they were generally unspecific and insensitive. This may be due both to the low base rate of suicides, and to the omission of other more powerful non-clinical predictors, such as personality traits, hopelessness and social disruption. It is concluded that secondary psychiatric prevention may still be justified, although it will be applied to large numbers of patients who will not eventually commit suicide or repeat self-poisoning.

Correlates of suicide and violence risk 1: the suicide risk measure

A measure of suicide risk was developed using items reported to discriminate suicidal patients from controls in various studies. The new self-report scale was administered to 82 outpatients, 157 inpatients, and 83 college students. Using total scores, significant differences were found between the college sample and the two patient samples. The scale also discriminated between patients who reported one or more past suicide attempts and those who reported none. An independent cross-validation showed that half of the items continued to discriminate between patient and control groups. Sensitivity and specificity estimates were also determined. The test does not attempt to predict a specific rare event, i.e., suicide. It attempts to describe the degree to which a given individual reveals a set of characteristics that are similar to a suicide prototype.

Correlates of suicide and violence risk: III. A two-stage model of countervailing forces

Questionnaires and self-report scales were administered to 100 psychiatric inpatients. The scales measured such variables as depression, hopelessness, impulsivity, mental and life problems, family violence, personality characteristics, and dyscontrol tendencies. These were correlated with indices of suicide risk and violence risk. Most variables were found to correlate significantly with both suicide and violence risk. Partial correlation analyses revealed that 10 variables correlated significantly with suicide risk but not violence risk, while four variables correlated significantly with violence risk but not suicide risk. A two-stage model of countervailing forces, incorporating concepts from both psychoanalysis and ethology, is presented as a way of interpreting the results.

Post-diction of suicide in a group of depressive patients

Prediction of suicides was made on the basis of 99 psychiatric patients included in a study of anti-depressive therapy during 1961-63. These patients were followed up until 31 December 1984. Eight of them committed suicide. The prediction was based on two test methods, the Serial Colour-Word Test (CWT), which registers style of adaptation to a conflicting situation, and the Meta-Contrast Technique (MCT), where incongruous or threatening stimulation is introduced by degrees into neutral pictures and various defense misrepresentations of the threat are spotted. Both tests could predict suicide, but MCT was most successful. The sensitivity of the test was 0.80 and the specificity 0.75. The main risk symptom appeared to be depressive retardation together with lack of functional defensive structures.