Is suicide predictable? Paul St John-Smith Short Courses in Psychiatry 15/10/2008

Risk/Prediction in in-patients Retrospective case study Powell, Geddes et al (2000) compared 112 people who committed suicide while in-patients in psychiatric hospitals with 112 randomly selected control patients in order to identify risk factors for suicide in psychiatric in-patients and to evaluate their power in detecting people at risk of suicide. The suicide rate in psychiatric in-patients was 13.7 (95% CI ) per 10,000 admissions.

The predictive factors with likelihood ratios greater than 2, following adjustment were: Actual suicide attempt = 4.9 Family history of suicide = 4.6 Planned suicide attempt = 4.1 Recent bereavement = 4.0 Presence of delusion = 2.3 Chronic mental illness = 2.2 Predictive Factors

They concluded that even in this high-risk group clinical utility of these factors is limited because of their low sensitivity and low specificity. With the rarity of in-patient suicide, even in this high-risk group, the positive predictive value of these factors is low and the number of false positives would be more than 99%. Only 2 of the 112 patients who committed suicide in the study had a predicted risk of above 5%. Conclusions

The US Preventative Services Task Force (2004) concluded that there was limited evidence that the currently available screening tools, including tools to identify those at high risk accurately identified those at risk of suicide in primary care settings. They found no or insufficient evidence that screening or treatment those at high risk reduced suicide attempts or mortality. US PSTF

The known risk factors for suicide include: male gender, suicidal ideation on admission, diagnosis, outcome on discharge and the number of previous suicide attempts. Risk Factors: Goldstein (i)

Goldstein et al (1991) assessed these risk factors for suicide and the actual suicides over a 10-year period in 1906 patients with affective disorder admitted to a psychiatric hospital. They were unable to identify any one of the 46 patients who committed suicide during this period with a 50% probability. They could identify only one of them with a 15% probability. However, 15% is the same as the risk of suicide in depressive disorders. Risk Factors: Goldstein (i)

A suicide screening tool CALCULATIONS BY PAUL ST JOHN SMITH CONSULTANT PSYCHIATRIST 6/5/2008 Cooper, J., Kapur, N., Dunning, J. et al (2006) A clinical tool for assessing risk after self-harm. Annals of Emergency Medicine, 48,

Cooper et al (2006) studied >9000 patients who attended emergency departments with self-harm and developed the Manchester Self-Harm Rule using 4 clinical correlates: –i.e. any history of self-harm, –previous psychiatric treatment, –benzodiazepine use in the current attempt, –and any current psychiatric treatment. Manchester Scale

They found a sensitivity of 94% and a specificity of 25% in identifying high- risk patients. Amongst them17% re-attended and 22 died by suicide within 6 months. ROC

Let us apply the above instrument to screen a population of 100, 000 people with severe mental illness and hence having a 30 times the general population rate (10/100, 000/ year) of suicide i.e. 300/100, 000/year. Calculation 1

The following table demonstrates the figures and numbers of false positives and false negatives given the above sensitivity (94%) and specificity (25%) of the instrument. Table

Suicide risk present Suicide risk absent Total Screening test positive a True positive 282 b False positive 74,775 a + b 75,057 Screening test negative c False negative 18 d True negative 24,925 c + d 24,943 Totala + c 300 b + d 99, 700 a + b + c + d 100, X 2 TABLE

Among this group of 100,000 people with mental illness, 300 people will commit suicide in one year, and 99,700 will not. With the 25% specificity of this (the best?) instrument available you will correctly identify 24, 925 as not at risk of suicide. You also identify 75% of the 99,700 i.e. 74, 775 as at risk for suicide when they are not. Calculation 2

SENSITIVITY Among the 300 who will commit suicide, the instrument will correctly identify 94%, i.e However, this instrument will not identify 6% of them, i.e. 18 people who will commit suicide.

Thus, the positive predictive value (PPV) i.e. the chance of having a disease given a positive test result (PPV = a / a+b) is 282 / 75,057 = 0.38%, PPV

The negative predictive value (NPV) NPV = the chance of not having a disease given a negative test result (NPV = d / d+c) is 24,925 / 24,943 = 99.9%

NPV Thus, this instrument will accurately identify non-suicidal patients 99% of the time. However, compared to 75, 057 people not ruled out as not at risk of suicide, only 282 would really be at risk of suicide.

CONCLUSION If we assume that hospital admission would prevent all suicides, one would have to admit 75, 057 people in order to prevent 282 suicides and would still miss 18 suicides.

Predicting Rare Events

Prediction; Fact of Fiction? Prediction is a fact in so far as suicides will occur sometimes! Some people are more AT RISK (probable) than others Who exactly, amongst those at risk, is difficult to predict <1.0% but we can predict it will happen sometimes ! We predict positive completion wrongly 99% of the time Who exactly will complete suicide and when and where remain problems for clinical staff.

Conclusion Total Prevention of suicide = 100% PPV i.e. prediction (who and when) + Totally effective treatment/strategies.