Accuracy of Clinician Predictions of Future Self‐Harm: A Systematic Review and Meta‐Analysis of Predictive Studies

Assessment of a patient after hospital‐treated self‐harm or psychiatric hospitalization often includes a risk assessment, resulting in a classification of high risk versus low risk for a future episode of self‐harm. Through systematic review and a series of meta‐analyses looking at unassisted clinician risk classification (eight studies; N = 22,499), we found pooled estimates for sensitivity 0.31 (95% CI: 0.18–0.50), specificity 0.85 (0.75–0.92), positive predictive value 0.22 (0.21–0.23), and negative predictive value 0.89 (0.86–0.92). Clinician classification was too inaccurate to be clinically useful. After‐care should therefore be allocated on the basis of a needs rather than risk assessment.     

Clusters of Suicidal Events Among Young People: Do Clusters from One Time Period Predict Later Clusters?

We sought to compare clusters of suicidal events between two different time periods and examine the extent to which earlier clusters predict later clusters. We included data on suicides and suicide attempts from New South Wales between July 2001 and June 2012 and Western Australia between January 2000 and December 2011. Suicide attempts included admissions to hospital for deliberate self‐harm and suicides were deaths due to deliberate self‐harm. We combined data on suicides and suicide attempts and grouped them into two equal time periods. We detected clusters in each period using Poisson discrete scan statistics adjusted for socio‐economic status. We estimated the predictive values of earlier clusters on later clusters. The results showed that clusters from earlier time period had a moderate power (36%) in predicting later clusters. During the later time period, some additional cluster areas (14%) were found and some earlier cluster areas subsided (64%). Historical clusters predict 36% of the subsequent clusters, which is probably not sufficient for targeting interventions. Our study highlights the need for other strategies to detect emerging clusters, for example, up‐to‐date data.