BMJ July 24th 2012 BMJ 2012;345:e4692
We cannot predict who will kill themselves. It is that simple. Most people who commit suicide are low risk (about 90% of suicides) and most high risk people do not commit suicide. One reason often put forward for the inability to predict suicide is the low base rate of suicide – about 11 suicides per 100,000 people a year.
But what about violence and antisocial behavior which is far more common? Are we any better at prediction? A systematic review and meta-analysis in the BMJ suggest not – well not by much. The authors found 68 studies which included nearly 25,000 people of whom about a quarter offended over four years.
Tools which predicted violence and sexual offending did better than tools which predicted non-violent offending. The positive predictive value for violent offending was 0.41 and the negative predictive value was 0.91. This means that when one of the prediction tools predicted someone was going to violently offend it would be right 4 times out of 10 and when it said someone wouldn’t then it would be right about 9 times out of 10.
Useful enough in clinical practice? As the authors state: