Letters

Only about 1 in 30 predictions of assault by discharged psychiatric patients will be correct

EDITORThe debate about the dangerousness of discharged psychiatric patients is interesting. ^{1 2} In public debate (and sometimes also among professionals) it is often claimed that discharged patients are responsible for a substantial number of violent assaults in society. This is sometimes held as a reason for more custodial, institutionalised treatment.

In 1987 Wistedt and I studied the possibility of using a prediction of the likelihood of violence, assessed at discharge from involuntary psychiatric care, as a means of reducing rates of violence in Swedish society.³ Our calculations showed that, at most, 100 serious assaults a year in Sweden were committed by patients who had been discharged during the previous year from involuntary psychiatric treatmentthat is, less than 1% of all patients discharged. Trieman et al estimated that 2% of their population of discharged patients committed serious violent acts within the five years after discharge (that is, 0.4% a year).²

When discussing the possibility of preventing these serious acts of violence the possibility of predicting statistically rare events must be taken into account.⁴ Beck showed in 1985 why the psychiatric profession fails in predicting violence after discharge from an institution.⁵ The reason is not lack of knowledge of risk factors important for future violence but rather a statistical impossibility.

The ability to make a correct prediction depends not only on the accuracy of the classification of future offenders (sensitivity) but also on the correct identification of all future non-offenders (specificity). If we assume that the psychiatric profession succeeded in predicting with 90% sensitivity and 90% specificity (a very high accuracy) our Swedish sample showed that for every correct prediction of violence after discharge from an institution, 11 patients would be falsely predicted to commit serious violence. Given the same predictive accuracy and Trieman et al's annual rate of violent patients discharged of 0.4%,² the ratio of correct to false positive predictions of assault would be about 1 in 30.

If the consequences for the patients who are positively predicted to be violent include some kind of extended incarceration the question of prediction becomes a moral issue since those interventions made would be not for the benefit of the patient but rather a precaution for society. From an ethical viewpoint, prediction of violence and restraining psychiatric patients into more custodial care is not only useless for society but bears extremely high costs for those many patients falsely predicted to become violent.

Tom Palmstierna. 
Centre for Dependency Disorders, St Göran's Hospital, PO Box 125 60, S-102 29 Stockholm, Sweden tom.palmstierna{at}mailbox.euromail.se