Sir: I agree with Szmukler (Psychiatric Bulletin, January 2000, 24, 6-10), about difficulties inherent in predicting rare events. This is especially so when attempting to predict homicide. The results of the National Confidential Inquiry (Reference Shaw, Appleby and AmosShaw et al, 1999) suggested that 15 homicides by people with schizophrenia who have had contact with psychiatric services occur each year. If we estimate that 80% of people with schizophrenia have contact with psychiatric services at some point during their lives this would suggest that the incidence of homicide by people with schizophrenia is about 0.094 per 1000 per year.

I am not aware of any instrument devised to identify patients at high risk of committing homicide, but those designed to detect violent incidents report their sensitivity and specificity to be around 0.8. If these figures were applied to an instrument for predicting homicide it would have a positive predictive value 0.0002. This value is the “proportion of patients with positive test results who are correctly diagnosed” (Reference AltmanAltman, 1991). In other words, for everyone identified correctly, 5000 people will be identified as being at high risk of committing a homicide but will not do so.

The rarity of homicide suggests that even highly sensitive methods for detecting risk fail to adequately distinguish those who commit homicide from the vast majority who will not. While risk assessment is an important part of providing high quality of care to all patients in contact with mental health services, thinking about patients in terms of the likelihood of their committing homicide is not.