Background: UK government proposals to reduce the risks posed by people with "dangerous" severe personality disorders (DSPD) include a new legal framework for indeterminate detention. We aimed to establish the degree to which those operating the framework will be able to predict which people will act violently in the future.
Methods: We reviewed published reports in which the accuracy of a clinical judgment or a statistically derived rating of dangerousness was validated by its use to predict the violent behaviour of adults in the community. We calculated the sensitivity and specificity of the procedures used by every study. We then applied these sensitivities and specificities to the purported base rates of violence in people with DSPD.
Findings: 23 studies fulfilled the criteria, and for 21 of these the sensitivity and specificity of the procedures used by the investigators could be calculated. Using the average positive predictive power of these procedures, six people would have to be detained to prevent one violent act. Making predictions over shorter periods did not improve their accuracy.
Interpretation: In practice, the number of people that need to be detained is likely to be higher than we reported. Differences between populations in respect of which predictions are being made and the population on which an instrument was validated will reduce the accuracy of that instrument. Not all of the necessary information will always be available.