Forensic psychiatry is principally concerned with assessing and managing the risk to others (usually of violence) by people with a mental disorder. A variety of lengthy risk assessment instruments help consolidate this expertise, but these instruments do not find favour in day-to-day psychiatry.

Based on the work of Fazel and colleagues, ^{Reference Fazel, Långström, Hjern, Grann and Lichtenstein1,Reference Singh and Fazel2} we set out to determine whether five ‘yes or no’ questions (male gender; less than 32 years old; previous criminal convictions; and comorbid alcohol misuse and drug misuse) could predict later actual physical violence to others. We analysed case notes on consecutively discharged patients from a medium secure forensic unit (52 patients, 46 male); an out-patient addictions service (51 patients; 26 male); and a crisis resolution and home treatment service (25 patients, 17 male), in a ‘pseudo-prospective’ method for a record of physical violence after applying these five questions as a screen to the case records 5 years earlier, from January 2006. Records with insufficient detail or length of history were excluded, and the screen was viewed as a positive predictor if three or more questions were answered ‘yes’.

We found 30 (of 128) patients were violent in the 5 years studied, with 83% being predicted as violent by our screen (sensitivity), and a false negative rate of 17%. The positive predictive value was poor at 38% but the negative predictive value (i.e. that a negative prediction was correct) was impressive at 92%. The factors predicting later violence were being male (93%); having a history of violence (80%, not a ‘Fazel’ question); a history of drug misuse (77%); a prior criminal conviction (70%); a history of alcohol misuse (60%); poor treatment adherence (52%, not a ‘Fazel’ question); and being less than 32 years old (50%). A history of self-harm was only seen in 20% of those who were violent later.

The rates of 5-year violence in the three separate groups were 35% in the forensic sample, 6% in addictions, and 36% in the acute community crisis resolution home treatment group. We acknowledge the preponderance of males in our sample will skew the results, given it is a screening question.

Our results raise two interesting points. First, that these five simple questions might aid clinical decision-making concerning which patients will not pose a risk of later violence, but does not elucidate prediction on who will become violent. This screen might therefore be useful as part of a stepped approach in a busy clinical environment when considering who to refer for more in-depth assessment. Second, as Turner & Salter have already noted, ^{Reference Turner and Salter3} we conclude it is hard to define who is ‘a forensic patient’ when we compare patterns across our three samples.