There are two recognized approaches to reducing firearm violence. One is to regulate guns and ammunition. The other is to regulate the people who might use guns and ammunition to harm others (1). In part because of limits on states' ability to regulate gun ownership, every new outrage involving firearms provokes questions concerning how we can prevent guns from falling into the wrong hands. Having a mental disorder is widely seen as making some people more dangerous. A “people regulation” approach to gun violence by those with mental disorders requires, first, reliably identifying dangerous people and, second, preventing those people from obtaining firearms. Neither is currently possible. Making this fact more explicit might focus attention and resources on areas where more could be achieved.

Identifying dangerous people without the wisdom of hindsight is difficult, as Dr. Swanson points out in the Open Forum in this issue (2), and the limits of the predictive accuracy of violence risk assessment are now widely recognized. On the basis of the rate of serious violence observed in the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study (3)—and using the most accurate techniques available, as well as treating false-positive and false-negative errors as equally to be avoided—such approaches can identify a person who will act violently only at a cost of placing in the same “risk class” a total of 14 persons, on average, who will not (4). This number rises as the rate of violence in the population falls. If the object is to limit access to firearms only for persons who would use a gun to commit violence, the base rate will be lower (because not all violence is gun based) and the number of people placed in the same risk class who will not act violently with a gun will be even higher. There has been some improvement in the predictive validity of risk assessment techniques over the past 30 years. But this improvement does not permit the reliable identification, in the population at large, of people with mental disorders who will commit firearms offenses.

If the difficulties in identifying dangerous individuals could somehow be overcome, what could then be done? Intensive outpatient support improves quality of life but there is little evidence that it reduces violence (5). Using psychiatric measures as risk factors in background checks stigmatizes people with mental illness and raises the question of which factors should be included. Two categories used in the Gun Control Act—being found “not guilty by reason of insanity” and having a history of being civilly committed—speak to perceived riskiness (not always riskiness to others) at the time of a past crime or hospital admission but do not speak clearly to future risk. Twelve years after its launch, the impact of the National Instant Background Check System (NICS) remains unclear. The Office of Technology Assessment estimated that for 80% of the targeted population with mental disorders, records may not exist (6). As of 2007 only 22 states provided mental health data to the NICS (7). Licensed gun outlets can in any case avoid background checks by obtaining “Brady permits,” and dealers selling secondhand firearms at gun shows are exempt.

There are good reasons to prevent some users of psychiatric services, some of the time, from having access to weapons. This is different from saying that a “dangerous people” approach to gun ownership applied to a psychiatric population can be expected to prevent future outrages. Dr. Swanson is right, for many reasons, to suggest that services should be improved. But we don't know how much gun violence would thereby be prevented and, as a result, cannot assume that a further decline in the U.S. homicide rate would be a likely consequence. From a psychiatric standpoint, the strongest statement that can safely be made in relation to recent notorious incidents is that improved services might have prevented some of them. Research into risk assessment will continue to advance, but this will remain true. The difficulty is not attributable to a lack of data.