Predicting Mental Patient Risk

Psychiatrists are frequently called upon to predict the likelihood that someone is a physical threat to themselves or others as a consequence of a mental illness. They aren't very good at it. The literature on this subject is notorious for showing that expert predictions are no better than, or worse than random chance, that pyschiatric office staff does just as well as the psychiatrists, and that paper and pencil tests are better predictors than experts. Yet, the show goes on because the important decisions need to be made by somebody.

Against this backdrop, even crude tools with a proven track record look good. Thus, a new study the shows a prediction method with some statistical validity, is welcome.

Psychologist John Monahan of the University of Virginia in Charlottesville calls the new paper “one of the most original and important studies on psychiatric symptoms and violence published in the past decade.” Monahan has studied the relationship between psychiatric disorders and violence for more than 30 years.

The approach shows that the frequency with which someone cycles from high points to low points of symptoms predicts the risk of harm to self or others:

Employing a statistical technique called dynamic systems modeling, the new work shows that among psychiatric patients with documented histories of committing violent acts, those whose symptoms of emotional distress rapidly and repeatedly fluctuated from mild to severe during a 26-week period were particularly apt to assault others or to threaten them with a weapon . . . . In cases of rapid symptom fluctuation, patients went from peaks to valleys of emotional health about every two to four weeks. . . . Violence occurred less frequently among patients whose symptoms fluctuated from high to low points over longer stretches, which often lasted about 10 weeks.

Overall, patients whose symptoms rapidly ebbed and flowed were almost three times as likely to become violent than those whose symptoms oscillated slowly. Patients who displayed rapidly fluctuating psychiatric symptoms that also worsened during the study were especially likely to commit two or more violent acts during the study period. . . .

[E]arlier studies have examined only whether symptoms present at one point in time predict violence at a later point. . . . Odgers and her coworkers studied 132 adults for 26 weeks after they were treated and released from a psychiatric hospital emergency room in 2000. Record reviews showed that all participants exhibited intense hostility and, in the two months before treatment, had heavily used alcohol and illicit drugs and had been involved in serious violence. Patients’ violent acts involved physical injuries, sexual assault or the use of weapons.

Each week during the study, the researchers interviewed a close family member or friend of each patient, as well as the patients themselves, for information on participants’ recent symptoms and violent behaviors. By the end of the 26 weeks, 78 patients had committed at least one violent act.

In further research, Odgers plans to examine whether rapid symptom fluctuations begin before or after patients commit violent acts. She also wants to explore the relationship between symptom swings and stressful events, including job loss and fights with romantic partners. . . . Many factors, including hostility and paranoia, are already known to make small but statistically significant contributions to predicting whether psychiatric patients will commit violence[.]