Machine learning can help identify what puts some schizophrenia patients at greater risk for violence

April 4, 2021

Researchers are breaking down the link between mental illness and violence. (AP Photo/Wilson Ring)

University of Zurich researchers have found that the length of time spent in a psychiatric hospital and the age of a first diagnosis were the most significant variables that separate violent and nonviolent people who live with schizophrenia spectrum disorder.

A recent study, published March 9 in Comprehensive Psychiatry, utilized data from 370 patient records and placed 519 variables through a machine-learning algorithm to identify factors that could put a person living with schizophrenia spectrum disorder at greater risk of exhibiting violent behavior. The model was able to classify violent and nonviolent cases almost three-quarters of the time. But the researchers cautioned that because the algorithm couldn't account for more than a quarter of cases, the results should be carefully scrutinized.

Although some studies have found that untreated people living with schizophrenia spectrum disorder may be at a higher risk of committing violent crime, only one in 20 violent crimes is committed by someone with a severe mental illness. In addition, those with severe mental illnesses such as schizophrenia are more likely to be the victims of violent crimes than they are to commit violence themselves, with one 2005 study showing a likelihood of victimization 11 times greater than that of the general population. Other analyses have indicated that substance abuse may be a factor in a large portion of violent crimes committed by those with severe mental illness.

Perhaps the most surprising facet of the study was the dearth of evidence that early environmental factors lead to more aggressive or violent behavior, said Martina Sonnweber, a lead author on the analysis and a researcher at the University of Zurich's Department of Forensic Psychiatry, especially since traumatic childhood experiences are often thought to be associated with more severe cases of mental illness.

"We could not find many childhood or adolescence related factors," Sonnweber told The Academic Times. "Since psychology and psychiatry are so interested in childhood and attachment and early stressors, this seems quite odd."

Although the researchers didn't spot correlations between violence and adverse childhood experiences, they did note that social isolation in adulthood could lead to an elevated risk of violence in some patients with mental illness. And even if social factors do not solely contribute to a person's capacity for violence, a lack of social services, financial assistance, community support and clinical care could all play a role in further isolating patients, the authors said.

"It would be conceivable that as soon as someone has schizophrenia and shows transgressive behavior, violence and criminality, some protective system would intervene, and the patients would not be sent back to their possibly existing isolation and socially precarious situation," Sonnweber said. "This could be done by better linking them to a social worker or [a similar] kind of system."

For decades, the psychiatric community has fiercely debated the potential link between severe psychiatric disorders and violence, often coming up with contradictory evidence. The debate has been further complicated by the ways in which researchers have so far collected their data. Some studies have used arrest and crime statistics to identify correlations without accounting for the fact people with mental illnesses are overrepresented in the criminal justice system and are often in jail for nonviolent, minor offenses. 

Without proper psychiatric treatment or medicine, people with a mental illness who endure prolonged jail sentences can also experience worsened symptoms that can persist long after they are released, according to the National Alliance on Mental Illness.

One of the goals of the University of Zurich study was to disentangle the boundaries between biological and environmental factors that can make some people's experience with schizophrenia more severe. Schizophrenia spectrum disorder represents a broad range of symptoms and outcomes, and it is likely composed of hundreds of heterogeneous risk factors, making it that much more difficult to separate the disorder's environmental and biological risk factors.

Whereas most psychiatric studies involve separately testing a small number of finely tuned hypotheses, the University of Zurich researchers' machine-learning approach allowed them to evaluate multiple factors at the same time to locate nonlinear correlations. It's important to isolate the variables that contribute to violence, the researchers said, so that psychiatrists can provide additional resources and prevention tools to people who may be more likely to commit violence.

"But this is a complicated problem: most of the patients are not violent, and it would be obviously wrong to put them in a register in advance," Sonnweber said. "Here an ethical discourse is immensely important."

The study, "Violent and non-violent offending in patients with schizophrenia: Exploring influences and differences via machine learning," published March 9 in Comprehensive Psychiatry, was authored by Martina Sonnweber, Steffen Lau and Johannes Kirchebner, University of Zurich.

We use cookies to improve your experience on our site and to show you relevant advertising.