Study: Machine learning shows promise toward accurately identifying suicidal behavior

By Heather Mack

Digital tools using machine learning to analyze a person’s spoken or written words could be instrumental in aiding mental health clinicians in assessments determining whether that person is suicidal, researchers have found.

A new study published in the journal Suicide and Life-Threatening Behavior found machine learning is 93 percent accurate in correctly identifying a suicidal person, and is 85 percent accurate in determining differential diagnosis of mental illness. The study, led by researchers at the Cincinnati Children’s Hospital Medical Center, looked at 379 patients who were recruited from three different sites – two academic medical centers and a rural community hospital.

“Death by suicide demonstrates profound personal suffering and societal failure,” writes lead author Dr. John Pestian, who is also a professor of biomedical informatics and psychiatry at Cincinnati Children’s. “While basic sciences provide the opportunity to understand biological markers related to suicide, computer science provides opportunities to understand suicide thought markers.”

The study took place over an 18-month period, and the participants were divided into one of three groups: suicidal, mentally ill but not suicidal, or controls. Participants were asked to complete standardized behavioral rating scales and answer questions such as “Do you have hope?” “Are you angry?” and “Does it hurt emotionally?” Researchers then used machine-learning algorithms to measure and blend together written or spoken thought that could indicate suicide or mental illness. 

Given the accuracy in the results, Pestian said the study shows strong evidence that machine learning could be a useful objective tool that clinicians and others around people with suicidal tendencies could use to determine whether they should intervene.

“These computational approaches may provide novel opportunities for large-scale innovations in suicidal care,” Pestian concluded. “The methodology described here can be readily translated into such settings as school, shelters, youth clubs, juvenile justice centers, and community centers, where earlier identification may help reduce suicide attempts and deaths.”