• The Language of Suicide

    New research analyzes suicide notes for clues to self-destructive mindsets

    Every two hours in America, a young person commits suicide. Every few minutes, another young person tries.

    “At Cincinnati Children’s, about 40 suicidal kids a week come into our Emergency Department,” says John Pestian, PhD, MBA, a leading researcher in the Division of Biomedical Informatics. “This is an extraordinarily complex problem because both biology and psychology are interacting. Better ways to identify suicidal behavior are needed.”

    The impact of suicide is deeper than many people realize. National data show that families are more than twice as likely to lose an adolescent child to suicide than to cancer. Nearly 4,400 young people – ages 15 to 24 – commit suicide each year, making intentional self-harm the third leading cause of death for this age group, according to the American Association of Suicidology. That compares to nearly 1,700 people ages 15 to 24 who die of cancer each year, according to the Centers for Disease Control and Prevention.

    Yet science has been unable to find effective ways to prevent suicide.

    An emerging field

    Pestian is among a handful of investigators searching for new ways to address this devastating problem. After amassing one of the world’s largest collections of suicide notes, Pestian is applying the latest in natural language processing science to better understand what a suicidal person is thinking. The goal: to develop an evidence-based tool that pediatricians, psychiatrists, social workers and others can use to predict the likelihood of future suicide attempts.

    “People have a natural propensity for self-preservation. Many of us may wish we were dead at some point in our lives, but we do not carry out the act,” Pestian says. “So what moves a person from self-preservation to self-annihilation? The best artifacts of those thoughts are the written and spoken words. The words that people leave behind are the closest thing we have to their final thoughts.”

    Efforts to study suicide notes date back at least as far as the 1960s. But no one has attempted to apply this level of computational rigor to the task.

    Over nearly eight years, Pestian and his colleagues collected and digitized more than 1,300 suicide notes. Family survivors, members of suicide prevention groups and mental health experts helped annotate the collection, assigning emotions to key words and phrases. It was complex, difficult work that Pestian says “could not have been achieved without the collaboration and commitment of many individuals.” The result was a database that identifies patterns and generates questions that can help mental health professionals more accurately determine which patients are serious about suicide.

    In a recently completed clinical trial, Pestian’s team reported that their computer algorithm was 90 percent accurate at distinguishing suicidal patients from non-suicidal control cases. Clinicians interpreting the same results were about 50 percent accurate. Why such a difference? Pestian calls it “psychological phenomology.” That is, when we hear something, we associate it with past experiences. Computers consider the structure of the message, not just the content.

    Machine intelligence

    The still-experimental computer algorithm takes into account: sentence lengths; key words and phrases; specific references to time, people or religion; even the overall length of responses. It uses the patterns gleaned from past suicide notes to look for similar patterns in the speech of people at risk of suicide.

    The goal is to produce a software product that will support a clinician’s decisions by providing real-time information. But getting to that point will require more work.

    Three organizations are teaming up to fund continued development: Cincinnati Children’s, AssureRx, a medical center spinoff, and Diamond Healthcare Corp., which operates mental health facilities. A multicenter clinical trial, using an Innovation Fund award from Cincinnati Children’s, will evaluate the algorithms in larger, more diverse groups, including adults.

    “We’ve tested it here. Now we want to test it in other hospitals,” Pestian says.

    Why focus on suicide?

    Pestian says his interest in suicide research has been entirely academic. It started when emergency medicine colleagues asked him to look into developing a better tool to use in the ED.

    “The more I studied this, the more I saw how the complexity devastates the survivors,” Pestian says. “I thought this was a good opportunity to make a difference, to do good.”

  • John Pestian, PhD, MBA, leading researcher in the Division of Biomedical Informatics.

    John Pestian, PhD, MBA, leading researcher in the Division of Biomedical Informatics.