By looking for risk-patterns, machine learning could predict when people feel suicidal.
Each year in the United States, some 40,000 people commit suicide. The number rose by a whopping 24% between 1999 and 2014, and that rise is fairly indicative of one thing: We are not very good at predicting or preventing self-harm. A recent study concludes that despite generations of data, we are no closer to understanding and preventing suicides. There is hope though; new research indicates that machine learning algorithms, as found in artificial intelligence systems, could read warning signs and help improve our ability to predict at-risk behavior.
A new survey in Psychological Bulletin asked researchers to look at 365 studies from the past 50 years that included 3,428 different measurements of risk factors, such as genes, mental illness and abuse. Surprisingly perhaps, the researchers concluded that no single risk factor could be identified as a warning light, for predicting suicidal tendencies or thoughts. The problem is that most research on the subject looks at risk factors predicatively, and the studies in question spanned over ten years. It’s difficult to say outright based on almost any factor, whether someone will kill themselves ten years later. Most psychologists are interested in predicting short term suicidal tendencies – and this is where the research is lacking.
“Few would expect hopelessness measured as an isolated trait-like factor to accurately predict suicide death over the course of a decade,” the researchers write. “But many might expect that, among older males who own a gun and have a prior history of self-injury and very little social support, a rapid elevation in hopelessness after the unexpected death of a spouse would greatly increase suicide death risk for a few hours or days. Yet, most of the existing literature has tested the former hypothesis rather than the latter.”
The researchers recommend developing a machine learning program that analyzes thousands of risk factors to better help find patterns that could allow for predicting suicide over a range of time spans, from weeks to years. “Clinical predictions are really bad,” says Jessica Ribeiro, a clinician and psychologist at Florida State University, “We know that, but that doesn’t mean we actually accept that our own predictions are bad.”
Colin Walsh, data scientist at Vanderbilt University Medical Center, along with FSU’s Joseph Franklin and Ribeiro, looked at millions of anonymous health records and compared around three thousand clear cases of nonfatal suicide attempts. Then they let a computer process the data and find distinct patterns that would be able to predict suicide attempts within various time frames. Walsh hopes to see a “hybrid” approach to this in the future, in which clinicians factor computer recommendations into their judgment.