A review and meta-analysis in PLOS Medicine concludes that machine learning tools do not reliably predict suicide or hospital-treated self-harm and should not be used to screen or triage people for high-intensity interventions.
The study, led by Matthew J. Spittal at the University of Melbourne, with collaborators from the University of Adelaide, KU Leuven, and the University of Oxford, examined 53 studies published between 2015 and 2025. Together, these studies analyzed over 35 million health records and 249,000 cases of suicide or self-harm.
The authors summarize their findings:
“The accuracy of machine learning algorithms for predicting suicidal behaviour is too low to be useful for screening (case finding) or for prioritising high-risk individuals for interventions (treatment allocation).”
“Machine learning algorithms incorrectly classify more than half the people who subsequently present to hospital for self-harm or die by suicide as low risk,” the authors write in their summary. “A classification of high risk poorly forecasts who will engage in suicide or self-harm.”
You've landed on a MIA journalism article that is funded by MIA supporters. To read the full article, sign up as a MIA Supporter. All active donors get full access to all MIA content, and free passes to all Mad in America events.
Current MIA supporters can log in below.(If you can't afford to support MIA in this way, email us at [email protected] and we will provide you with access to all donor-supported content.)
And who is surprised by these results?
Report comment