Machine learning algorithms, often seen as promising tools for revolutionizing mental health care, may not be as effective as hoped when it comes to predicting suicidal behavior. A comprehensive new study published in PLOS Medicine reveals that the accuracy of these AI models is too low to be clinically useful for screening or prioritizing high-risk individuals. Led by Matthew Spittal of the University of Melbourne, the research team conducted a systematic review andmeta-analysisof 53 studies from around the world. These studies applied machine learning algorithms to vast datasets of over 35 million medical records, including nearly 250,000 cases of suicide or hospital-treated self-harm. The goal was to assess whether AI could outperform traditional risk assessment tools in identifying individuals most at risk of future suicide or self-harm. While the algorithms demonstrated high specificity—meaning they were good at identifying people unlikely to self-harm—they showed only modest sensitivity, failing to correctly identify many individuals who later presented with suicidal behavior. “Specifically, the researchers found that these algorithms wrongly classified as low risk more than half of those who subsequently presented to health services for self-harm or died by suicide,” the study noted. Even among those classified as high-risk, only 6% went on to die by suicide, and fewer than 20% returned for treatment after self-harm, suggesting a high rate of false positives. “We found that the predictive properties of these machine learning algorithms were poor and no better than traditional risk assessment scales,” the authors said. “The overall quality of the research in this area was poor, with most studies at either high or unclear risk of bias.” The findings align with existing clinical practice guidelines, which already caution against using risk assessment scores to determine aftercare strategies. Reference:Spittal MJ, Guo XA, Kang L, Kirtley OJ, Clapperton A, Hawton K, et al. (2025) Machine learning algorithms and their predictive accuracy for suicide and self-harm: Systematic review and meta-analysis. PLoS Med 22(9): e1004581. https://doi.org/10.1371/journal.pmed.1004581