Overview

• The analysis, published September 11, 2025, pooled results from 53 studies covering more than 35 million health records and nearly 250,000 suicide or hospital-treated self-harm cases.
• Algorithms showed high specificity but weak ability to identify those who would later self-harm or die by suicide, reflecting modest sensitivity on clinically relevant measures.
• More than half of people who subsequently presented with self-harm or died by suicide were labeled low risk by the models.
• Among those flagged as high risk, only 6% later died by suicide and fewer than 20% returned for treatment of self-harm.
• The authors judged most underlying studies at high or unclear risk of bias and concluded the tools perform no better than traditional risk scales, offering no basis to change clinical guidelines.