Machine-learning algorithms underestimate self-harm risk, Pg11

Key Highlights:

• Machine-learning algorithms designed to predict suicide or self-harm risk have a success rate below 50%.
• Over half of those at risk are misclassified as "low risk" by these algorithms.
• The review encompassed 53 studies testing these algorithms.
• Authors recommend needs-based assessments for all patients instead of relying on algorithms.

Detailed Insights:

• The low accuracy of these algorithms raises concerns about their reliability in mental health risk assessment.
• Misclassification can lead to at-risk individuals not receiving timely intervention and support.
• Needs-based assessments would allow clinicians to make more informed decisions about treatment allocation.
• This study highlights the limitations of using AI in complex mental health predictions.

Scientific/Technical Concepts Involved:

• Machine Learning: A type of AI where systems learn from data without explicit programming.
• Algorithms: Step-by-step procedures or formulas used by computers to solve problems.
• Self-harm: Intentional injury to oneself, irrespective of the intent to die.