Can machine-learning methods really help predict suicide?

Area of Science:

• Psychiatry
• Computer Science
• Data Science

Background:

• Traditional statistical methods have limitations in creating clinically useful suicide prediction models.
• Recent interest in machine learning (ML) for suicide research stems from these limitations.
• This review examines ML approaches in suicide prediction studies.

Purpose of the Study:

• To review recent suicide prediction studies, including those using ML.
• To evaluate the added value of novel ML approaches in suicide research.
• To understand the current contribution of ML to suicide prediction.

Main Methods:

• Review of recent prediction studies in the suicide literature.
• Inclusion of studies employing machine learning techniques.
• Analysis of reported performance metrics for ML and conventional methods.

Main Results:

• ML models show only modest improvements in area under the curve compared to conventional methods.
• Sensitivities of ML models are comparable to traditional predictive methods.
• Positive predictive value for suicide prediction remains low (around 1%) in cohort studies.

Conclusions:

• Machine learning and artificial intelligence present opportunities in mental health and clinical care for suicidal patients.
• Careful consideration is needed to avoid repeating methodological errors of existing approaches.
• ML-based suicide prediction studies have not yet significantly advanced the field or proven clinical utility.