Machine learning algorithms and their predictive accuracy for suicide and self-harm: Systematic review and...

Area of Science:

• Computational psychiatry
• Clinical informatics
• Public health research

Background:

• Rapid advancements in machine learning (ML) for predicting suicidal behaviors.
• Need to systematically evaluate ML algorithm accuracy for suicide and hospital-treated self-harm prediction.

Purpose of the Study:

• To conduct a systematic review and meta-analysis of ML algorithms predicting suicide and hospital-treated self-harm.
• To assess the diagnostic accuracy of these ML algorithms.

Main Methods:

• Systematic search of multiple databases (PubMed, PsycINFO, Scopus, etc.) until April 2025.
• Inclusion of case-control, case-cohort, or cohort studies on suicide or hospital-treated self-harm.
• Exclusion of studies with self-reported outcomes or other designs.
• Statistical assessment of accuracy using methods for diagnostic accuracy studies.

Main Results:

• Fifty-three studies met inclusion criteria.
• Area under the ROC curve ranged from 0.69 to 0.93.
• Sensitivity: 45%-82%; Specificity: 91%-95%; Positive Likelihood Ratio: 6.5-9.9.
• Positive predictive values varied significantly with population prevalence (0.1% to 66%).

Conclusions:

• ML algorithm accuracy for predicting suicidal behavior is currently too low for effective screening or intervention prioritization.
• For hospital-treated self-harm, management should focus on needs-based assessment, addressing modifiable risk factors, and effective aftercare.