Prediction Models for Suicide Attempts and Deaths: A Systematic Review and Simulation

Area of Science:

• Health Informatics
• Psychiatry
• Epidemiology

Background:

• Suicide prediction models utilize large-scale data and predictive algorithms to identify at-risk individuals.
• These models are being developed for major healthcare systems like the US Department of Defense and Veterans Affairs.

Purpose of the Study:

• To evaluate the diagnostic accuracy of suicide prediction models.
• To simulate the impact of implementing these models on population-level suicide rates.

Main Methods:

• A systematic literature search was performed across multiple databases (MEDLINE, PsycINFO, Embase, Cochrane Library) up to August 21, 2018.
• 17 cohort studies with 64 unique prediction models and over 14 million participants were included.
• Two reviewers independently screened and evaluated eligible studies.

Main Results:

• Global classification accuracy of models was generally good (≥0.80).
• However, the predictive validity for suicide mortality was extremely low (≤0.01).
• Simulations indicated very low positive predictive values, even with varied population characteristics.

Conclusions:

• Current suicide prediction models demonstrate high overall classification accuracy but near-zero accuracy in predicting future suicide events.
• Significant concerns remain, hindering their readiness for widespread clinical application in healthcare systems.