Polyphenic risk score shows robust predictive ability for long-term future suicidality.

Area of Science:

• Psychiatry
• Public Health
• Computational Medicine

Background:

• Suicides are preventable public health crises.
• Accurate risk assessment is crucial for effective suicide prevention.
• The Convergent Functional Information for Suicidality (CFI-S) instrument was developed for quantitative suicidality risk assessment.

Purpose of the Study:

• To report a four-year follow-up of patients assessed with the CFI-S in an urban Emergency Department.
• To evaluate the predictive accuracy of the CFI-S for suicidality occurrence, severity, and imminence.
• To enhance the predictive capabilities of the CFI-S using machine learning approaches.

Main Methods:

• A naturalistic cohort of 482 patients from an urban Emergency Department were followed for four years.
• The predictive performance of the CFI-S was assessed using Receiver Operating Characteristic (ROC) curves, Pearson correlation, and Cox regression.
• Machine learning, including deep neural networks and similarity network classifiers, was applied to a discovery and testing cohort.

Main Results:

• The CFI-S significantly predicted suicidality over four years (occurrence ROC AUC 80%, severity correlation 0.44, imminence HR 1.33).
• Key predictive phenes included feeling useless, past suicidality, and social isolation.
• Machine learning enhanced prediction accuracy (ROC AUC increased to 90%) and enabled risk visualization.

Conclusions:

• The CFI-S is a valuable tool for screening and predicting suicidality.
• Machine learning enhancements can further improve the accuracy of suicidality risk prediction.
• Addressing social determinants identified as top risk factors is vital for suicide prevention efforts.