Advancing Psychiatric Safety With the Predictive Risk Identification for Mental Health Events Tool: Retrospective...

Area of Science:

• Psychiatric Informatics
• Machine Learning in Healthcare
• Patient Safety Technology

Background:

• Patient safety incidents are a significant concern in psychiatric settings, with current early warning systems (EWS) being underdeveloped and often reactive.
• Traditional risk assessment tools, like the Dynamic Appraisal of Situational Aggression-Inpatient Version (DASA-IV), demonstrate limited predictive accuracy for adverse events.

Purpose of the Study:

• To introduce the Predictive Risk Identification for Mental Health Events (PRIME) tool, a novel deep learning-based EWS.
• To utilize longitudinal psychiatric electronic medical record (EMR) data for anticipating adverse events within 24-hour windows.

Main Methods:

• A retrospective cohort study at Waypoint Centre for Mental Health Care, including 4651 patients and over 400,000 encounters.
• Training recurrent neural networks with attention mechanisms on multivariate time-series EMR data for risk forecasting.
• Benchmarking PRIME against the DASA-IV tool and other machine learning models using AUC and recall metrics.

Main Results:

• The long short-term memory network with an attention mechanism achieved a high predictive performance with an AUC of 0.83.
• PRIME demonstrated a significant improvement over the DASA-IV tool, with an AUC difference of 0.20 (0.83 vs 0.61).
• The study highlighted the potential of machine learning models for proactive risk management in mental health.

Conclusions:

• PRIME is among the first deep learning-based EWS evaluated for psychiatric inpatient care, offering interpretable, rolling predictions.
• The tool shows a pathway toward safer and more proactive mental health interventions by outperforming existing clinical tools.
• Future research should focus on assessing PRIME's equity implications and its integration into clinical workflows.