Explainable artificial intelligence models for predicting risk of suicide using health administrative data in Quebec

Area of Science:

• Public Health
• Data Science
• Mental Health Research

Background:

• Suicide is a complex global challenge requiring advanced prevention strategies.
• Artificial intelligence (AI) and machine learning (ML) offer potential for enhanced suicide risk detection.
• Interpretable AI is crucial for enabling data-driven decisions in mental health services planning.

Purpose of the Study:

• To develop and interpret sex-specific ML models for predicting population suicide risk.
• To utilize a large-scale dataset from the Quebec Integrated Chronic Disease Surveillance System (QICDSS).
• To identify key individual, programmatic, systemic, and community factors associated with suicide risk.

Main Methods:

• A case-control study design using QICDSS data (2002-2019).
• Inclusion of 103 features and validation of supervised ML algorithms (LR, RF, XGBoost, MLP).
• Application of Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• Sex-specific ML models were developed, with varying sensitivity and precision (PPV) for males and females.
• XGBoost demonstrated the best precision for both sexes.
• The study highlights the potential of explainable AI in population-level suicide prevention.

Conclusions:

• Explainable AI models can serve as valuable tools for informing suicide prevention policies and actions.
• Routine data on individual, programmatic, systemic, and community factors are useful for predictive modeling.
• Future work includes developing user-friendly interfaces for stakeholders to facilitate decision-making.