Comprehensive Summary
This study investigates how interpretable machine learning methods can predict suicide risk in patients diagnosed with major depressive disorder (MDD). The researchers analyzed data from 273 Greek patients, both inpatients and outpatients, using a combination of sociodemographic, clinical, and psychological variables, including depression severity, attachment style, and self-compassion. They tested six popular classifiers (LightGBM, XGBoost, Random Forest, SVM, Logistic Regression, and Multilayer Perceptron) through comparative analysis and post-hoc explainability using SHAP values. Among these, the LightGBM model achieved the best performance with a ROC-AUC of 0.895, identifying both risk and protective factors for suicidality. High depression scores, early onset of depressive symptoms, insecure attachment, and a history of suicide attempts were the strongest predictors of suicide risk, while psychotherapy, employment, and family support acted as protective factors. The explainable AI framework provided interpretable insights, highlighting how different variables influenced predictions across individual patients.
Outcomes and Implications
This study demonstrates the potential of interpretable AI to enhance suicide risk assessment in patients with MDD, offering clinicians a transparent and data-driven way to identify individuals at highest risk. By combining psychological factors like attachment style and self-compassion with clinical history, AI tools could improve early intervention and personalize treatment plans. The explainability component, achieved through SHAP analysis, allows practitioners to understand not only which factors contribute to risk but also why they contribute to risk, fostering trust and ethical integration of AI in psychiatric care. Although promising, the authors emphasize the need for broader validation across diverse populations and cultural contexts. With further refinement, interpretable machine learning models could become valuable adjuncts to clinical decision-making, improving prevention strategies and ultimately saving lives.