[Artificial intelligence-enabled clinical decision support systems for suicide risk assessment in emergency departments using Quebec health administrative data]

The integration of artificial intelligence (AI) into mental health services has generated growing interest regarding its potential role in suicide risk assessment and prevention. Drawing on recent research using Quebec's health administrative data, this article examines the potential of clinical decision support systems (CDSS) to strengthen population-based suicide prevention strategies, particularly in high-risk settings such as emergency departments. We distinguish between individual suicide risk prediction, population risk stratification, and clinical decision support systems, which use these data to support-rather than replace-clinical judgment. Although predictive models may improve the anticipation of adverse outcomes, their implementation raises important ethical, organizational, and clinical challenges. Three issues are particularly salient: the opacity of algorithmic models and the need for explainable AI, the risk of reproducing social inequalities embedded in the data used to train predictive models, and the challenges associated with the clinical acceptability and integration of these tools into routine practice. We argue that AI should not be viewed as a substitute for clinical judgment but rather as a complementary tool capable of enhancing suicide prevention and continuity of care. Despite promising developments, current predictive models remain of limited accuracy for individual suicide prediction, supporting their use only as an aid to clinical decision-making. Furthermore, successful implementation requires transparent governance, appropriate professional training, and an interdisciplinary approach that explicitly considers the social determinants of health. This approach should also be guided by the principles of patient partnership, as articulated in the Montreal Model, ensuring that people with lived experience actively participate in the development, evaluation, and continuous improvement of these technologies. From this perspective, the development of clinical decision support systems capable of supporting suicide risk stratification may contribute to improving patient safety, provided that their implementation remains grounded in the fundamental values of patient care and guided by rigorous ethical oversight. Future research should clarify the medico-legal responsibilities associated with following-or not following-algorithmic recommendations. It should also evaluate these systems using multiple outcomes, including clinical effectiveness, continuity of care, healthcare utilization, and acceptability among both clinicians and patients. In addition, implementation strategies will need to address integration with electronic health records, regulatory requirements, model maintenance, and implementation costs. Finally, the anticipated benefits of these systems will need to be confirmed through prospective implementation studies.