Artificial Intelligence (AI) for Substance Use Disorder Screening and Management: A Narrative Review of Barriers and Facilitators to Clinical Implementation

Benedictus Atsu Hlormenu, BPHN, RN
University of Miami

Co-authors: Benedictus Atsu Hlormenu¹, Isaac Frimpong¹, Denise C. Vidot^1
1University of Miami

Background: Substance use disorder (SUD) continues to be one of the most pressing global public health challenges, even though machine learning and artificial intelligence (AI) have produced numerous predictive models for SUD risk stratification, screening, and management across alcohol, opioid, and cannabis use disorders. Few of these models have progressed beyond development, validation, and implementation into real-world clinical practice.

Objective: This narrative review synthesizes current evidence on AI applications for SUD and CUD screening and management using a hybrid implementation science framework organized across four domains: data and model characteristics, clinician and workflow integration, patient-level factors, and system-level and regulatory considerations.

Methods: A comprehensive literature search was conducted using goggle scholar, Pubmed, Web of Science, PsyINFO, arXiv and Embase for articles published in English from January 1, 2015 to July 1, 2026 with emphasis on articles published from 2018 related to development, evaluation, implementation, deployment, or acceptance of AI or machine learning tools relevant to substance use disorder screening or management.

Results: At the data and model level, opioid use disorder has the most advanced evidence base for predictive modeling and fairness evaluation, whereas alcohol use disorder has a larger but methodologically heterogeneous literature with substantial validation limitations. Cannabis use disorder remains at an earlier stage related to its clinical implementation focused models. Though not limited in methodological rigor, most cannabis predictive studies rely on smaller, single site samples failing the AI predictive modeling and fairness evaluation across a broader spectrum, creating a gap in implementation into daily clinical practice. Few studies include multi level datasets, varied demographic populations, real-time evaluation, or standardized validation frameworks, impending replicability, and evidence-based clinical translation in patient care. This limitation therefore reduces model generatability, creates weak external validation, limits fair clinical applicability, and ultimately slows cannabis AI-driven decision support tools for screening and management of cannabis use disorder. Clinician and workflow factors highlight the importance of EHR integration while patient-level implementation is shaped by stigma, concerns about data sharing, digital access, and trust. System-level implementation is further influenced by the unique regulatory requirements governing substance use disorder data confidentiality under 42 CFR Part 2.

Conclusion: AI tools for SUD screening and management show substantial predictive potential, but their integration into routine care remains limited, especially with respect to cannabis use. Successful implementation requires further research into external validation, systematic evaluation across diverse populations, and continued adaptation to the evolving regulatory landscape surrounding substance use disorder data.