AI-Enabled Prescribing in Opioid Use Disorder Care: Decision Support, Risk Scores, Large Language Model Copilots, and Digital Therapeutics (2024-2025).

Abstract

This narrative, practice-oriented review maps the 2024-2025 landscape of AI-enabled prescribing in opioid use disorder (OUD) care, where the evidence base is most developed, while noting cautious extension to other substance use disorders (SUDs). We link concrete clinical decisions to fit-for-purpose tools across four domains: EHR-embedded prediction and safety nudges; PDMP-derived analytics and risk scores; clinician-facing large language model (LLM) copilots; and prescription digital therapeutics, just-in-time adaptive interventions (JITAIs), and personal sensing systems. The regulatory analysis uses U.S. frameworks as worked examples and translates the same governance questions to European settings through the EU AI Act, MDR/IVDR medical software requirements, GDPR, the European Health Data Space, and the HMA-EMA data and AI workplan. Evidence maturity is uneven. EHR-based clinical decision support for naloxone co-prescribing and protocolized buprenorphine initiation has the most established implementation base. PDMP risk scores are clinically visible but incompletely validated and should remain advisory. Digital therapeutics and JITAI approaches may support engagement and retention, but they should not automate dosing. LLM copilots are most defensible as audited, citation-first conformance checkers rather than autonomous prescribers; patient-facing agents should be restricted to education, skills support, navigation, and escalation. Across settings, the highest-yield systems are transparent, narrowly scoped, workflow-native, and evaluated with prescribing-centered outcomes such as dose optimization, 28/90/180-day retention, naloxone co-prescribing and fills, and overdose-related safety. We propose safeguards that keep AI explainable, equitable, transferable across health systems, and subordinate to clinical judgment.