LARGE LANGUAGE MODELS IN MEDICATION MANAGEMENT: CLINICAL UTILITY, SAFETY ASSURANCE, AND PHARMACIST-LED GOVERNANCE

Abstract

Large language models (LLMs) can produce fluent medication explanations, draft patient-facing materials, summarize records, and assist with information retrieval. Their linguistic competence has encouraged proposals for pharmaceutical-care use, yet fluency is not equivalent to clinical reliability. This integrative narrative review evaluates LLMs as governed components of medication-management systems rather than autonomous digital pharmacists. Evidence was synthesized across pharmacy practice, clinical informatics, patient safety, human factors, and AI governance. The analysis distinguishes low-risk language transformation from high-risk clinical judgment and examines drug information, medication reconciliation, interaction screening, adherence support, pharmacovigilance, documentation, education, and multilingual communication. Published evaluations suggest potential workflow utility in constrained tasks but substantial variation across models, prompts, versions, questions, evaluators, and settings. Hazards include confabulation, fabricated citations, critical omission, automation bias, privacy loss, unequal language performance, model drift, and weak reproducibility. Retrieval-augmented generation can improve traceability but cannot eliminate retrieval or synthesis failure. The Pharmacist-Led LLM Assurance Framework for Medication Management (PLLAMM) organizes implementation around five stages: Govern, Map, Measure, Manage, and Monitor. Current evidence supports only conditional utility under pharmacist supervision for bounded, auditable tasks; it does not justify unsupervised medication counseling, autonomous dosing, interaction management, or therapeutic decision-making.