Safety, Accuracy, and Reliability of Large Language Models for Medication Advice and Drug Information: A Mini Systematic Evidence Map (2023–2026)

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

Background: Large language models (LLMs) are increasingly used to answer medication-related questions, but concerns remain regarding the accuracy, completeness, reliability, and safety of patient-facing drug information. Objective: To map recent evidence evaluating LLM-generated medication advice and drug information. Methods: A rapid mini systematic evidence map of English-language original studies published from 2023 through August 2026 was conducted using structured PubMed and publisher searches. Eligible studies assessed generative LLMs in medication counseling, drug information, medication safety, over-the-counter medication use, patient instructions, or related contexts and reported accuracy, safety, completeness, clarity, or reliability outcomes. Results: Ten original studies were mapped. Evidence covered general drug information, telepharmacy, OTC medication counseling, renal disease, pregnancy, medication instructions, and oral anticoagulant therapy. Newer and knowledge-grounded models generally performed better than earlier ChatGPT versions, but clinically important omissions, inconsistent safety classifications, incomplete counseling, and potential risks of harm remained. Conclusion: Large language models show potential as supportive medication-information tools, but current evidence does not support their use as standalone sources for clinically consequential medication decisions. Verification by qualified healthcare professionals or authoritative drug-information resources remains important.

Links and sources

Comments