Large Language Models in Chronic Obstructive Pulmonary Disease Care and Health Management: A Scoping Review

Abstract

Background: Chronic obstructive pulmonary disease (COPD) requires sustained patient education, self-management, medication use, pulmonary rehabilitation, symptom monitoring, and guideline-concordant clinical management. Large language models (LLMs) may expand access to conversational health information, but their value depends on more than factual accuracy. Objective: To map empirical evidence on LLMs and LLM-based generative artificial intelligence systems in COPD-related care and health management, summarize application domains and evaluation outcomes, and identify priorities for future research. Methods: A scoping review was conducted in accordance with Joanna Briggs Institute guidance and reported using PRISMA-ScR. PubMed/MEDLINE, Embase, Web of Science Core Collection, Scopus, CINAHL, and PsycINFO were searched from inception through September 10, 2026. Empirical studies evaluating an LLM or LLM-based generative AI system in COPD-related patient education, health information, pulmonary rehabilitation, medication management, or guideline-based clinical management were eligible. Data were charted and synthesized descriptively and thematically. Results: The search identified 327 records; 34 duplicates were removed, and 293 records underwent title/abstract screening. Of these, 281 were excluded, and 12 full-text reports were assessed. Four full-text reports were excluded, leaving eight studies. Five studies focused primarily on patient education or health information, one on pulmonary rehabilitation, one on COPD medication management, and one on guideline-based clinical management. Accuracy and information quality were the most common outcomes. LLMs often produced clinically acceptable responses, but limitations included incomplete information, high reading grade levels, outdated guideline references, response variability, and differences across models and task formats. Only two studies directly involved people with COPD, and one of these demonstrated a marked gap between favorable expert ratings and patients’ ability to formulate disease-specific queries. Conclusions: Evidence on LLMs in COPD is expanding but remains dominated by model-response evaluation. Current findings support LLMs as supplementary, human-supervised tools rather than autonomous sources of COPD management advice. Future research should prioritize actionability, real-world usability, health literacy, equity, behavioral outcomes, clinical outcomes, and safety under routine patient use.