Application of Large Language Models in Chronic Disease Care: Mixed Methods Systematic Review and Thematic Synthesis

Chronic diseases account for nearly three-quarters of global deaths and demand continuous, personalized long-term management; yet traditional care models often fall short in delivering such sustained support. Large language models, with advanced conversational and analytical capabilities, present promising opportunities to address the problem by offering scalable, interactive support. However, a comprehensive synthesis of evidence across diverse study designs, which moves beyond isolated technical metrics to evaluate large language models through a structured, theory-driven lens, remains limited.

This study aimed to synthesize quantitative, qualitative, and mixed methods evidence on technical performance, application scenarios, and documented challenges of large language models in chronic disease care, and to critically evaluate these findings through a theory-driven, 3D framework informed by Orem's Self-Care Theory.

The mixed methods systematic review adhered to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 and SWiM (Synthesis Without Meta-Analysis) guidelines. A comprehensive computer-based search was conducted across PubMed, Web of Science, Embase, Cochrane Library, CINAHL, Wiley Online Library, SpringerLink, ScienceDirect, China National Knowledge Infrastructure, Wanfang Data, VIP Database, and the China Biology Medicine Disc from inception to April 2026. Two independent researchers performed study screening, data extraction, and quality appraisal using the Mixed Methods Appraisal Tool 2018. Given significant clinical and methodological heterogeneity across the included studies, a quantitative meta-analysis was not appropriate; instead, thematic synthesis was used following Thomas and Harden's 3-step approach, with NVivo 14 (Lumivero) used for line-by-line coding and theme development, all informed by Orem's 3D framework.

A total of 20 studies were included, all rated as moderate or high quality. The thematic synthesis revealed three core themes aligned with the proposed framework: (1) foundational safety, privacy, and fairness (hallucination risks and data concerns); (2) self-care enablement through perceived usefulness (patient education, decision support, and self-management); and (3) design and system integration challenges (readability mismatches and workflow gaps). Patient education and clinical decision support were the most common application scenarios. Key technical enhancements (retrieval-augmented generation [RAG] and fine-tuning) primarily strengthened the second theme, while barriers such as content readability, hallucination risks, and ethical ambiguities limited real-world readiness.

This review is the first to integrate Orem's Self-Care Theory into a 3D evaluative framework for large language models in chronic care, thereby moving beyond fragmented, technology-centric assessments toward a structured, nursing-informed, and theory-driven synthesis. Unlike prior reviews that primarily focused on isolated technical metrics or broad feasibility, this synthesis provides a layered, discipline-grounded evaluation distinguishing foundational safety, self-care enablement, and system integration. These findings show evidence mainly supports the intermediate layer, while safeguards and operational integration remain deficient, guiding nursing research toward safety assurances, health-literacy-adaptive design, and implementation science for equitable, patient-centered care.