<![CDATA[Hypertension and type 2 diabetes mellitus (T2DM) are two non-communicable diseases with a high burden in primary healthcare and have become important areas of research on clinical decision support using large language models (LLMs) since 2024. Retrieval-Augmented Generation (RAG) is increasingly used to ground LLM outputs in clinical guidelines, yet the specific contribution of embedding models as an independent variable remains underexplored. This scoping review aimed to map the 2024–2026 literature on the use of LLMs to support clinical decision-making for hypertension and T2DM and to identify existing research gaps. The review followed the Arksey and O’Malley framework and PRISMA-ScR reporting guidelines. Searches were conducted in Google Scholar, PubMed/PubMed Central, arXiv, and IEEE Xplore using Boolean combinations of terms related to LLMs, RAG, hypertension, and T2DM. Fifteen studies met the inclusion criteria. The findings indicate that RAG consistently improves answer accuracy and reduces hallucinations compared with prompting-only approaches, although its benefits decrease as the capabilities of the underlying model increase. Only one study directly manipulated the embedding model as an experimental variable and found trade-offs between general and domain-specific embeddings in retrieval sensitivity and specificity. No studies evaluated RAG or embedding models using Indonesian clinical texts. These findings support the potential of RAG for clinical decision support while highlighting the need for further research on embedding models in the Indonesian medical context.]]>
Copyrights © 2026
