Comprehensive Literature Review: Large Language Models in Medical Decision Support Systems

Abstract

Large Language Models (LLMs) have become a game-changing artificial intelligence (AI) technology that can revolutionize clinical decision support systems (CDSS) by leveraging its ability to process and understand natural language, to reason, and to interpret multimodal data. This review offers a detailed analysis of how LLMs are currently applied to medical decision-making tools including diagnosis, treatment planning, interpretation of medical imaging, clinical documentation, medication safety and personalized healthcare suggestions. This study aims to critically discuss the recent developments regarding Retrieval-Augmented Generation (RAG), Chain-of-Thought (CoT) reasoning, knowledge graph incorporation, multimodal fusion, and multi-agent collaborative AI systems, which are all aimed at boosting the clinical accuracy, minimizing hallucinations, and increasing explainability of AI systems. Moreover, the review explores the key limitations of LLMs in healthcare, such as potential biases, hallucinations, interpretability, regulatory compliance, ethical issues, privacy protection, and robustness in face of dataset shift. A comparative study of current medical LLM models reveals that, although these systems offer substantial advances in diagnostic reasoning and streamlining workflows, the role of the clinician remains indispensable, highlighting the need for a human-centric approach to AI in healthcare. Finally, future research directions are discussed to facilitate next-generation intelligent healthcare infrastructures for reliable and clinically deployable applications: uncertainty-aware reasoning, domainspecialized medical LLMs, trustworthy explainable AI, federated medical LLMs, and collaborative multi-agent healthcare systems.