A Survey of LLM Reasoning in Healthcare and Medicine: from Individual Modeling to Collaborative Agents

The rapid advancement of Large Language Models (LLMs) has initiated a profound transformation in intelligent healthcare and medicine. However, the successful clinical deployment of these models is significantly dependent on their ability to perform reliable and explainable medical reasoning rather than surface-level text generation. This survey provides a comprehensive overview of the current landscape of LLM reasoning in the medical domain, systematically reviewing foundational datasets, training-time optimization strategies, and test-time reasoning methodologies. Initially, research efforts focused on individual modeling, treating a single LLM as an isolated expert that performs evidence synthesis and hypothesis generation within a monolithic architecture. While this approach improved interpretability through techniques like Chain-of-Thought (CoT) prompting, it faced inherent limitations in robustness and handling complex, multi-hop clinical inferences. Consequently, the field is witnessing a strategic technical transition toward collaborative agents. In this emerging paradigm, multiple LLM-based agents are assigned complementary roles (e.g., diagnosis, evidence retrieval, and verification) to interact through structured communication. This collaborative framework more closely mirrors realworld clinical practice, offering a system-level approach to enhancing transparency, error recovery, and decision-making stability in AI-assisted healthcare.