Personalizing communication with the patient: large language models

The paper analyzes the role and prospects of large language models (LLMs) in the transformation of modern healthcare, with a focus on improving doctor-patient interactions. The spectrum of LLM applications is considered: from automating administrative tasks to supporting patients in self-education and managing their health. It reveals the ability to semantic adaptation, to translate complex medical terminology into a language understandable to the patient, which supports the concept of shared decision-making. Practical cases of LLM application are highlighted, including monitoring chronic disease, supporting adherence to drug therapy, and providing instructions in emergency situations. Accepting the problems of accuracy of publicly available LLMs, the possibility of generating false information (“hallucinations”), data bias, and ethical and regulatory challenges related to data privacy and accountability for the information provided, technological aspects such as Retrieval-Augmented Generation search architectures and Chain of Thought techniques to improve the accuracy and clinical relevance of LLM-generated responses, and voice interfaces as a means of improving the accessibility of these technologies to diverse populations are disclosed. The need for interdisciplinary collaboration and a clear regulatory framework for safe and effective implementation of technologies in clinical practice is emphasized.