New chapter in pediatric medicine: technological evolution, application, and evaluation system of large language models

With significant breakthroughs in natural language processing technology, large language models (LLMs) based on deep learning have demonstrated considerable potential in the medical field in recent years. Through pre-training on massive textual corpora, these models are capable of understanding and generating human-like language, providing innovative tools for tasks such as medical literature retrieval, clinical note generation, and diagnostic assistance. In particular, within the domain of pediatrics, LLMs offer promising applications for enhancing the efficiency and safety of diagnosis and treatment through intelligent patient communication, personalized educational support, and optimized treatment planning. This article reviews recent advancements in LLM technology, encompassing the developmental trajectory and scaling of general-purpose models, the tailored training of medical specialized models, and the emergence of multimodal and mixture-of-expert architectures. It further highlights practical applications in pediatric contexts, including dosage calculation, subspecialty-specific clinical decision support, and automated medical record structuring, while also examining evaluation metrics, ethical-legal challenges, and considerations for multilingual and low-resource settings.

the paper emphasizes the importance of interdisciplinary collaboration and outlines future directions for safely and equitably integrating LLMs into pediatric medical practice.

• Large language models (LLMs) are increasingly used in medicine for text generation, clinical documentation, and knowledge retrieval. • Pediatric applications of LLMs have been less systematically reviewed compared with those in adult medicine.

• This review provides an integrative overview of LLM development, clinical implementation, and evaluation in pediatric contexts. • It identifies unique challenges in pediatrics, including age-dependent variability and the need for family-centered care, and proposes design principles for future child-specific LLM benchmarks.