This review article maps recent progress in large language models, focusing on their development, scaling, and specialization for medical use. It summarizes how pre-training on massive textual corpora enables these models to understand and generate human-like language and lists emerging trends including tailored medical models, multimodal systems, and mixture-of-expert architectures. The authors position these technological shifts within the pediatric domain and note the need for domain-specific design and evaluation.
Practical pediatric applications are described across several clinical tasks. The review highlights roles for large language models in medical literature retrieval, clinical note generation, diagnostic assistance, intelligent patient communication, personalized educational support, and optimized treatment planning. Specific pediatric uses discussed include dosage calculation, subspecialty clinical decision support, and automated medical record structuring. The authors draw attention to pediatric-specific constraints such as age-dependent variability and the centrality of family-centered care when designing and deploying models.
The paper also surveys evaluation approaches and nontechnical challenges. It examines evaluation metrics and the design of benchmarks, and underscores ethical and legal challenges including safety, equity, and multilingual and low-resource settings. The authors call for interdisciplinary collaboration among clinicians, technologists, ethicists and regulators to create child-specific benchmarks and governance frameworks that reflect pediatric needs.
In conclusion, the review advocates cautious, evidence driven implementation of large language models in pediatric practice and outlines future directions for research and policy to ensure safe and equitable deployment. The article is published as a review in Eur J Pediatr 2025 Dec 1;184(12):809 with doi: 10.1007/s00431-025-06602-x. The authors declare no competing interests.
