Abstract: This study aims to break through the limitations of traditional static case design by constructing a ″dynamic prompt model″ based on large language models and exploring its application and effectiveness in pediatric medical education. The model encompasses disease course progression information (prodromal, acute, and recovery phases), a complete clinical decision-making chain (history taking, physical examination, auxiliary investigations, diagnosis, and treatment), and progressively advanced cognitive objectives (remembering, understanding, applying, analyzing, and evaluating). The disease list includes two categories: common pediatric diseases and rare diseases. The study selected 45 fourth-year pediatric medical students from China Medical University as research participants to conduct a 4-week learning program. The results indicate that after learning with dynamic cases, the virtual case scores significantly increased from (75.31±15.21) to (82.22±11.43). Critical thinking ability improved from (102.67±10.93) to (110.13±12.61), with system analysis skills rising from (42.85±3.91) to (45.13±4.61) and knowledge exploration willingness increasing from (31.55±3.74) to (36.32±5.11), all P<0.05. The realism scores of different cases were relatively high (all>4), and the difficulty ratings were reasonably distributed. In the experience evaluation, the ″enhancing learning″ dimension scored the highest (4.02±0.47). Therefore, the ″dynamic prompt model″ effectively improved medical students′ clinical thinking and critical thinking abilities, providing an efficient and innovative tool for pediatric medical education.

Key words: Pediatrics, Large language models, Dynamic prompts, Virtual diagnosis and treatment, Clinical reasoning ability