LLMs compared on pediatric infection questions

Aykac and colleagues compared LLM answers to pediatric infectious disease questions. They found no statistically significant differences between models in domain scores. Pediatric Research published the study.

The team tested ChatGPT-5, DeepSeek-V3.2-Exp and Gemini 2.5 Flash on 12 clinician-written scenarios. Each scenario included open questions about cause, clinical features, diagnosis, treatment and complications. Two pediatric infectious disease specialists scored the responses independently.

Answer scores

ChatGPT-5 scored 10.0 out of 10 in every domain. DeepSeek scored 9.83 for diagnosis and 9.92 for treatment; Gemini scored 9.75 and 8.83, respectively. All domain comparisons had p values above 0.05. Comparisons of accuracy, completeness and safety also showed no significant differences.

Twelve simulated scenarios cannot establish model equivalence from nonsignificant results. Specialists rated safety in the written answers; the researchers did not measure adverse events in patients.

Study notes

This is a comparative evaluation of answers to simulated cases. The public abstract does not provide funding or conflict-of-interest details.

Sources

1. [1]
   論文：Aykac et al. Performance of large language models in pediatric infectious disease education: a multi-condition comparative study
2. [2]
   データベース：PubMed 公開要旨

Images: generated with GPT Image 2.5 (illustrative).