LLMs score highly in expert-reviewed pediatric infection scenarios

Researchers compared ChatGPT-5, DeepSeek-V3.2-Exp and Gemini 2.5 Flash using 12 clinician-style scenarios covering six pediatric infectious disease conditions. Each scenario included questions on etiology, clinical features, diagnostics, treatment and complications, for 60 questions in total. Two pediatric infectious disease specialists independently assessed the responses using a Likert scoring tool. The study used no patient data or human participants.

All three models received high scores across the evaluated domains. ChatGPT-5 scored 10.0 in each domain; DeepSeek-V3.2-Exp scored 9.83 for diagnostics and 9.92 for treatment, with 10.0 in the other domains. Gemini 2.5 Flash scored 8.83 for treatment and 9.75 for diagnostics, and 10.0 for clinical features and complications. Differences between models were not statistically significant. Model-level accuracy, completeness and safety comparisons were also not significant.

The findings suggest these models can organize information in structured educational cases, but do not establish performance in real patient care. The assessment was limited to 12 scenarios and expert ratings; the authors say LLMs should complement, not replace, clinician judgment.