Truace
Subscribe

Performance of large language models in pediatric infectious disease education: a multi-condition comparative study

Background: Evaluating how appropriate current large language models (LLMs) respond to common pediatric infectious disease (PID) questions is important for understanding their potential role in education and clinical support. Therefore, we evaluated and compared the clinical quality of answers produced by different LLMs when given professionally written, clinician-style case descriptions, using a purpose-built multi-domain evaluation tool. Methods: Three different LLMs, including DeepSeek-V3.2-Exp, ChatGPT-5, an…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1803· cite & verify
Performance of large language models in pediatric infectious disease education: a multi-condition comparative study
Reading Wikipedia in the Classroom for Secondary School Students 11 by James Rhoda. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers tested DeepSeek-V3.2-Exp, ChatGPT-5 and Gemini 2.5 Flash on 12 clinician-style pediatric infectious disease scenarios. Each had five open-ended questions, and two specialists scored the answers on a Likert-based tool across five clinical domains.

The results suggest current chatbots can organize clinical information well in standardized cases, which could help medical education. The scenarios were structured and expert-written, so the findings do not show how the models handle messy real-world care, and the authors stress clinician supervision.

Main points

  1. ChatGPT-5 received perfect domain scores of 10.0 across the five domains.
  2. DeepSeek-V3.2-Exp had minor reductions in diagnostics (9.83) and treatment (9.92).
  3. Gemini 2.5 Flash varied more, scoring 8.83 in treatment and 9.75 in diagnostics.
  4. No domain-level differences between models were statistically significant (all p > 0.05).

The gain

In 12 structured pediatric infectious disease scenarios, three LLMs earned high expert ratings for accuracy, completeness and safety, suggesting a role in education and early clinical reasoning under supervision.

The rundown

The study used 12 pediatric infectious disease scenarios, each with five questions on etiology, clinical features, diagnostics, treatment options and complications. Two pediatric infectious disease specialists independently reviewed every response.

Model-level scores for ChatGPT-5, DeepSeek-V3.2-Exp and Gemini 2.5 Flash were accuracy 6.0, 5.98 and 5.85, completeness 3.0, 2.97 and 2.87, and safety 1.0, 1.0 and 0.97. The accuracy, completeness and safety differences were not significant (p = 0.074, 0.066 and 0.14).

The authors conclude that the best use is within clinician-supervised workflows, where AI outputs complement rather than replace human clinical judgment.

Sources

  1. Peer-reviewedPediatric Research2026-10-17

The debate