ChatGPT and Other Large Language Models for Parents' Questions About Childhood Asthma: A Comparative Study

Abstract

<bold>Background:</bold> Parents of children with asthma increasingly seek health information online. Large Language Models (LLMs) represent a new way to access medical information, yet the quality of their responses remains unexplored. <bold>Methods:</bold> We evaluated responses from 10 different LLMs (free and paid versions of ChatGPT, Claude, Copilot, and Gemini) to the 10 most common questions about childhood asthma collected from parents. Five pediatric pulmonologists evaluated medical accuracy, appropriateness, and reproducibility of 100 responses using 5-point Likert scales. Additionally, 100 parents assessed comprehensibility and appropriateness of randomly selected responses. <bold>Results:</bold> LLM responses showed good medical accuracy (median 4/5, IQR 4-5), with 91% rated as good or very good. Paid versions performed significantly better than free models (p<0.01). Parents (median age 41 years, 84% mothers) rated responses highly for comprehensibility (93% easy/very easy) and appropriateness (90% appropriate/very appropriate). However, reproducibility was lower, with only 48% of responses maintaining similarity after one month. <bold>Conclusions:</bold> LLMs provide medically accurate and comprehensible responses to parents' questions about childhood asthma, though response consistency over time remains a challenge.

Ähnliche Arbeiten