Real-world user interactions with large language models for explaining home electrocardiogram reports

With the development of portable electrocardiogram (ECG) devices, users can conveniently obtain physiological data in home settings, yet they still face an interpretation gap because of limited medical knowledge. Large language models (LLMs) may help bridge this gap through natural-language explanations, but large-scale real-world evidence on how users interact with them in home settings remains limited. This study retrospectively analyzed 109,017 conversational sessions comprising 323,470 dialogue turns from 26,419 users of the Wen-Xin-Wu-Yang system over 16 months, examining user engagement, intent evolution, confirmation and concern expressions, and interaction repair. 60.01% of interactions occurred outside regular working hours, and 57.90% of users were from lower-tier cities and regions. Users demonstrated relatively high retention and moderate interaction depth, while their needs gradually extended from result interpretation to medical knowledge, action-oriented guidance, and health management. During Action-to-Knowledge transitions, explicit confirmation of system reliability decreased, whereas concern expressions increased during the middle stages of conversations before partially declining. In uncertain interactions, users tended to simplify their expressions. Overall, LLMs integrated into AI-ECG systems can connect physiological data with medical knowledge and these findings provide design implications for progressive explanation, recognition of risk concerns, support for interaction repair, and continued physician review.

Verbatim fromOpenAlex· Abstract via OpenAlex (publisher-deposited); © the authors/publisher