LLMs Show Promise for Personal Health Record Queries
A study on arXiv (2605.18937) evaluates using large language models (LLMs, specifically Gemini 3.0 Flash) to answer patient health queries based on Personal Health Records (PHRs). Researchers tested 2,257 user queries from three distributions: short web searches, longer chatbot-style questions, and patient calls to healthcare teams. Queries were matched with de-identified PHRs from a pool of 1,945. Gemini responses were generated without PHR context, with a basic summary, and with full clinical notes. Evaluation used the SHARP rating framework. The study assesses LLMs' potential to empower patients by simplifying complex PHR data.
Key facts
- arXiv paper 2605.18937
- Uses Gemini 3.0 Flash LLM
- 2,257 user queries from 3 distributions
- 1,945 de-identified PHRs
- Three context levels: none, basic summary, full clinical notes
- Evaluated with SHARP framework
- Focus on patient health queries
- Assesses PHR utility for personalized health AI
Entities
Institutions
- arXiv