A Dataset and Resources for Identifying Patient Health Literacy Information from Clinical Notes.

Area of Science:

• Health Informatics
• Natural Language Processing
• Clinical Documentation

Background:

• Health literacy significantly impacts patient outcomes.
• Existing screening tools for health literacy are often impractical and inconsistently applied.
• Difficulty in documenting health literacy in structured electronic health records hinders care.
• Unstructured clinical notes offer rich, contextual health literacy data but lack annotated resources.

Purpose of the Study:

• To address the need for annotated resources for automated health literacy detection.
• To introduce HEALIX, the first publicly available annotated health literacy dataset from clinical notes.
• To facilitate the development of advanced natural language processing (NLP) models for health literacy assessment.

Main Methods:

• HEALIX dataset curated using social worker note sampling, keyword filtering, and LLM-based active learning.
• Dataset comprises 589 clinical notes across 9 types.
• Notes annotated with three health literacy levels: low, normal, and high.
• Benchmarking of zero-shot and few-shot prompting strategies on four open-source LLMs.

Main Results:

• The HEALIX dataset provides a valuable resource for NLP research in health literacy.
• Demonstrated the utility of HEALIX for benchmarking LLM performance in health literacy classification.
• Initial benchmarking showed varying performance of LLMs based on prompting strategies.

Conclusions:

• Automated health literacy detection from clinical notes is a feasible and promising approach.
• The HEALIX dataset is crucial for advancing research and development in this area.
• Facilitates improved documentation and understanding of patient health literacy within electronic health records.