Extracting Multifaceted Characteristics of Patients With Chronic Disease Comorbidity: Framework Development Using...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Chronic Disease Management

Background:

• Chronic multimorbidity research is vital due to population aging.
• Current patient data extraction methods are inefficient and error-prone.
• Accurate patient characteristic extraction is a key challenge in comorbidity studies.

Purpose of the Study:

• To develop a comprehensive framework for extracting patient demographic and disease characteristics in multimorbidity.
• To leverage large language models (LLMs) for feature extraction from electronic health records (EHRs).
• To evaluate the framework's performance across seven key feature dimensions.

Main Methods:

• Developed a multifaceted feature extraction framework for multimorbidity patients.
• Utilized large language models (LLMs) for prompt-based feature extraction from EHRs.
• Conducted rigorous manual verification for straightforward and intricate features, assessing outcomes quantitatively and qualitatively.

Main Results:

• Achieved an overall F1-score of 99.6% for straightforward feature extraction (e.g., basic information at 100%).
• Demonstrated an overall F1-score of 94.4% for intricate feature extraction.
• Highlighted accurate information extraction as a key advantage, while noting format consistency as a challenge.

Conclusions:

• The scalable framework integrates EHR data for 1225 multimorbidity patients across 41 diseases.
• Enables efficient, zero-sample target feature extraction using LLMs for research and policy.
• Significantly improves research efficiency, reduces labor, and enhances study reliability through high accuracy.