Related Experiment Video
Updated: Sep 12, 2025
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
682
Mobility Functional Status Ascertainment in Electronic Health Records using Large Language Models
Xingyi Liu1, Muskan Garg1, Heling Jia1
1Mayo Clinic.
Research Square
|August 6, 2025
Summary
Large Language Models (LLMs) can accurately extract patient mobility status from clinical notes in Electronic Health Records (EHRs). This approach enhances precision medicine by standardizing functional status data for research and clinical use.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Healthcare
- Geriatric Medicine
Background:
- Global population aging necessitates precise functional status assessment for personalized medicine.
- Electronic Health Records (EHRs) contain rich, yet largely unstructured, patient mobility data.
- Standardizing mobility information from clinical notes is crucial for clinical applications and research.
Purpose of the Study:
- To investigate the efficacy of Large Language Models (LLMs) in extracting and standardizing patient mobility status from unstructured EHR clinical notes.
- To evaluate different LLM prompting strategies for mobility data extraction and impairment classification.
- To assess the trustworthiness and generalizability of an LLM-based approach across multiple healthcare institutions.
Main Methods:
- Annotation of 600 clinical notes from three healthcare institutions focusing on mobility and impairment expressions.
- Utilizing the open-source Llama 3 model with zero-shot, few-shot, and task decomposition prompting techniques.
- Performance evaluation through error analysis and calculation of patient-level accuracy and F1-scores for mobility extraction and impairment classification.
Main Results:
- Mobility Extraction achieved a micro-average accuracy of 0.952 and an F1-score of 0.962.
- Impairment Classification achieved a micro-average accuracy of 0.912 and an F1-score of 0.948.
- Error analysis indicated clinically reasonable inferences, even in ambiguous cases, with a local, deterministic setup enhancing trustworthiness and generalizability.
Conclusions:
- LLM-based solutions are feasible for extracting functional mobility status from unstructured EHR data.
- This methodology supports the integration of mobility data into precision medicine initiatives and clinical research.
- The developed approach demonstrates cross-institution generalizability and enhances data privacy and consistency.
