CoChat

Mobility functional status ascertainment in electronic health records using large language models

Open access
Liu, Xingyi, Garg, Muskan, Jia, Heling, Sauver, Jennifer St., Pagali, Sandeep R., Sohn, Sunghwan · 2026 · Scientific Reports

Published version of a preprint

The DOI matches its Crossref record; 24 of the 24 references it cites were found in a registry or an index; 3 of the 17 sources its claims cite were checked for support.

Abstract

This study explores using Large Language Models (LLMs) to extract and standardize mobility status from unstructured EHR data (i.e., clinical notes). We annotated 600 clinical notes from three healthcare institutions located in southeastern Minnesota and west-central Wisconsin, focusing on expressions of mobility and associated impairment. Leveraging the open-source Llama 3 model, we tested various prompting strategies, including zero-shot, few-shot, and task decomposition, and evaluated their performance. Our final model achieved a patient-level micro-average F1-score of 0.876 [95% CI 0.858–0.894] for Mobility Extraction and 0.897 [95% CI 0.878–0.917] for Impairment Classification. A local, deterministic setup improved trustworthiness by ensuring consistent outputs, safeguarding privacy, and demonstrating cross-institution generalizability. These findings highlight the feasibility of LLM-based solutions for extracting mobility functional status from unstructured EHR data, supporting both clinical applications and research.

Overview

CoChat Summary

TL;DR

This study used the open-source Llama 3 model with prompt engineering to extract and classify mobility functional status from 600 annotated clinical notes across three healthcare institutions. The best standalone configuration was error-informed prompt refinement, which reached patient-level micro-average F1-scores of 0.876 for Mobility Extraction and 0.897 for Impairment Classification.

Key findings

  • The dataset included 600 clinical notes from 256 unique patients; the cohort had a mean age of 75.6 years (SD 13.2) and was 53.5% female (137 patients).
  • Under the baseline zero-shot setup, section-level micro-average F1-scores for Mobility Extraction were 0.663, 0.638, and 0.516 for Sites 1–3; for Impairment Classification, they were 0.888, 0.857, and 0.833.
  • The best configuration was Error-Informed Prompt Refinement; its section-level micro-average F1-scores for Mobility Extraction were 0.729, 0.723, and 0.608 at Sites 1–3, and for Impairment Classification were 0.832, 0.814, and 0.785.
  • At the patient level, the final model achieved micro-average F1-scores of 0.876 [95% CI 0.858–0.894] for Mobility Extraction and 0.897 [95% CI 0.878–0.917] for Impairment Classification; under the paper’s relaxed evaluation, these rose to 0.962 and 0.948, respectively.

Methods

  • Design: Cross-institution evaluation of an LLM-based annotation pipeline using Llama 3 with zero-shot, few-shot, error-informed prompt refinement, and task decomposition strategies.
  • Data/sample: 600 notes from three institutions in southeastern Minnesota and west-central Wisconsin; after sectioning, the dataset comprised 3,810 sections (1,153 Site 1, 1,075 Site 2, 1,582 Site 3). The notes included 200 PT/OT notes each from Sites 1 and 2 and 200 unrestricted clinical notes from Site 3.
  • Main analysis: Two tasks were evaluated: Mobility Extraction (Mentioned vs Not Mentioned) and Impairment Classification (Impaired vs Unimpaired). Performance was measured with precision, recall, F1-score, micro-averages, and 95% bootstrap confidence intervals, at section-, note-, and patient-levels.

Limitations

  • The paper states there was no direct comparison with conventional clinical NLP systems such as cTAKES or CLAMP.
  • The paper states the model was not fine-tuned on the annotated dataset; this was a deliberate choice because the dataset was relatively small and fine-tuning could overfit or be computationally expensive.
  • Evident from the text: the development/prompt refinement was done on Site 1, so generalizability beyond the two validation sites is promising but still limited to the specific institutions and note types studied.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-026-37025-9

Cites

24 of 24 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Analysis of mobility classes at section- and note-levels and distributions of sections for mobility classes. The mobility classes are abbreviated as: CB (“Changing and maintaining body position”), CO (“Carrying, moving, and handling objects”), WM (“Walking and moving”), MT (“Moving around using transportation”), and MU (“Mobility, unspecified”). (a) UpSet plot of sections by mobility class combination (n = 3,810). (b) UpSet plot of notes by mobility class combination (n = 600). (c) Pie chart of the proportion of sections mentioning 0–5 classes: 57.0% none, 23.1% one, 12.8% two, 5.8% three, 1.2% four, 0.3% five. (d) Pie chart of the proportion of notes mentioning 0–5 classes: 48.0% none, 29.3% one, 9.0% two, 8.5% three, 4.5% four, 0.7% five. (e–i) Distribution of sections for CB, CO, WM, MT and MU, respectively, showing the three most frequent sections.
Fig. 1 Analysis of mobility classes at section- and note-levels and distributions of sections for mobility classes. The mobility classes are abbreviated as: CB (“Changing and maintaining body position”), CO (“Carrying, moving, and handling objects”), WM (“Walking and moving”), MT (“Moving around using transportation”), and MU (“Mobility, unspecified”). (a) UpSet plot of sections by mobility class combination (n = 3,810). (b) UpSet plot of notes by mobility class combination (n = 600). (c) Pie chart of the proportion of sections mentioning 0–5 classes: 57.0% none, 23.1% one, 12.8% two, 5.8% three, 1.2% four, 0.3% five. (d) Pie chart of the proportion of notes mentioning 0–5 classes: 48.0% none, 29.3% one, 9.0% two, 8.5% three, 4.5% four, 0.7% five. (e–i) Distribution of sections for CB, CO, WM, MT and MU, respectively, showing the three most frequent sections.CC BY-NC-ND · PubMed Central
Fig. 2 Section-level F1-scores with 95% CIs for mobility extraction across five functional status classes and three institutions under different prompt and task configurations. (a) Comparison of zero-shot baseline, three few-shot learning strategies and error-informed prompt refinement approach. (b) Comparison of zero-shot baseline and two task decomposition setups. (CB: Changing and maintaining body position, CO: Carrying, moving, and handling objects, WM: Walking and moving, MT: Moving around using transportation, MU: Mobility, unspecified).
Fig. 2 Section-level F1-scores with 95% CIs for mobility extraction across five functional status classes and three institutions under different prompt and task configurations. (a) Comparison of zero-shot baseline, three few-shot learning strategies and error-informed prompt refinement approach. (b) Comparison of zero-shot baseline and two task decomposition setups. (CB: Changing and maintaining body position, CO: Carrying, moving, and handling objects, WM: Walking and moving, MT: Moving around using transportation, MU: Mobility, unspecified).CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

3 Valid · 1 Not valid · 7 Not checked · 107 Own results and interpretations

44 more claims

Create a free account to see all 118 claims and to verify the 1 reference still open.

See all claims — free account
  • Probably not a substantial claim. 36
  • It may state two separate claims as one. 44

Create a free account to see these claims.

Citations

Cites

Show all 24 references

Research works printed but not linked: 1

  1. Goel, A. et al. Llms accelerate annotation for medical information extraction. In Machine Learning for Health (ML4H) 82–100 (PMLR, 2023).

Cited by

See all 2

Metadata

Authors
Liu, X., Garg, M., Jia, H., Sauver, J. S., Pagali, S. R., & Sohn, S.
Affiliations
Mayo Clinic (US)
Corresponding author
Liu, Xingyi
Venue
Scientific Reports
Published
2026-01-23
Volume
16
Issue
1
Pages
6045-6045
ISSN
2045-2322 (ISSN-L)
OpenAlex
W7125499897
Cited by
2
References
21
Citations by year
2026: 2
FWCI
6.73
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by-nc-nd)
  • Research Square (acceptedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Institutes of Health (R01 AG068007)
  • Eric and Wendy Schmidt Fund for AI Research and Innovation
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Medicine › Health Informatics › Biomedical Text Mining and Ontologies
Keywords
large language models, electronic health records, clinical notes, clinical text, Llama 3, prompting strategies, zero-shot learning, few-shot learning, task decomposition, clinical annotation, aging, precision medicine

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Mobility functional status ascertainment in electronic health records using large language models. https://app.cochat.ai/papers/10.1038/s41598-026-37025-9
The paper
Liu, X., Garg, M., Jia, H., Sauver, J. S., Pagali, S. R., & Sohn, S. (2026). Mobility functional status ascertainment in electronic health records using large language models. Scientific Reports. https://doi.org/10.1038/s41598-026-37025-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.