Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
April 20, 2026ESMO Real World Data and Digital OncologyOpen Access

Accelerating real-world data collection using large language models in rare neoplasms: a bone sarcoma example

View Full Paper
Ask AI
Bookmark
Share

Authors

PTP. TeteryczSRS. RynkunBSB. Szostakowski

Discussion

Loading...

Member takes

Overview

Evaluates large language models for extracting data from medical notes in bone sarcoma patients, indicating potential for improving clinical research efficiency.

Key Points

  • The aim is to assess the performance of small large language models in extracting structured information from Polish medical notes of bone sarcoma patients.
  • Evaluated multiple small large language models as information extractors on Polish medical notes.
  • Select electronic health records of 302 bone sarcoma patients from 2016 to 2022.
  • Annotated key variables: pathology type, tumor size, localization, grade, and primary resection.
  • Implemented ensemble voting strategies to enhance data extraction accuracy.
  • Single-model accuracy ranged from 17.5% to 30.3%, highly dependent on prompts.
  • Accuracy for tumor localization reached up to 36.2%.
  • Majority of extracted non-concordant values were non-valid.
  • The ensemble approach significantly improved accuracy to 83.6%, peaking at 90.0% for resection type.

Cite This Study

Teterycz et al. (2026) studied this question.

synapsesocial.com/papers/69e5c2d003c2939914028bf9https://doi.org/10.1016/j.esmorw.2026.100705
View Full Paper
Ask AI
Bookmark
Share