Pneumatic retinopexy identification in electronic medical records using an on-premises large language model - Takeaways - MDSpire
Coming Soon: Introducing MDSpire News. Learn more
Conexiant’s news site is now MDSpire News. Learn more

Identifying Pneumatic Retinopexy in Electronic Medical Records with an On-Premises Large Language Model

  • By

  • James Tsimiklis

  • Shrirajh Satheakeerthy

  • Jai Paris

  • Weng Onn Chan

  • October 9, 2026

Share

  • 1

    This retrospective diagnostic accuracy study assessed an on-premises, open-weight LLM for identifying documented pneumatic retinopexy in ophthalmology records from 2 hospitals.

  • 2

    Among 542 keyword-enriched records, 110 (20.3%) documented pneumatic retinopexy, and the model classified 108 true positives, 407 true negatives, 25 false positives, and 2 false negatives.

  • 3

    The model had 0.98 sensitivity, 0.94 specificity, 0.81 positive predictive value, and 0.99 negative predictive value against manual review.

  • 4

    Of 25 false-positive classifications, 14 involved pneumatic displacement for submacular hemorrhage and 11 involved gas tamponade associated with pars plana vitrectomy.

  • 5

    The authors support human-supervised case finding and targeted verification of positive classifications, with external multicenter validation and prospective workflow assessment needed before implemen

Original Source(s)

Related Content